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 <linux/skbuff.h> 87 #include <net/sock.h> 88 #include <linux/errno.h> 89 #include <linux/timer.h> 90 #include <linux/init.h> 91 #include <asm/system.h> 92 #include <asm/uaccess.h> 93 #include <net/checksum.h> 94 #include <net/xfrm.h> 95 #include <net/inet_common.h> 96 97 /* 98 * Build xmit assembly blocks 99 */ 100 101 struct icmp_bxm { 102 struct sk_buff *skb; 103 int offset; 104 int data_len; 105 106 struct { 107 struct icmphdr icmph; 108 __be32 times[3]; 109 } data; 110 int head_len; 111 struct ip_options replyopts; 112 unsigned char optbuf[40]; 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 }; 184 185 /* 186 * ICMP control array. This specifies what to do with each ICMP. 187 */ 188 189 struct icmp_control { 190 void (*handler)(struct sk_buff *skb); 191 short error; /* This ICMP is classed as an error message */ 192 }; 193 194 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 195 196 /* 197 * The ICMP socket(s). This is the most convenient way to flow control 198 * our ICMP output as well as maintain a clean interface throughout 199 * all layers. All Socketless IP sends will soon be gone. 200 * 201 * On SMP we have one ICMP socket per-cpu. 202 */ 203 static struct sock *icmp_sk(struct net *net) 204 { 205 return net->ipv4.icmp_sk[smp_processor_id()]; 206 } 207 208 static inline struct sock *icmp_xmit_lock(struct net *net) 209 { 210 struct sock *sk; 211 212 local_bh_disable(); 213 214 sk = icmp_sk(net); 215 216 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 217 /* This can happen if the output path signals a 218 * dst_link_failure() for an outgoing ICMP packet. 219 */ 220 local_bh_enable(); 221 return NULL; 222 } 223 return sk; 224 } 225 226 static inline void icmp_xmit_unlock(struct sock *sk) 227 { 228 spin_unlock_bh(&sk->sk_lock.slock); 229 } 230 231 /* 232 * Send an ICMP frame. 233 */ 234 235 /* 236 * Check transmit rate limitation for given message. 237 * The rate information is held in the destination cache now. 238 * This function is generic and could be used for other purposes 239 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. 240 * 241 * Note that the same dst_entry fields are modified by functions in 242 * route.c too, but these work for packet destinations while xrlim_allow 243 * works for icmp destinations. This means the rate limiting information 244 * for one "ip object" is shared - and these ICMPs are twice limited: 245 * by source and by destination. 246 * 247 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate 248 * SHOULD allow setting of rate limits 249 * 250 * Shared between ICMPv4 and ICMPv6. 251 */ 252 #define XRLIM_BURST_FACTOR 6 253 int xrlim_allow(struct dst_entry *dst, int timeout) 254 { 255 unsigned long now, token = dst->rate_tokens; 256 int rc = 0; 257 258 now = jiffies; 259 token += now - dst->rate_last; 260 dst->rate_last = now; 261 if (token > XRLIM_BURST_FACTOR * timeout) 262 token = XRLIM_BURST_FACTOR * timeout; 263 if (token >= timeout) { 264 token -= timeout; 265 rc = 1; 266 } 267 dst->rate_tokens = token; 268 return rc; 269 } 270 271 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 272 int type, int code) 273 { 274 struct dst_entry *dst = &rt->u.dst; 275 int rc = 1; 276 277 if (type > NR_ICMP_TYPES) 278 goto out; 279 280 /* Don't limit PMTU discovery. */ 281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 282 goto out; 283 284 /* No rate limit on loopback */ 285 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 286 goto out; 287 288 /* Limit if icmp type is enabled in ratemask. */ 289 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) 290 rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit); 291 out: 292 return rc; 293 } 294 295 /* 296 * Maintain the counters used in the SNMP statistics for outgoing ICMP 297 */ 298 void icmp_out_count(struct net *net, unsigned char type) 299 { 300 ICMPMSGOUT_INC_STATS(net, type); 301 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 302 } 303 304 /* 305 * Checksum each fragment, and on the first include the headers and final 306 * checksum. 