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/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 96 /* 97 * Build xmit assembly blocks 98 */ 99 100 struct icmp_bxm { 101 struct sk_buff *skb; 102 int offset; 103 int data_len; 104 105 struct { 106 struct icmphdr icmph; 107 __u32 times[3]; 108 } data; 109 int head_len; 110 struct ip_options replyopts; 111 unsigned char optbuf[40]; 112 }; 113 114 /* 115 * Statistics 116 */ 117 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics); 118 119 /* An array of errno for error messages from dest unreach. */ 120 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 121 122 struct icmp_err icmp_err_convert[] = { 123 { 124 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 125 .fatal = 0, 126 }, 127 { 128 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 129 .fatal = 0, 130 }, 131 { 132 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 133 .fatal = 1, 134 }, 135 { 136 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 137 .fatal = 1, 138 }, 139 { 140 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 141 .fatal = 0, 142 }, 143 { 144 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 145 .fatal = 0, 146 }, 147 { 148 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 149 .fatal = 1, 150 }, 151 { 152 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 153 .fatal = 1, 154 }, 155 { 156 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 157 .fatal = 1, 158 }, 159 { 160 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 161 .fatal = 1, 162 }, 163 { 164 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 165 .fatal = 1, 166 }, 167 { 168 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 169 .fatal = 0, 170 }, 171 { 172 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 173 .fatal = 0, 174 }, 175 { 176 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 177 .fatal = 1, 178 }, 179 { 180 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 181 .fatal = 1, 182 }, 183 { 184 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 185 .fatal = 1, 186 }, 187 }; 188 189 /* Control parameters for ECHO replies. */ 190 int sysctl_icmp_echo_ignore_all; 191 int sysctl_icmp_echo_ignore_broadcasts; 192 193 /* Control parameter - ignore bogus broadcast responses? */ 194 int sysctl_icmp_ignore_bogus_error_responses; 195 196 /* 197 * Configurable global rate limit. 198 * 199 * ratelimit defines tokens/packet consumed for dst->rate_token bucket 200 * ratemask defines which icmp types are ratelimited by setting 201 * it's bit position. 202 * 203 * default: 204 * dest unreachable (3), source quench (4), 205 * time exceeded (11), parameter problem (12) 206 */ 207 208 int sysctl_icmp_ratelimit = 1 * HZ; 209 int sysctl_icmp_ratemask = 0x1818; 210 211 /* 212 * ICMP control array. This specifies what to do with each ICMP. 213 */ 214 215 struct icmp_control { 216 int output_entry; /* Field for increment on output */ 217 int input_entry; /* Field for increment on input */ 218 void (*handler)(struct sk_buff *skb); 219 short error; /* This ICMP is classed as an error message */ 220 }; 221 222 static struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 223 224 /* 225 * The ICMP socket(s). This is the most convenient way to flow control 226 * our ICMP output as well as maintain a clean interface throughout 227 * all layers. All Socketless IP sends will soon be gone. 228 * 229 * On SMP we have one ICMP socket per-cpu. 230 */ 231 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL; 232 #define icmp_socket __get_cpu_var(__icmp_socket) 233 234 static __inline__ int icmp_xmit_lock(void) 235 { 236 local_bh_disable(); 237 238 if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) { 239 /* This can happen if the output path signals a 240 * dst_link_failure() for an outgoing ICMP packet. 