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/config.h> 68 #include <linux/module.h> 69 #include <linux/types.h> 70 #include <linux/jiffies.h> 71 #include <linux/kernel.h> 72 #include <linux/fcntl.h> 73 #include <linux/socket.h> 74 #include <linux/in.h> 75 #include <linux/inet.h> 76 #include <linux/inetdevice.h> 77 #include <linux/netdevice.h> 78 #include <linux/string.h> 79 #include <linux/netfilter_ipv4.h> 80 #include <net/snmp.h> 81 #include <net/ip.h> 82 #include <net/route.h> 83 #include <net/protocol.h> 84 #include <net/icmp.h> 85 #include <net/tcp.h> 86 #include <net/udp.h> 87 #include <net/raw.h> 88 #include <linux/skbuff.h> 89 #include <net/sock.h> 90 #include <linux/errno.h> 91 #include <linux/timer.h> 92 #include <linux/init.h> 93 #include <asm/system.h> 94 #include <asm/uaccess.h> 95 #include <net/checksum.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 __u32 times[3]; 109 } data; 110 int head_len; 111 struct ip_options replyopts; 112 unsigned char optbuf[40]; 113 }; 114 115 /* 116 * Statistics 117 */ 118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly; 119 120 /* An array of errno for error messages from dest unreach. */ 121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 122 123 struct icmp_err icmp_err_convert[] = { 124 { 125 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 126 .fatal = 0, 127 }, 128 { 129 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 130 .fatal = 0, 131 }, 132 { 133 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 134 .fatal = 1, 135 }, 136 { 137 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 138 .fatal = 1, 139 }, 140 { 141 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 142 .fatal = 0, 143 }, 144 { 145 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 146 .fatal = 0, 147 }, 148 { 149 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 150 .fatal = 1, 151 }, 152 { 153 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 154 .fatal = 1, 155 }, 156 { 157 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 158 .fatal = 1, 159 }, 160 { 161 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 162 .fatal = 1, 163 }, 164 { 165 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 166 .fatal = 1, 167 }, 168 { 169 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 170 .fatal = 0, 171 }, 172 { 173 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 174 .fatal = 0, 175 }, 176 { 177 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 178 .fatal = 1, 179 }, 180 { 181 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 182 .fatal = 1, 183 }, 184 { 185 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 186 .fatal = 1, 187 }, 188 }; 189 190 /* Control parameters for ECHO replies. */ 191 int sysctl_icmp_echo_ignore_all; 192 int sysctl_icmp_echo_ignore_broadcasts = 1; 193 194 /* Control parameter - ignore bogus broadcast responses? */ 195 int sysctl_icmp_ignore_bogus_error_responses; 196 197 /* 198 * Configurable global rate limit. 199 * 200 * ratelimit defines tokens/packet consumed for dst->rate_token bucket 201 * ratemask defines which icmp types are ratelimited by setting 202 * it's bit position. 