307 */ 308 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 309 struct sk_buff *skb) 310 { 311 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 312 __wsum csum; 313 314 csum = skb_copy_and_csum_bits(icmp_param->skb, 315 icmp_param->offset + offset, 316 to, len, 0); 317 318 skb->csum = csum_block_add(skb->csum, csum, odd); 319 if (icmp_pointers[icmp_param->data.icmph.type].error) 320 nf_ct_attach(skb, icmp_param->skb); 321 return 0; 322 } 323 324 static void icmp_push_reply(struct icmp_bxm *icmp_param, 325 struct ipcm_cookie *ipc, struct rtable **rt) 326 { 327 struct sock *sk; 328 struct sk_buff *skb; 329 330 sk = icmp_sk(dev_net((*rt)->u.dst.dev)); 331 if (ip_append_data(sk, icmp_glue_bits, icmp_param, 332 icmp_param->data_len+icmp_param->head_len, 333 icmp_param->head_len, 334 ipc, rt, MSG_DONTWAIT) < 0) 335 ip_flush_pending_frames(sk); 336 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 337 struct icmphdr *icmph = icmp_hdr(skb); 338 __wsum csum = 0; 339 struct sk_buff *skb1; 340 341 skb_queue_walk(&sk->sk_write_queue, skb1) { 342 csum = csum_add(csum, skb1->csum); 343 } 344 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 345 (char *)icmph, 346 icmp_param->head_len, csum); 347 icmph->checksum = csum_fold(csum); 348 skb->ip_summed = CHECKSUM_NONE; 349 ip_push_pending_frames(sk); 350 } 351 } 352 353 /* 354 * Driving logic for building and sending ICMP messages. 355 */ 356 357 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 358 { 359 struct ipcm_cookie ipc; 360 struct rtable *rt = skb_rtable(skb); 361 struct net *net = dev_net(rt->u.dst.dev); 362 struct sock *sk; 363 struct inet_sock *inet; 364 __be32 daddr; 365 366 if (ip_options_echo(&icmp_param->replyopts, skb)) 367 return; 368 369 sk = icmp_xmit_lock(net); 370 if (sk == NULL) 371 return; 372 inet = inet_sk(sk); 373 374 icmp_param->data.icmph.checksum = 0; 375 376 inet->tos = ip_hdr(skb)->tos; 377 daddr = ipc.addr = rt->rt_src; 378 ipc.opt = NULL; 379 ipc.shtx.flags = 0; 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_rtable(skb_in); 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); 430 431 /* 432 * Find the original header. It is expected to be valid, of course. 433 * Check this, icmp_send is called from the most obscure devices 434 * sometimes. 435 */ 436 iph = ip_hdr(skb_in); 437 438 if ((u8 *)iph < skb_in->head || 439 (skb_in->network_header + sizeof(*iph)) > skb_in->tail) 440 goto out; 441 442 /* 443 * No replies to physical multicast/broadcast 444 */ 445 if (skb_in->pkt_type != PACKET_HOST) 446 goto out; 447 448 /* 449 * Now check at the protocol level 450 */ 451 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 452 goto out; 453 454 /* 455 * Only reply to fragment 0. We byte re-order the constant 456 * mask for efficiency. 457 */ 458 if (iph->frag_off & htons(IP_OFFSET)) 459 goto out; 460 461 /* 462 * If we send an ICMP error to an ICMP error a mess would result.. 463 */ 464 if (icmp_pointers[type].error) { 465 /* 466 * We are an error, check if we are replying to an 467 * ICMP error 468 */ 469 if (iph->protocol == IPPROTO_ICMP) { 470 u8 _inner_type, *itp; 471 472 itp = skb_header_pointer(skb_in, 473 skb_network_header(skb_in) + 474 (iph->ihl << 2) + 475 offsetof(struct icmphdr, 476 type) - 477 skb_in->data, 478 sizeof(_inner_type), 479 &_inner_type); 480 if (itp == NULL) 481 goto out; 482 483 /* 484 * Assume any unknown ICMP type is an error. This 485 * isn't specified by the RFC, but think about it.. 486 */ 487 if (*itp > NR_ICMP_TYPES || 488 icmp_pointers[*itp].error) 489 goto out; 490 } 491 } 492 493 sk = icmp_xmit_lock(net); 494 if (sk == NULL) 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 rcu_read_lock(); 506 if (rt->fl.iif && 507 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 508 dev = dev_get_by_index_rcu(net, rt->fl.iif); 509 510 if (dev) 511 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 512 else 513 saddr = 0; 514 rcu_read_unlock(); 515 } 516 517 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 518 IPTOS_PREC_INTERNETCONTROL) : 519 iph->tos; 520 521 if (ip_options_echo(&icmp_param.replyopts, skb_in)) 522 goto out_unlock; 523 524 525 /* 526 * Prepare data for ICMP header. 