241 */ 242 local_bh_enable(); 243 return 1; 244 } 245 return 0; 246 } 247 248 static void icmp_xmit_unlock(void) 249 { 250 spin_unlock_bh(&icmp_socket->sk->sk_lock.slock); 251 } 252 253 /* 254 * Send an ICMP frame. 255 */ 256 257 /* 258 * Check transmit rate limitation for given message. 259 * The rate information is held in the destination cache now. 260 * This function is generic and could be used for other purposes 261 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov. 262 * 263 * Note that the same dst_entry fields are modified by functions in 264 * route.c too, but these work for packet destinations while xrlim_allow 265 * works for icmp destinations. This means the rate limiting information 266 * for one "ip object" is shared - and these ICMPs are twice limited: 267 * by source and by destination. 268 * 269 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate 270 * SHOULD allow setting of rate limits 271 * 272 * Shared between ICMPv4 and ICMPv6. 273 */ 274 #define XRLIM_BURST_FACTOR 6 275 int xrlim_allow(struct dst_entry *dst, int timeout) 276 { 277 unsigned long now; 278 int rc = 0; 279 280 now = jiffies; 281 dst->rate_tokens += now - dst->rate_last; 282 dst->rate_last = now; 283 if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout) 284 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout; 285 if (dst->rate_tokens >= timeout) { 286 dst->rate_tokens -= timeout; 287 rc = 1; 288 } 289 return rc; 290 } 291 292 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code) 293 { 294 struct dst_entry *dst = &rt->u.dst; 295 int rc = 1; 296 297 if (type > NR_ICMP_TYPES) 298 goto out; 299 300 /* Don't limit PMTU discovery. */ 301 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 302 goto out; 303 304 /* No rate limit on loopback */ 305 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 306 goto out; 307 308 /* Limit if icmp type is enabled in ratemask. */ 309 if ((1 << type) & sysctl_icmp_ratemask) 310 rc = xrlim_allow(dst, sysctl_icmp_ratelimit); 311 out: 312 return rc; 313 } 314 315 /* 316 * Maintain the counters used in the SNMP statistics for outgoing ICMP 317 */ 318 static void icmp_out_count(int type) 319 { 320 if (type <= NR_ICMP_TYPES) { 321 ICMP_INC_STATS(icmp_pointers[type].output_entry); 322 ICMP_INC_STATS(ICMP_MIB_OUTMSGS); 323 } 324 } 325 326 /* 327 * Checksum each fragment, and on the first include the headers and final 328 * checksum. 329 */ 330 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 331 struct sk_buff *skb) 332 { 333 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 334 unsigned int csum; 335 336 csum = skb_copy_and_csum_bits(icmp_param->skb, 337 icmp_param->offset + offset, 338 to, len, 0); 339 340 skb->csum = csum_block_add(skb->csum, csum, odd); 341 if (icmp_pointers[icmp_param->data.icmph.type].error) 342 nf_ct_attach(skb, icmp_param->skb); 343 return 0; 344 } 345 346 static void icmp_push_reply(struct icmp_bxm *icmp_param, 347 struct ipcm_cookie *ipc, struct rtable *rt) 348 { 349 struct sk_buff *skb; 350 351 ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param, 352 icmp_param->data_len+icmp_param->head_len, 353 icmp_param->head_len, 354 ipc, rt, MSG_DONTWAIT); 355 356 if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) { 357 struct icmphdr *icmph = skb->h.icmph; 358 unsigned int csum = 0; 359 struct sk_buff *skb1; 360 361 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) { 362 csum = csum_add(csum, skb1->csum); 363 } 364 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 365 (char *)icmph, 366 icmp_param->head_len, csum); 367 icmph->checksum = csum_fold(csum); 368 skb->ip_summed = CHECKSUM_NONE; 369 ip_push_pending_frames(icmp_socket->sk); 370 } 371 } 372 373 /* 374 * Driving logic for building and sending ICMP messages. 