203 * 204 * default: 205 * dest unreachable (3), source quench (4), 206 * time exceeded (11), parameter problem (12) 207 */ 208 209 int sysctl_icmp_ratelimit = 1 * HZ; 210 int sysctl_icmp_ratemask = 0x1818; 211 int sysctl_icmp_errors_use_inbound_ifaddr; 212 213 /* 214 * ICMP control array. This specifies what to do with each ICMP. 215 */ 216 217 struct icmp_control { 218 int output_entry; /* Field for increment on output */ 219 int input_entry; /* Field for increment on input */ 220 void (*handler)(struct sk_buff *skb); 221 short error; /* This ICMP is classed as an error message */ 222 }; 223 224 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 225 226 /* 227 * The ICMP socket(s). This is the most convenient way to flow control 228 * our ICMP output as well as maintain a clean interface throughout 229 * all layers. All Socketless IP sends will soon be gone. 230 * 231 * On SMP we have one ICMP socket per-cpu. 232 */ 233 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL; 234 #define icmp_socket __get_cpu_var(__icmp_socket) 235 236 static __inline__ int icmp_xmit_lock(void) 237 { 238 local_bh_disable(); 239 240 if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) { 241 /* This can happen if the output path signals a 242 * dst_link_failure() for an outgoing ICMP packet. 243 */ 244 local_bh_enable(); 245 return 1; 246 } 247 return 0; 248 } 249 250 static void icmp_xmit_unlock(void) 251 { 252 spin_unlock_bh(&icmp_socket->sk->sk_lock.slock); 253 } 254 255 /* 256 * Send an ICMP frame. 257 */ 258 259 /* 260 * Check transmit rate limitation for given message. 261 * The rate information is held in the destination cache now. 262 * This function is generic and could be used for other purposes 263 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. 264 * 265 * Note that the same dst_entry fields are modified by functions in 266 * route.c too, but these work for packet destinations while xrlim_allow 267 * works for icmp destinations. This means the rate limiting information 268 * for one "ip object" is shared - and these ICMPs are twice limited: 269 * by source and by destination. 270 * 271 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate 272 * SHOULD allow setting of rate limits 273 * 274 * Shared between ICMPv4 and ICMPv6. 275 */ 276 #define XRLIM_BURST_FACTOR 6 277 int xrlim_allow(struct dst_entry *dst, int timeout) 278 { 279 unsigned long now; 280 int rc = 0; 281 282 now = jiffies; 283 dst->rate_tokens += now - dst->rate_last; 284 dst->rate_last = now; 285 if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout) 286 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout; 287 if (dst->rate_tokens >= timeout) { 288 dst->rate_tokens -= timeout; 289 rc = 1; 290 } 291 return rc; 292 } 293 294 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code) 295 { 296 struct dst_entry *dst = &rt->u.dst; 297 int rc = 1; 298 299 if (type > NR_ICMP_TYPES) 300 goto out; 301 302 /* Don't limit PMTU discovery. */ 303 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 304 goto out; 305 306 /* No rate limit on loopback */ 307 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 308 goto out; 309 310 /* Limit if icmp type is enabled in ratemask. */ 311 if ((1 << type) & sysctl_icmp_ratemask) 312 rc = xrlim_allow(dst, sysctl_icmp_ratelimit); 313 out: 314 return rc; 315 } 316 317 /* 318 * Maintain the counters used in the SNMP statistics for outgoing ICMP 319 */ 320 static void icmp_out_count(int type) 321 { 322 if (type <= NR_ICMP_TYPES) { 323 ICMP_INC_STATS(icmp_pointers[type].output_entry); 324 ICMP_INC_STATS(ICMP_MIB_OUTMSGS); 325 } 326 } 327 328 /* 329 * Checksum each fragment, and on the first include the headers and final 330 * checksum. 