527 */ 528 529 icmp_param.data.icmph.type = type; 530 icmp_param.data.icmph.code = code; 531 icmp_param.data.icmph.un.gateway = info; 532 icmp_param.data.icmph.checksum = 0; 533 icmp_param.skb = skb_in; 534 icmp_param.offset = skb_network_offset(skb_in); 535 inet_sk(sk)->tos = tos; 536 ipc.addr = iph->saddr; 537 ipc.opt = &icmp_param.replyopts; 538 ipc.shtx.flags = 0; 539 540 { 541 struct flowi fl = { 542 .nl_u = { 543 .ip4_u = { 544 .daddr = icmp_param.replyopts.srr ? 545 icmp_param.replyopts.faddr : 546 iph->saddr, 547 .saddr = saddr, 548 .tos = RT_TOS(tos) 549 } 550 }, 551 .proto = IPPROTO_ICMP, 552 .uli_u = { 553 .icmpt = { 554 .type = type, 555 .code = code 556 } 557 } 558 }; 559 int err; 560 struct rtable *rt2; 561 562 security_skb_classify_flow(skb_in, &fl); 563 if (__ip_route_output_key(net, &rt, &fl)) 564 goto out_unlock; 565 566 /* No need to clone since we're just using its address. */ 567 rt2 = rt; 568 569 err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0); 570 switch (err) { 571 case 0: 572 if (rt != rt2) 573 goto route_done; 574 break; 575 case -EPERM: 576 rt = NULL; 577 break; 578 default: 579 goto out_unlock; 580 } 581 582 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET)) 583 goto relookup_failed; 584 585 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL) 586 err = __ip_route_output_key(net, &rt2, &fl); 587 else { 588 struct flowi fl2 = {}; 589 struct dst_entry *odst; 590 591 fl2.fl4_dst = fl.fl4_src; 592 if (ip_route_output_key(net, &rt2, &fl2)) 593 goto relookup_failed; 594 595 /* Ugh! */ 596 odst = skb_dst(skb_in); 597 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src, 598 RT_TOS(tos), rt2->u.dst.dev); 599 600 dst_release(&rt2->u.dst); 601 rt2 = skb_rtable(skb_in); 602 skb_dst_set(skb_in, odst); 603 } 604 605 if (err) 606 goto relookup_failed; 607 608 err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL, 609 XFRM_LOOKUP_ICMP); 610 switch (err) { 611 case 0: 612 dst_release(&rt->u.dst); 613 rt = rt2; 614 break; 615 case -EPERM: 616 goto ende; 617 default: 618 relookup_failed: 619 if (!rt) 620 goto out_unlock; 621 break; 622 } 623 } 624 625 route_done: 626 if (!icmpv4_xrlim_allow(net, rt, type, code)) 627 goto ende; 628 629 /* RFC says return as much as we can without exceeding 576 bytes. */ 630 631 room = dst_mtu(&rt->u.dst); 632 if (room > 576) 633 room = 576; 634 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen; 635 room -= sizeof(struct icmphdr); 636 637 icmp_param.data_len = skb_in->len - icmp_param.offset; 638 if (icmp_param.data_len > room) 639 icmp_param.data_len = room; 640 icmp_param.head_len = sizeof(struct icmphdr); 641 642 icmp_push_reply(&icmp_param, &ipc, &rt); 643 ende: 644 ip_rt_put(rt); 645 out_unlock: 646 icmp_xmit_unlock(sk); 647 out:; 648 } 649 650 651 /* 652 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 653 */ 654 655 static void icmp_unreach(struct sk_buff *skb) 656 { 657 struct iphdr *iph; 658 struct icmphdr *icmph; 659 int hash, protocol; 660 const struct net_protocol *ipprot; 661 u32 info = 0; 662 struct net *net; 663 664 net = dev_net(skb_dst(skb)->dev); 665 666 /* 667 * Incomplete header ? 668 * Only checks for the IP header, there should be an 669 * additional check for longer headers in upper levels. 