375 */ 376 377 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 378 { 379 struct sock *sk = icmp_socket->sk; 380 struct inet_sock *inet = inet_sk(sk); 381 struct ipcm_cookie ipc; 382 struct rtable *rt = (struct rtable *)skb->dst; 383 u32 daddr; 384 385 if (ip_options_echo(&icmp_param->replyopts, skb)) 386 goto out; 387 388 if (icmp_xmit_lock()) 389 return; 390 391 icmp_param->data.icmph.checksum = 0; 392 icmp_out_count(icmp_param->data.icmph.type); 393 394 inet->tos = skb->nh.iph->tos; 395 daddr = ipc.addr = rt->rt_src; 396 ipc.opt = NULL; 397 if (icmp_param->replyopts.optlen) { 398 ipc.opt = &icmp_param->replyopts; 399 if (ipc.opt->srr) 400 daddr = icmp_param->replyopts.faddr; 401 } 402 { 403 struct flowi fl = { .nl_u = { .ip4_u = 404 { .daddr = daddr, 405 .saddr = rt->rt_spec_dst, 406 .tos = RT_TOS(skb->nh.iph->tos) } }, 407 .proto = IPPROTO_ICMP }; 408 if (ip_route_output_key(&rt, &fl)) 409 goto out_unlock; 410 } 411 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type, 412 icmp_param->data.icmph.code)) 413 icmp_push_reply(icmp_param, &ipc, rt); 414 ip_rt_put(rt); 415 out_unlock: 416 icmp_xmit_unlock(); 417 out:; 418 } 419 420 421 /* 422 * Send an ICMP message in response to a situation 423 * 424 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 425 * MAY send more (we do). 426 * MUST NOT change this header information. 427 * MUST NOT reply to a multicast/broadcast IP address. 428 * MUST NOT reply to a multicast/broadcast MAC address. 429 * MUST reply to only the first fragment. 430 */ 431 432 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info) 433 { 434 struct iphdr *iph; 435 int room; 436 struct icmp_bxm icmp_param; 437 struct rtable *rt = (struct rtable *)skb_in->dst; 438 struct ipcm_cookie ipc; 439 u32 saddr; 440 u8 tos; 441 442 if (!rt) 443 goto out; 444 445 /* 446 * Find the original header. It is expected to be valid, of course. 447 * Check this, icmp_send is called from the most obscure devices 448 * sometimes. 449 */ 450 iph = skb_in->nh.iph; 451 452 if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail) 453 goto out; 454 455 /* 456 * No replies to physical multicast/broadcast 457 */ 458 if (skb_in->pkt_type != PACKET_HOST) 459 goto out; 460 461 /* 462 * Now check at the protocol level 463 */ 464 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 465 goto out; 466 467 /* 468 * Only reply to fragment 0. We byte re-order the constant 469 * mask for efficiency. 470 */ 471 if (iph->frag_off & htons(IP_OFFSET)) 472 goto out; 473 474 /* 475 * If we send an ICMP error to an ICMP error a mess would result.. 476 */ 477 if (icmp_pointers[type].error) { 478 /* 479 * We are an error, check if we are replying to an 480 * ICMP error 481 */ 482 if (iph->protocol == IPPROTO_ICMP) { 483 u8 _inner_type, *itp; 484 485 itp = skb_header_pointer(skb_in, 486 skb_in->nh.raw + 487 (iph->ihl << 2) + 488 offsetof(struct icmphdr, 489 type) - 490 skb_in->data, 491 sizeof(_inner_type), 492 &_inner_type); 493 if (itp == NULL) 494 goto out; 495 496 /* 497 * Assume any unknown ICMP type is an error. This 498 * isn't specified by the RFC, but think about it.. 499 */ 500 if (*itp > NR_ICMP_TYPES || 501 icmp_pointers[*itp].error) 502 goto out; 503 } 504 } 505 506 if (icmp_xmit_lock()) 507 return; 508 509 /* 510 * Construct source address and options. 511 */ 512 513 saddr = iph->daddr; 514 if (!