331 */ 332 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 333 struct sk_buff *skb) 334 { 335 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 336 unsigned int csum; 337 338 csum = skb_copy_and_csum_bits(icmp_param->skb, 339 icmp_param->offset + offset, 340 to, len, 0); 341 342 skb->csum = csum_block_add(skb->csum, csum, odd); 343 if (icmp_pointers[icmp_param->data.icmph.type].error) 344 nf_ct_attach(skb, icmp_param->skb); 345 return 0; 346 } 347 348 static void icmp_push_reply(struct icmp_bxm *icmp_param, 349 struct ipcm_cookie *ipc, struct rtable *rt) 350 { 351 struct sk_buff *skb; 352 353 if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param, 354 icmp_param->data_len+icmp_param->head_len, 355 icmp_param->head_len, 356 ipc, rt, MSG_DONTWAIT) < 0) 357 ip_flush_pending_frames(icmp_socket->sk); 358 else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) { 359 struct icmphdr *icmph = skb->h.icmph; 360 unsigned int csum = 0; 361 struct sk_buff *skb1; 362 363 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) { 364 csum = csum_add(csum, skb1->csum); 365 } 366 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 367 (char *)icmph, 368 icmp_param->head_len, csum); 369 icmph->checksum = csum_fold(csum); 370 skb->ip_summed = CHECKSUM_NONE; 371 ip_push_pending_frames(icmp_socket->sk); 372 } 373 } 374 375 /* 376 * Driving logic for building and sending ICMP messages. 377 */ 378 379 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 380 { 381 struct sock *sk = icmp_socket->sk; 382 struct inet_sock *inet = inet_sk(sk); 383 struct ipcm_cookie ipc; 384 struct rtable *rt = (struct rtable *)skb->dst; 385 u32 daddr; 386 387 if (ip_options_echo(&icmp_param->replyopts, skb)) 388 goto out; 389 390 if (icmp_xmit_lock()) 391 return; 392 393 icmp_param->data.icmph.checksum = 0; 394 icmp_out_count(icmp_param->data.icmph.type); 395 396 inet->tos = skb->nh.iph->tos; 397 daddr = ipc.addr = rt->rt_src; 398 ipc.opt = NULL; 399 if (icmp_param->replyopts.optlen) { 400 ipc.opt = &icmp_param->replyopts; 401 if (ipc.opt->srr) 402 daddr = icmp_param->replyopts.faddr; 403 } 404 { 405 struct flowi fl = { .nl_u = { .ip4_u = 406 { .daddr = daddr, 407 .saddr = rt->rt_spec_dst, 408 .tos = RT_TOS(skb->nh.iph->tos) } }, 409 .proto = IPPROTO_ICMP }; 410 if (ip_route_output_key(&rt, &fl)) 411 goto out_unlock; 412 } 413 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type, 414 icmp_param->data.icmph.code)) 415 icmp_push_reply(icmp_param, &ipc, rt); 416 ip_rt_put(rt); 417 out_unlock: 418 icmp_xmit_unlock(); 419 out:; 420 } 421 422 423 /* 424 * Send an ICMP message in response to a situation 425 * 426 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 427 * MAY send more (we do). 428 * MUST NOT change this header information. 429 * MUST NOT reply to a multicast/broadcast IP address. 430 * MUST NOT reply to a multicast/broadcast MAC address. 431 * MUST reply to only the first fragment. 432 */ 433 434 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info) 435 { 436 struct iphdr *iph; 437 int room; 438 struct icmp_bxm icmp_param; 439 struct rtable *rt = (struct rtable *)skb_in->dst; 440 struct ipcm_cookie ipc; 441 u32 saddr; 442 u8 tos; 443 444 if (!rt) 445 goto out; 446 447 /* 448 * Find the original header. It is expected to be valid, of course. 449 * Check this, icmp_send is called from the most obscure devices 450 * sometimes. 451 */ 452 iph = skb_in->nh.iph; 453 454 if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail) 455 goto out; 456 457 /* 458 * No replies to physical multicast/broadcast 459 */ 460 if (skb_in->pkt_type != PACKET_HOST) 461 goto out; 462 463 /* 464 * Now check at the protocol level 465 */ 466 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 467 goto out; 468 469 /* 470 * Only reply to fragment 0. We byte re-order the constant 471 * mask for efficiency. 