670 */ 671 672 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 673 goto out_err; 674 675 icmph = icmp_hdr(skb); 676 iph = (struct iphdr *)skb->data; 677 678 if (iph->ihl < 5) /* Mangled header, drop. */ 679 goto out_err; 680 681 if (icmph->type == ICMP_DEST_UNREACH) { 682 switch (icmph->code & 15) { 683 case ICMP_NET_UNREACH: 684 case ICMP_HOST_UNREACH: 685 case ICMP_PROT_UNREACH: 686 case ICMP_PORT_UNREACH: 687 break; 688 case ICMP_FRAG_NEEDED: 689 if (ipv4_config.no_pmtu_disc) { 690 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n", 691 &iph->daddr); 692 } else { 693 info = ip_rt_frag_needed(net, iph, 694 ntohs(icmph->un.frag.mtu), 695 skb->dev); 696 if (!info) 697 goto out; 698 } 699 break; 700 case ICMP_SR_FAILED: 701 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n", 702 &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 "%pI4 sent an invalid ICMP " 734 "type %u, code %u " 735 "error to a broadcast: %pI4 on %s\n", 736 &ip_hdr(skb)->saddr, 737 icmph->type, icmph->code, 738 &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(net, 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(dev_net(skb->dev), 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(skb)->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 timespec 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 getnstimeofday(&tv); 862 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + 863 tv.tv_nsec / NSEC_PER_MSEC); 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(dev_net(skb_dst(skb)->dev), 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(skb); 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 %pI4 from %s/%pI4\n", 957 mp, dev->name, &rt->rt_src); 958 } 959 } 960 rcu_read_unlock(); 961 in_dev_put(in_dev); 962 out:; 963 } 964 965 static void icmp_discard(struct sk_buff *skb) 966 { 967 } 968 969 /* 970 * Deal with incoming ICMP packets. 971 */ 972 int icmp_rcv(struct sk_buff *skb) 973 { 974 struct icmphdr *icmph; 975 struct rtable *rt = skb_rtable(skb); 976 struct net *net = dev_net(rt->u.dst.dev); 977 978 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 979 struct sec_path *sp = skb_sec_path(skb); 980 int nh; 981 982 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 983 XFRM_STATE_ICMP)) 984 goto drop; 985 986 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 987 goto drop; 988 989 nh = skb_network_offset(skb); 990 skb_set_network_header(skb, sizeof(*icmph)); 991 992 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 993 goto drop; 994 995 skb_set_network_header(skb, nh); 996 } 997 998 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS); 999 1000 switch (skb->ip_summed) { 1001 case CHECKSUM_COMPLETE: 1002 if (!csum_fold(skb->csum)) 1003 break; 1004 /* fall through */ 1005 case CHECKSUM_NONE: 1006 skb->csum = 0; 1007 if (__skb_checksum_complete(skb)) 1008 goto error; 1009 } 1010 1011 if (!pskb_pull(skb, sizeof(*icmph))) 1012 goto error; 1013 1014 icmph = icmp_hdr(skb); 1015 1016 ICMPMSGIN_INC_STATS_BH(net, icmph->type); 1017 /* 1018 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1019 * 1020 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1021 * discarded. 1022 */ 1023 if (icmph->type > NR_ICMP_TYPES) 1024 goto error; 1025 1026 1027 /* 1028 * Parse the ICMP message 1029 */ 1030 1031 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1032 /* 1033 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1034 * silently ignored (we let user decide with a sysctl). 1035 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1036 * discarded if to broadcast/multicast. 