(rt->rt_flags & RTCF_LOCAL)) 515 saddr = 0; 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 ende; 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_in->nh.raw - skb_in->data; 535 icmp_out_count(icmp_param.data.icmph.type); 536 inet_sk(icmp_socket->sk)->tos = tos; 537 ipc.addr = iph->saddr; 538 ipc.opt = &icmp_param.replyopts; 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 if (ip_route_output_key(&rt, &fl)) 560 goto out_unlock; 561 } 562 563 if (!icmpv4_xrlim_allow(rt, type, code)) 564 goto ende; 565 566 /* RFC says return as much as we can without exceeding 576 bytes. */ 567 568 room = dst_mtu(&rt->u.dst); 569 if (room > 576) 570 room = 576; 571 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen; 572 room -= sizeof(struct icmphdr); 573 574 icmp_param.data_len = skb_in->len - icmp_param.offset; 575 if (icmp_param.data_len > room) 576 icmp_param.data_len = room; 577 icmp_param.head_len = sizeof(struct icmphdr); 578 579 icmp_push_reply(&icmp_param, &ipc, rt); 580 ende: 581 ip_rt_put(rt); 582 out_unlock: 583 icmp_xmit_unlock(); 584 out:; 585 } 586 587 588 /* 589 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 590 */ 591 592 static void icmp_unreach(struct sk_buff *skb) 593 { 594 struct iphdr *iph; 595 struct icmphdr *icmph; 596 int hash, protocol; 597 struct net_protocol *ipprot; 598 struct sock *raw_sk; 599 u32 info = 0; 600 601 /* 602 * Incomplete header ? 603 * Only checks for the IP header, there should be an 604 * additional check for longer headers in upper levels. 605 */ 606 607 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 608 goto out_err; 609 610 icmph = skb->h.icmph; 611 iph = (struct iphdr *)skb->data; 612 613 if (iph->ihl < 5) /* Mangled header, drop. */ 614 goto out_err; 615 616 if (icmph->type == ICMP_DEST_UNREACH) { 617 switch (icmph->code & 15) { 618 case ICMP_NET_UNREACH: 619 case ICMP_HOST_UNREACH: 620 case ICMP_PROT_UNREACH: 621 case ICMP_PORT_UNREACH: 622 break; 623 case ICMP_FRAG_NEEDED: 624 if (ipv4_config.no_pmtu_disc) { 625 LIMIT_NETDEBUG( 626 printk(KERN_INFO "ICMP: %u.%u.%u.%u: " 627 "fragmentation needed " 628 "and DF set.\n", 629 NIPQUAD(iph->daddr))); 630 } else { 631 info = ip_rt_frag_needed(iph, 632 ntohs(icmph->un.frag.mtu)); 633 if (!info) 634 goto out; 635 } 636 break; 637 case ICMP_SR_FAILED: 638 LIMIT_NETDEBUG( 639 printk(KERN_INFO "ICMP: %u.%u.%u.%u: Source " 640 "Route Failed.\n", 641 NIPQUAD(iph->daddr))); 642 break; 643 default: 644 break; 645 } 646 if (icmph->code > NR_ICMP_UNREACH) 647 goto out; 648 } else if (icmph->type == ICMP_PARAMETERPROB) 649 info = ntohl(icmph->un.gateway) >> 24; 650 651 /* 652 * Throw it at our lower layers 653 * 654 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 655 * header. 656 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 657 * transport layer. 658 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 659 * transport layer. 660 */ 661 662 /* 663 * Check the other end isnt violating RFC 1122. Some routers send 664 * bogus responses to broadcast frames. If you see this message 665 * first check your netmask matches at both ends, if it does then 666 * get the other vendor to fix their kit. 667 */ 668 669 if (!sysctl_icmp_ignore_bogus_error_responses && 670 inet_addr_type(iph->daddr) == RTN_BROADCAST) { 671 if (net_ratelimit()) 672 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP " 673 "type %u, code %u " 674 "error to a broadcast: %u.%u.%u.%u on %s\n", 675 NIPQUAD(skb->nh.iph->saddr), 676 icmph->type, icmph->code, 677 NIPQUAD(iph->daddr), 678 skb->dev->name); 679 goto out; 680 } 681 682 /* Checkin full IP header plus 8 bytes of protocol to 683 * avoid additional coding at protocol handlers. 684 */ 685 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 686 goto out; 687 688 iph = (struct iphdr *)skb->data; 689 protocol = iph->protocol; 690 691 /* 692 * Deliver ICMP message to raw sockets. Pretty useless feature? 