472 */ 473 if (iph->frag_off & htons(IP_OFFSET)) 474 goto out; 475 476 /* 477 * If we send an ICMP error to an ICMP error a mess would result.. 478 */ 479 if (icmp_pointers[type].error) { 480 /* 481 * We are an error, check if we are replying to an 482 * ICMP error 483 */ 484 if (iph->protocol == IPPROTO_ICMP) { 485 u8 _inner_type, *itp; 486 487 itp = skb_header_pointer(skb_in, 488 skb_in->nh.raw + 489 (iph->ihl << 2) + 490 offsetof(struct icmphdr, 491 type) - 492 skb_in->data, 493 sizeof(_inner_type), 494 &_inner_type); 495 if (itp == NULL) 496 goto out; 497 498 /* 499 * Assume any unknown ICMP type is an error. This 500 * isn't specified by the RFC, but think about it.. 501 */ 502 if (*itp > NR_ICMP_TYPES || 503 icmp_pointers[*itp].error) 504 goto out; 505 } 506 } 507 508 if (icmp_xmit_lock()) 509 return; 510 511 /* 512 * Construct source address and options. 513 */ 514 515 saddr = iph->daddr; 516 if (!(rt->rt_flags & RTCF_LOCAL)) { 517 if (sysctl_icmp_errors_use_inbound_ifaddr) 518 saddr = inet_select_addr(skb_in->dev, 0, RT_SCOPE_LINK); 519 else 520 saddr = 0; 521 } 522 523 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 524 IPTOS_PREC_INTERNETCONTROL) : 525 iph->tos; 526 527 if (ip_options_echo(&icmp_param.replyopts, skb_in)) 528 goto ende; 529 530 531 /* 532 * Prepare data for ICMP header. 533 */ 534 535 icmp_param.data.icmph.type = type; 536 icmp_param.data.icmph.code = code; 537 icmp_param.data.icmph.un.gateway = info; 538 icmp_param.data.icmph.checksum = 0; 539 icmp_param.skb = skb_in; 540 icmp_param.offset = skb_in->nh.raw - skb_in->data; 541 icmp_out_count(icmp_param.data.icmph.type); 542 inet_sk(icmp_socket->sk)->tos = tos; 543 ipc.addr = iph->saddr; 544 ipc.opt = &icmp_param.replyopts; 545 546 { 547 struct flowi fl = { 548 .nl_u = { 549 .ip4_u = { 550 .daddr = icmp_param.replyopts.srr ? 551 icmp_param.replyopts.faddr : 552 iph->saddr, 553 .saddr = saddr, 554 .tos = RT_TOS(tos) 555 } 556 }, 557 .proto = IPPROTO_ICMP, 558 .uli_u = { 559 .icmpt = { 560 .type = type, 561 .code = code 562 } 563 } 564 }; 565 if (ip_route_output_key(&rt, &fl)) 566 goto out_unlock; 567 } 568 569 if (!icmpv4_xrlim_allow(rt, type, code)) 570 goto ende; 571 572 /* RFC says return as much as we can without exceeding 576 bytes. */ 573 574 room = dst_mtu(&rt->u.dst); 575 if (room > 576) 576 room = 576; 577 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen; 578 room -= sizeof(struct icmphdr); 579 580 icmp_param.data_len = skb_in->len - icmp_param.offset; 581 if (icmp_param.data_len > room) 582 icmp_param.data_len = room; 583 icmp_param.head_len = sizeof(struct icmphdr); 584 585 icmp_push_reply(&icmp_param, &ipc, rt); 586 ende: 587 ip_rt_put(rt); 588 out_unlock: 589 icmp_xmit_unlock(); 590 out:; 591 } 592 593 594 /* 595 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 596 */ 597 598 static void icmp_unreach(struct sk_buff *skb) 599 { 600 struct iphdr *iph; 601 struct icmphdr *icmph; 602 int hash, protocol; 603 struct net_protocol *ipprot; 604 struct sock *raw_sk; 605 u32 info = 0; 606 607 /* 608 * Incomplete header ? 609 * Only checks for the IP header, there should be an 610 * additional check for longer headers in upper levels. 611 */ 612 613 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 614 goto out_err; 615 616 icmph = skb->h.icmph; 617 iph = (struct iphdr *)skb->data; 618 619 if (iph->ihl < 5) /* Mangled header, drop. */ 620 goto out_err; 621 622 if (icmph->type == ICMP_DEST_UNREACH) { 623 switch (icmph->code & 15) { 624 case ICMP_NET_UNREACH: 625 case ICMP_HOST_UNREACH: 626 case ICMP_PROT_UNREACH: 627 case ICMP_PORT_UNREACH: 628 break; 629 case ICMP_FRAG_NEEDED: 630 if (ipv4_config.no_pmtu_disc) { 631 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: " 632 "fragmentation needed " 633 "and DF set.