1037 */ 1038 if ((icmph->type == ICMP_ECHO || 1039 icmph->type == ICMP_TIMESTAMP) && 1040 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1041 goto error; 1042 } 1043 if (icmph->type != ICMP_ECHO && 1044 icmph->type != ICMP_TIMESTAMP && 1045 icmph->type != ICMP_ADDRESS && 1046 icmph->type != ICMP_ADDRESSREPLY) { 1047 goto error; 1048 } 1049 } 1050 1051 icmp_pointers[icmph->type].handler(skb); 1052 1053 drop: 1054 kfree_skb(skb); 1055 return 0; 1056 error: 1057 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 1058 goto drop; 1059 } 1060 1061 /* 1062 * This table is the definition of how we handle ICMP. 1063 */ 1064 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1065 [ICMP_ECHOREPLY] = { 1066 .handler = icmp_discard, 1067 }, 1068 [1] = { 1069 .handler = icmp_discard, 1070 .error = 1, 1071 }, 1072 [2] = { 1073 .handler = icmp_discard, 1074 .error = 1, 1075 }, 1076 [ICMP_DEST_UNREACH] = { 1077 .handler = icmp_unreach, 1078 .error = 1, 1079 }, 1080 [ICMP_SOURCE_QUENCH] = { 1081 .handler = icmp_unreach, 1082 .error = 1, 1083 }, 1084 [ICMP_REDIRECT] = { 1085 .handler = icmp_redirect, 1086 .error = 1, 1087 }, 1088 [6] = { 1089 .handler = icmp_discard, 1090 .error = 1, 1091 }, 1092 [7] = { 1093 .handler = icmp_discard, 1094 .error = 1, 1095 }, 1096 [ICMP_ECHO] = { 1097 .handler = icmp_echo, 1098 }, 1099 [9] = { 1100 .handler = icmp_discard, 1101 .error = 1, 1102 }, 1103 [10] = { 1104 .handler = icmp_discard, 1105 .error = 1, 1106 }, 1107 [ICMP_TIME_EXCEEDED] = { 1108 .handler = icmp_unreach, 1109 .error = 1, 1110 }, 1111 [ICMP_PARAMETERPROB] = { 1112 .handler = icmp_unreach, 1113 .error = 1, 1114 }, 1115 [ICMP_TIMESTAMP] = { 1116 .handler = icmp_timestamp, 1117 }, 1118 [ICMP_TIMESTAMPREPLY] = { 1119 .handler = icmp_discard, 1120 }, 1121 [ICMP_INFO_REQUEST] = { 1122 .handler = icmp_discard, 1123 }, 1124 [ICMP_INFO_REPLY] = { 1125 .handler = icmp_discard, 1126 }, 1127 [ICMP_ADDRESS] = { 1128 .handler = icmp_address, 1129 }, 1130 [ICMP_ADDRESSREPLY] = { 1131 .handler = icmp_address_reply, 1132 }, 1133 }; 1134 1135 static void __net_exit icmp_sk_exit(struct net *net) 1136 { 1137 int i; 1138 1139 for_each_possible_cpu(i) 1140 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1141 kfree(net->ipv4.icmp_sk); 1142 net->ipv4.icmp_sk = NULL; 1143 } 1144 1145 static int __net_init icmp_sk_init(struct net *net) 1146 { 1147 int i, err; 1148 1149 net->ipv4.icmp_sk = 1150 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL); 1151 if (net->ipv4.icmp_sk == NULL) 1152 return -ENOMEM; 1153 1154 for_each_possible_cpu(i) { 1155 struct sock *sk; 1156 1157 err = inet_ctl_sock_create(&sk, PF_INET, 1158 SOCK_RAW, IPPROTO_ICMP, net); 1159 if (err < 0) 1160 goto fail; 1161 1162 net->ipv4.icmp_sk[i] = sk; 1163 1164 /* Enough space for 2 64K ICMP packets, including 1165 * sk_buff struct overhead. 1166 */ 1167 sk->sk_sndbuf = 1168 (2 * ((64 * 1024) + sizeof(struct sk_buff))); 1169 1170 /* 1171 * Speedup sock_wfree() 1172 */ 1173 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1174 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1175 } 1176 1177 /* Control parameters for ECHO replies. */ 1178 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1179 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1180 1181 /* Control parameter - ignore bogus broadcast responses? */ 1182 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1183 1184 /* 1185 * Configurable global rate limit. 1186 * 1187 * ratelimit defines tokens/packet consumed for dst->rate_token 1188 * bucket ratemask defines which icmp types are ratelimited by 1189 * setting it's bit position. 1190 * 1191 * default: 1192 * dest unreachable (3), source quench (4), 1193 * time exceeded (11), parameter problem (12) 1194 */ 1195 1196 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1197 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1198 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1199 1200 return 0; 1201 1202 fail: 1203 for_each_possible_cpu(i) 1204 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1205 kfree(net->ipv4.icmp_sk); 1206 return err; 1207 } 1208 1209 static struct pernet_operations __net_initdata icmp_sk_ops = { 1210 .init = icmp_sk_init, 1211 .exit = icmp_sk_exit, 1212 }; 1213 1214 int __init icmp_init(void) 1215 { 1216 return register_pernet_subsys(&icmp_sk_ops); 1217 } 1218 1219 EXPORT_SYMBOL(icmp_err_convert); 1220 EXPORT_SYMBOL(icmp_send); 1221 EXPORT_SYMBOL(xrlim_allow); 1222