693 */ 694 695 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */ 696 hash = protocol & (MAX_INET_PROTOS - 1); 697 read_lock(&raw_v4_lock); 698 if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) { 699 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr, 700 iph->saddr, 701 skb->dev->ifindex)) != NULL) { 702 raw_err(raw_sk, skb, info); 703 raw_sk = sk_next(raw_sk); 704 iph = (struct iphdr *)skb->data; 705 } 706 } 707 read_unlock(&raw_v4_lock); 708 709 rcu_read_lock(); 710 ipprot = rcu_dereference(inet_protos[hash]); 711 if (ipprot && ipprot->err_handler) 712 ipprot->err_handler(skb, info); 713 rcu_read_unlock(); 714 715 out: 716 return; 717 out_err: 718 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 719 goto out; 720 } 721 722 723 /* 724 * Handle ICMP_REDIRECT. 725 */ 726 727 static void icmp_redirect(struct sk_buff *skb) 728 { 729 struct iphdr *iph; 730 unsigned long ip; 731 732 if (skb->len < sizeof(struct iphdr)) 733 goto out_err; 734 735 /* 736 * Get the copied header of the packet that caused the redirect 737 */ 738 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 739 goto out; 740 741 iph = (struct iphdr *)skb->data; 742 ip = iph->daddr; 743 744 switch (skb->h.icmph->code & 7) { 745 case ICMP_REDIR_NET: 746 case ICMP_REDIR_NETTOS: 747 /* 748 * As per RFC recommendations now handle it as a host redirect. 749 */ 750 case ICMP_REDIR_HOST: 751 case ICMP_REDIR_HOSTTOS: 752 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway, 753 iph->saddr, iph->tos, skb->dev); 754 break; 755 } 756 out: 757 return; 758 out_err: 759 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 760 goto out; 761 } 762 763 /* 764 * Handle ICMP_ECHO ("ping") requests. 765 * 766 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 767 * requests. 768 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 769 * included in the reply. 770 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 771 * echo requests, MUST have default=NOT. 772 * See also WRT handling of options once they are done and working. 773 */ 774 775 static void icmp_echo(struct sk_buff *skb) 776 { 777 if (!sysctl_icmp_echo_ignore_all) { 778 struct icmp_bxm icmp_param; 779 780 icmp_param.data.icmph = *skb->h.icmph; 781 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 782 icmp_param.skb = skb; 783 icmp_param.offset = 0; 784 icmp_param.data_len = skb->len; 785 icmp_param.head_len = sizeof(struct icmphdr); 786 icmp_reply(&icmp_param, skb); 787 } 788 } 789 790 /* 791 * Handle ICMP Timestamp requests. 792 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 793 * SHOULD be in the kernel for minimum random latency. 794 * MUST be accurate to a few minutes. 795 * MUST be updated at least at 15Hz. 796 */ 797 static void icmp_timestamp(struct sk_buff *skb) 798 { 799 struct timeval tv; 800 struct icmp_bxm icmp_param; 801 /* 802 * Too short. 803 */ 804 if (skb->len < 4) 805 goto out_err; 806 807 /* 808 * Fill in the current time as ms since midnight UT: 809 */ 810 do_gettimeofday(&tv); 811 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 + 812 tv.tv_usec / 1000); 813 icmp_param.data.times[2] = icmp_param.data.times[1]; 814 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 815 BUG(); 816 icmp_param.data.icmph = *skb->h.icmph; 817 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 818 icmp_param.data.icmph.code = 0; 819 icmp_param.skb = skb; 820 icmp_param.offset = 0; 821 icmp_param.data_len = 0; 822 icmp_param.head_len = sizeof(struct icmphdr) + 12; 823 icmp_reply(&icmp_param, skb); 824 out: 825 return; 826 out_err: 827 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 828 goto out; 829 } 830 831 832 /* 833 * Handle ICMP_ADDRESS_MASK requests. (RFC950) 834 * 835 * RFC1122 (3.2.2.9). A host MUST only send replies to 836 * ADDRESS_MASK requests if it's been configured as an address mask 837 * agent. Receiving a request doesn't constitute implicit permission to 838 * act as one. Of course, implementing this correctly requires (SHOULD) 839 * a way to turn the functionality on and off. Another one for sysctl(), 840 * I guess. -- MS 841 * 842 * RFC1812 (4.3.3.9). A router MUST implement it. 843 * A router SHOULD have switch turning it on/off. 844 * This switch MUST be ON by default. 