\n", 634 NIPQUAD(iph->daddr)); 635 } else { 636 info = ip_rt_frag_needed(iph, 637 ntohs(icmph->un.frag.mtu)); 638 if (!info) 639 goto out; 640 } 641 break; 642 case ICMP_SR_FAILED: 643 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source " 644 "Route Failed.\n", 645 NIPQUAD(iph->daddr)); 646 break; 647 default: 648 break; 649 } 650 if (icmph->code > NR_ICMP_UNREACH) 651 goto out; 652 } else if (icmph->type == ICMP_PARAMETERPROB) 653 info = ntohl(icmph->un.gateway) >> 24; 654 655 /* 656 * Throw it at our lower layers 657 * 658 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 659 * header. 660 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 661 * transport layer. 662 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 663 * transport layer. 664 */ 665 666 /* 667 * Check the other end isnt violating RFC 1122. Some routers send 668 * bogus responses to broadcast frames. If you see this message 669 * first check your netmask matches at both ends, if it does then 670 * get the other vendor to fix their kit. 671 */ 672 673 if (!sysctl_icmp_ignore_bogus_error_responses && 674 inet_addr_type(iph->daddr) == RTN_BROADCAST) { 675 if (net_ratelimit()) 676 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP " 677 "type %u, code %u " 678 "error to a broadcast: %u.%u.%u.%u on %s\n", 679 NIPQUAD(skb->nh.iph->saddr), 680 icmph->type, icmph->code, 681 NIPQUAD(iph->daddr), 682 skb->dev->name); 683 goto out; 684 } 685 686 /* Checkin full IP header plus 8 bytes of protocol to 687 * avoid additional coding at protocol handlers. 688 */ 689 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 690 goto out; 691 692 iph = (struct iphdr *)skb->data; 693 protocol = iph->protocol; 694 695 /* 696 * Deliver ICMP message to raw sockets. Pretty useless feature? 697 */ 698 699 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */ 700 hash = protocol & (MAX_INET_PROTOS - 1); 701 read_lock(&raw_v4_lock); 702 if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) { 703 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr, 704 iph->saddr, 705 skb->dev->ifindex)) != NULL) { 706 raw_err(raw_sk, skb, info); 707 raw_sk = sk_next(raw_sk); 708 iph = (struct iphdr *)skb->data; 709 } 710 } 711 read_unlock(&raw_v4_lock); 712 713 rcu_read_lock(); 714 ipprot = rcu_dereference(inet_protos[hash]); 715 if (ipprot && ipprot->err_handler) 716 ipprot->err_handler(skb, info); 717 rcu_read_unlock(); 718 719 out: 720 return; 721 out_err: 722 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 723 goto out; 724 } 725 726 727 /* 728 * Handle ICMP_REDIRECT. 729 */ 730 731 static void icmp_redirect(struct sk_buff *skb) 732 { 733 struct iphdr *iph; 734 unsigned long ip; 735 736 if (skb->len < sizeof(struct iphdr)) 737 goto out_err; 738 739 /* 740 * Get the copied header of the packet that caused the redirect 741 */ 742 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 743 goto out; 744 745 iph = (struct iphdr *)skb->data; 746 ip = iph->daddr; 747 748 switch (skb->h.icmph->code & 7) { 749 case ICMP_REDIR_NET: 750 case ICMP_REDIR_NETTOS: 751 /* 752 * As per RFC recommendations now handle it as a host redirect. 753 */ 754 case ICMP_REDIR_HOST: 755 case ICMP_REDIR_HOSTTOS: 756 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway, 757 iph->saddr, iph->tos, skb->dev); 758 break; 759 } 760 out: 761 return; 762 out_err: 763 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 764 goto out; 765 } 766 767 /* 768 * Handle ICMP_ECHO ("ping") requests. 