845 * 846 * Gratuitous replies, zero-source replies are not implemented, 847 * that complies with RFC. DO NOT implement them!!! All the idea 848 * of broadcast addrmask replies as specified in RFC950 is broken. 849 * The problem is that it is not uncommon to have several prefixes 850 * on one physical interface. Moreover, addrmask agent can even be 851 * not aware of existing another prefixes. 852 * If source is zero, addrmask agent cannot choose correct prefix. 853 * Gratuitous mask announcements suffer from the same problem. 854 * RFC1812 explains it, but still allows to use ADDRMASK, 855 * that is pretty silly. --ANK 856 * 857 * All these rules are so bizarre, that I removed kernel addrmask 858 * support at all. It is wrong, it is obsolete, nobody uses it in 859 * any case. --ANK 860 * 861 * Furthermore you can do it with a usermode address agent program 862 * anyway... 863 */ 864 865 static void icmp_address(struct sk_buff *skb) 866 { 867 #if 0 868 if (net_ratelimit()) 869 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n"); 870 #endif 871 } 872 873 /* 874 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain 875 * loudly if an inconsistency is found. 876 */ 877 878 static void icmp_address_reply(struct sk_buff *skb) 879 { 880 struct rtable *rt = (struct rtable *)skb->dst; 881 struct net_device *dev = skb->dev; 882 struct in_device *in_dev; 883 struct in_ifaddr *ifa; 884 885 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC)) 886 goto out; 887 888 in_dev = in_dev_get(dev); 889 if (!in_dev) 890 goto out; 891 rcu_read_lock(); 892 if (in_dev->ifa_list && 893 IN_DEV_LOG_MARTIANS(in_dev) && 894 IN_DEV_FORWARD(in_dev)) { 895 u32 _mask, *mp; 896 897 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask); 898 if (mp == NULL) 899 BUG(); 900 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) { 901 if (*mp == ifa->ifa_mask && 902 inet_ifa_match(rt->rt_src, ifa)) 903 break; 904 } 905 if (!ifa && net_ratelimit()) { 906 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from " 907 "%s/%u.%u.%u.%u\n", 908 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src)); 909 } 910 } 911 rcu_read_unlock(); 912 in_dev_put(in_dev); 913 out:; 914 } 915 916 static void icmp_discard(struct sk_buff *skb) 917 { 918 } 919 920 /* 921 * Deal with incoming ICMP packets. 922 */ 923 int icmp_rcv(struct sk_buff *skb) 924 { 925 struct icmphdr *icmph; 926 struct rtable *rt = (struct rtable *)skb->dst; 927 928 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS); 929 930 switch (skb->ip_summed) { 931 case CHECKSUM_HW: 932 if (!(u16)csum_fold(skb->csum)) 933 break; 934 NETDEBUG(if (net_ratelimit()) 935 printk(KERN_DEBUG "icmp v4 hw csum failure\n")); 936 case CHECKSUM_NONE: 937 if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0))) 938 goto error; 939 default:; 940 } 941 942 if (!pskb_pull(skb, sizeof(struct icmphdr))) 943 goto error; 944 945 icmph = skb->h.icmph; 946 947 /* 948 * 18 is the highest 'known' ICMP type. Anything else is a mystery 949 * 950 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 951 * discarded. 952 */ 953 if (icmph->type > NR_ICMP_TYPES) 954 goto error; 955 956 957 /* 958 * Parse the ICMP message 959 */ 960 961 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 962 /* 963 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 964 * silently ignored (we let user decide with a sysctl). 