769 * 770 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 771 * requests. 772 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 773 * included in the reply. 774 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 775 * echo requests, MUST have default=NOT. 776 * See also WRT handling of options once they are done and working. 777 */ 778 779 static void icmp_echo(struct sk_buff *skb) 780 { 781 if (!sysctl_icmp_echo_ignore_all) { 782 struct icmp_bxm icmp_param; 783 784 icmp_param.data.icmph = *skb->h.icmph; 785 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 786 icmp_param.skb = skb; 787 icmp_param.offset = 0; 788 icmp_param.data_len = skb->len; 789 icmp_param.head_len = sizeof(struct icmphdr); 790 icmp_reply(&icmp_param, skb); 791 } 792 } 793 794 /* 795 * Handle ICMP Timestamp requests. 796 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 797 * SHOULD be in the kernel for minimum random latency. 798 * MUST be accurate to a few minutes. 799 * MUST be updated at least at 15Hz. 800 */ 801 static void icmp_timestamp(struct sk_buff *skb) 802 { 803 struct timeval tv; 804 struct icmp_bxm icmp_param; 805 /* 806 * Too short. 807 */ 808 if (skb->len < 4) 809 goto out_err; 810 811 /* 812 * Fill in the current time as ms since midnight UT: 813 */ 814 do_gettimeofday(&tv); 815 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 + 816 tv.tv_usec / 1000); 817 icmp_param.data.times[2] = icmp_param.data.times[1]; 818 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 819 BUG(); 820 icmp_param.data.icmph = *skb->h.icmph; 821 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 822 icmp_param.data.icmph.code = 0; 823 icmp_param.skb = skb; 824 icmp_param.offset = 0; 825 icmp_param.data_len = 0; 826 icmp_param.head_len = sizeof(struct icmphdr) + 12; 827 icmp_reply(&icmp_param, skb); 828 out: 829 return; 830 out_err: 831 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 832 goto out; 833 } 834 835 836 /* 837 * Handle ICMP_ADDRESS_MASK requests. (RFC950) 838 * 839 * RFC1122 (3.2.2.9). A host MUST only send replies to 840 * ADDRESS_MASK requests if it's been configured as an address mask 841 * agent. Receiving a request doesn't constitute implicit permission to 842 * act as one. Of course, implementing this correctly requires (SHOULD) 843 * a way to turn the functionality on and off. Another one for sysctl(), 844 * I guess. -- MS 845 * 846 * RFC1812 (4.3.3.9). A router MUST implement it. 847 * A router SHOULD have switch turning it on/off. 848 * This switch MUST be ON by default. 849 * 850 * Gratuitous replies, zero-source replies are not implemented, 851 * that complies with RFC. DO NOT implement them!!! All the idea 852 * of broadcast addrmask replies as specified in RFC950 is broken. 853 * The problem is that it is not uncommon to have several prefixes 854 * on one physical interface. Moreover, addrmask agent can even be 855 * not aware of existing another prefixes. 856 * If source is zero, addrmask agent cannot choose correct prefix. 857 * Gratuitous mask announcements suffer from the same problem. 858 * RFC1812 explains it, but still allows to use ADDRMASK, 859 * that is pretty silly. --ANK 860 * 861 * All these rules are so bizarre, that I removed kernel addrmask 862 * support at all. It is wrong, it is obsolete, nobody uses it in 863 * any case. --ANK 864 * 865 * Furthermore you can do it with a usermode address agent program 866 * anyway... 867 */ 868 869 static void icmp_address(struct sk_buff *skb) 870 { 871 #if 0 872 if (net_ratelimit()) 873 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n"); 874 #endif 875 } 876 877 /* 878 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain 879 * loudly if an inconsistency is found. 880 */ 881 882 static void icmp_address_reply(struct sk_buff *skb) 883 { 884 struct rtable *rt = (struct rtable *)skb->dst; 885 struct net_device *dev = skb->dev; 886 struct in_device *in_dev; 887 struct in_ifaddr *ifa; 888 889 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC)) 890 goto out; 891 892 in_dev = in_dev_get(dev); 893 if (!in_dev) 894 goto out; 895 rcu_read_lock(); 896 if (in_dev->ifa_list && 897 IN_DEV_LOG_MARTIANS(in_dev) && 898 IN_DEV_FORWARD(in_dev)) { 899 u32 _mask, *mp; 900 901 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask); 902 BUG_ON(mp == NULL); 903 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) { 904 if (*mp == ifa->ifa_mask && 905 inet_ifa_match(rt->rt_src, ifa)) 906 break; 907 } 908 if (!ifa && net_ratelimit()) { 909 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from " 910 "%s/%u.%u.%u.%u\n", 911 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src)); 912 } 913 } 914 rcu_read_unlock(); 915 in_dev_put(in_dev); 916 out:; 917 } 918 919 static void icmp_discard(struct sk_buff *skb) 920 { 921 } 922 923 /* 924 * Deal with incoming ICMP packets. 925 */ 926 int icmp_rcv(struct sk_buff *skb) 927 { 928 struct icmphdr *icmph; 929 struct rtable *rt = (struct rtable *)skb->dst; 930 931 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS); 932 933 switch (skb->ip_summed) { 934 case CHECKSUM_HW: 935 if (!(u16)csum_fold(skb->csum)) 936 break; 937 /* fall through */ 938 case CHECKSUM_NONE: 939 skb->csum = 0; 940 if (__skb_checksum_complete(skb)) 941 goto error; 942 } 943 944 if (!pskb_pull(skb, sizeof(struct icmphdr))) 945 goto error; 946 947 icmph = skb->h.icmph; 948 949 /* 950 * 18 is the highest 'known' ICMP type. Anything else is a mystery 951 * 952 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 953 * discarded. 954 */ 955 if (icmph->type > NR_ICMP_TYPES) 956 goto error; 957 958 959 /* 960 * Parse the ICMP message 961 */ 962 963 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 964 /* 965 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 966 * silently ignored (we let user decide with a sysctl). 967 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 968 * discarded if to broadcast/multicast. 969 */ 970 if ((icmph->type == ICMP_ECHO || 971 icmph->type == ICMP_TIMESTAMP) && 972 sysctl_icmp_echo_ignore_broadcasts) { 973 goto error; 974 } 975 if (icmph->type != ICMP_ECHO && 976 icmph->type != ICMP_TIMESTAMP && 977 icmph->type != ICMP_ADDRESS && 978 icmph->type != ICMP_ADDRESSREPLY) { 979 goto error; 980 } 981 } 982 983 ICMP_INC_STATS_BH(icmp_pointers[icmph->type].input_entry); 984 icmp_pointers[icmph->type].handler(skb); 985 986 drop: 987 kfree_skb(skb); 988 return 0; 989 error: 990 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 991 goto drop; 992 } 993 994 /* 995 * This table is the definition of how we handle ICMP. 996 */ 997 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 998 [ICMP_ECHOREPLY] = { 999 .output_entry = ICMP_MIB_OUTECHOREPS, 1000 .input_entry = ICMP_MIB_INECHOREPS, 1001 .handler = icmp_discard, 1002 }, 1003 [1] = { 1004 .output_entry = ICMP_MIB_DUMMY, 1005 .input_entry = ICMP_MIB_INERRORS, 1006 .handler = icmp_discard, 1007 .error = 1, 1008 }, 1009 [2] = { 1010 .output_entry = ICMP_MIB_DUMMY, 1011 .input_entry = ICMP_MIB_INERRORS, 1012 .handler = icmp_discard, 1013 .error = 1, 1014 }, 1015 [ICMP_DEST_UNREACH] = { 1016 .output_entry = ICMP_MIB_OUTDESTUNREACHS, 1017 .input_entry = ICMP_MIB_INDESTUNREACHS, 1018 .handler = icmp_unreach, 1019 .error = 1, 1020 }, 1021 [ICMP_SOURCE_QUENCH] = { 1022 .output_entry = ICMP_MIB_OUTSRCQUENCHS, 1023 .input_entry = ICMP_MIB_INSRCQUENCHS, 1024 .handler = icmp_unreach, 1025 .error = 1, 1026 }, 1027 [ICMP_REDIRECT] = { 1028 .output_entry = ICMP_MIB_OUTREDIRECTS, 1029 .input_entry = ICMP_MIB_INREDIRECTS, 1030 .handler = icmp_redirect, 1031 .error = 1, 1032 }, 1033 [6] = { 1034 .output_entry = ICMP_MIB_DUMMY, 1035 .input_entry = ICMP_MIB_INERRORS, 1036 .handler = icmp_discard, 1037 .error = 1, 1038 }, 1039 [7] = { 1040 .output_entry = ICMP_MIB_DUMMY, 1041 .input_entry = ICMP_MIB_INERRORS, 1042 .handler = icmp_discard, 1043 .error = 1, 1044 }, 1045 [ICMP_ECHO] = { 1046 .output_entry = ICMP_MIB_OUTECHOS, 1047 .input_entry = ICMP_MIB_INECHOS, 1048 .handler = icmp_echo, 1049 }, 1050 [9] = { 1051 .output_entry = ICMP_MIB_DUMMY, 1052 .input_entry = ICMP_MIB_INERRORS, 1053 .handler = icmp_discard, 1054 .error = 1, 1055 }, 1056 [10] = { 1057 .output_entry = ICMP_MIB_DUMMY, 1058 .input_entry = ICMP_MIB_INERRORS, 1059 .handler = icmp_discard, 1060 .error = 1, 1061 }, 1062 [ICMP_TIME_EXCEEDED] = { 1063 .output_entry = ICMP_MIB_OUTTIMEEXCDS, 1064 .input_entry = ICMP_MIB_INTIMEEXCDS, 1065 .handler = icmp_unreach, 1066 .error = 1, 1067 }, 1068 [ICMP_PARAMETERPROB] = { 1069 .output_entry = ICMP_MIB_OUTPARMPROBS, 1070 .input_entry = ICMP_MIB_INPARMPROBS, 1071 .handler = icmp_unreach, 1072 .error = 1, 1073 }, 1074 [ICMP_TIMESTAMP] = { 1075 .output_entry = ICMP_MIB_OUTTIMESTAMPS, 1076 .input_entry = ICMP_MIB_INTIMESTAMPS, 1077 .handler = icmp_timestamp, 1078 }, 1079 [ICMP_TIMESTAMPREPLY] = { 1080 .output_entry = ICMP_MIB_OUTTIMESTAMPREPS, 1081 .input_entry = ICMP_MIB_INTIMESTAMPREPS, 1082 .handler = icmp_discard, 1083 }, 1084 [ICMP_INFO_REQUEST] = { 1085 .output_entry = ICMP_MIB_DUMMY, 1086 .input_entry = ICMP_MIB_DUMMY, 1087 .handler = icmp_discard, 1088 }, 1089 [ICMP_INFO_REPLY] = { 1090 .output_entry = ICMP_MIB_DUMMY, 1091 .input_entry = ICMP_MIB_DUMMY, 1092 .handler = icmp_discard, 1093 }, 1094 [ICMP_ADDRESS] = { 1095 .output_entry = ICMP_MIB_OUTADDRMASKS, 1096 .input_entry = ICMP_MIB_INADDRMASKS, 1097 .handler = icmp_address, 1098 }, 1099 [ICMP_ADDRESSREPLY] = { 1100 .output_entry = ICMP_MIB_OUTADDRMASKREPS, 1101 .input_entry = ICMP_MIB_INADDRMASKREPS, 1102 .handler = icmp_address_reply, 1103 }, 1104 }; 1105 1106 void __init icmp_init(struct net_proto_family *ops) 1107 { 1108 struct inet_sock *inet; 1109 int i; 1110 1111 for_each_cpu(i) { 1112 int err; 1113 1114 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, 1115 &per_cpu(__icmp_socket, i)); 1116 1117 if (err < 0) 1118 panic("Failed to create the ICMP control socket.\n"); 1119 1120 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC; 1121 1122 /* Enough space for 2 64K ICMP packets, including 1123 * sk_buff struct overhead. 1124 */ 1125 per_cpu(__icmp_socket, i)->sk->sk_sndbuf = 1126 (2 * ((64 * 1024) + sizeof(struct sk_buff))); 1127 1128 inet = inet_sk(per_cpu(__icmp_socket, i)->sk); 1129 inet->uc_ttl = -1; 1130 inet->pmtudisc = IP_PMTUDISC_DONT; 1131 1132 /* Unhash it so that IP input processing does not even 1133 * see it, we do not wish this socket to see incoming 1134 * packets. 1135 */ 1136 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk); 1137 } 1138 } 1139 1140 EXPORT_SYMBOL(icmp_err_convert); 1141 EXPORT_SYMBOL(icmp_send); 1142 EXPORT_SYMBOL(icmp_statistics); 1143 EXPORT_SYMBOL(xrlim_allow); 1144