965 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 966 * discarded if to broadcast/multicast. 967 */ 968 if (icmph->type == ICMP_ECHO && 969 sysctl_icmp_echo_ignore_broadcasts) { 970 goto error; 971 } 972 if (icmph->type != ICMP_ECHO && 973 icmph->type != ICMP_TIMESTAMP && 974 icmph->type != ICMP_ADDRESS && 975 icmph->type != ICMP_ADDRESSREPLY) { 976 goto error; 977 } 978 } 979 980 ICMP_INC_STATS_BH(icmp_pointers[icmph->type].input_entry); 981 icmp_pointers[icmph->type].handler(skb); 982 983 drop: 984 kfree_skb(skb); 985 return 0; 986 error: 987 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 988 goto drop; 989 } 990 991 /* 992 * This table is the definition of how we handle ICMP. 993 */ 994 static struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 995 [ICMP_ECHOREPLY] = { 996 .output_entry = ICMP_MIB_OUTECHOREPS, 997 .input_entry = ICMP_MIB_INECHOREPS, 998 .handler = icmp_discard, 999 }, 1000 [1] = { 1001 .output_entry = ICMP_MIB_DUMMY, 1002 .input_entry = ICMP_MIB_INERRORS, 1003 .handler = icmp_discard, 1004 .error = 1, 1005 }, 1006 [2] = { 1007 .output_entry = ICMP_MIB_DUMMY, 1008 .input_entry = ICMP_MIB_INERRORS, 1009 .handler = icmp_discard, 1010 .error = 1, 1011 }, 1012 [ICMP_DEST_UNREACH] = { 1013 .output_entry = ICMP_MIB_OUTDESTUNREACHS, 1014 .input_entry = ICMP_MIB_INDESTUNREACHS, 1015 .handler = icmp_unreach, 1016 .error = 1, 1017 }, 1018 [ICMP_SOURCE_QUENCH] = { 1019 .output_entry = ICMP_MIB_OUTSRCQUENCHS, 1020 .input_entry = ICMP_MIB_INSRCQUENCHS, 1021 .handler = icmp_unreach, 1022 .error = 1, 1023 }, 1024 [ICMP_REDIRECT] = { 1025 .output_entry = ICMP_MIB_OUTREDIRECTS, 1026 .input_entry = ICMP_MIB_INREDIRECTS, 1027 .handler = icmp_redirect, 1028 .error = 1, 1029 }, 1030 [6] = { 1031 .output_entry = ICMP_MIB_DUMMY, 1032 .input_entry = ICMP_MIB_INERRORS, 1033 .handler = icmp_discard, 1034 .error = 1, 1035 }, 1036 [7] = { 1037 .output_entry = ICMP_MIB_DUMMY, 1038 .input_entry = ICMP_MIB_INERRORS, 1039 .handler = icmp_discard, 1040 .error = 1, 1041 }, 1042 [ICMP_ECHO] = { 1043 .output_entry = ICMP_MIB_OUTECHOS, 1044 .input_entry = ICMP_MIB_INECHOS, 1045 .handler = icmp_echo, 1046 }, 1047 [9] = { 1048 .output_entry = ICMP_MIB_DUMMY, 1049 .input_entry = ICMP_MIB_INERRORS, 1050 .handler = icmp_discard, 1051 .error = 1, 1052 }, 1053 [10] = { 1054 .output_entry = ICMP_MIB_DUMMY, 1055 .input_entry = ICMP_MIB_INERRORS, 1056 .handler = icmp_discard, 1057 .error = 1, 1058 }, 1059 [ICMP_TIME_EXCEEDED] = { 1060 .output_entry = ICMP_MIB_OUTTIMEEXCDS, 1061 .input_entry = ICMP_MIB_INTIMEEXCDS, 1062 .handler = icmp_unreach, 1063 .error = 1, 1064 }, 1065 [ICMP_PARAMETERPROB] = { 1066 .output_entry = ICMP_MIB_OUTPARMPROBS, 1067 .input_entry = ICMP_MIB_INPARMPROBS, 1068 .handler = icmp_unreach, 1069 .error = 1, 1070 }, 1071 [ICMP_TIMESTAMP] = { 1072 .output_entry = ICMP_MIB_OUTTIMESTAMPS, 1073 .input_entry = ICMP_MIB_INTIMESTAMPS, 1074 .handler = icmp_timestamp, 1075 }, 1076 [ICMP_TIMESTAMPREPLY] = { 1077 .output_entry = ICMP_MIB_OUTTIMESTAMPREPS, 1078 .input_entry = ICMP_MIB_INTIMESTAMPREPS, 1079 .handler = icmp_discard, 1080 }, 1081 [ICMP_INFO_REQUEST] = { 1082 .output_entry = ICMP_MIB_DUMMY, 1083 .input_entry = ICMP_MIB_DUMMY, 1084 .handler = icmp_discard, 1085 }, 1086 [ICMP_INFO_REPLY] = { 1087 .output_entry = ICMP_MIB_DUMMY, 1088 .input_entry = ICMP_MIB_DUMMY, 1089 .handler = icmp_discard, 1090 }, 1091 [ICMP_ADDRESS] = { 1092 .output_entry = ICMP_MIB_OUTADDRMASKS, 1093 .input_entry = ICMP_MIB_INADDRMASKS, 1094 .handler = icmp_address, 1095 }, 1096 [ICMP_ADDRESSREPLY] = { 1097 .output_entry = ICMP_MIB_OUTADDRMASKREPS, 1098 .input_entry = ICMP_MIB_INADDRMASKREPS, 1099 .handler = icmp_address_reply, 1100 }, 1101 }; 1102 1103 void __init icmp_init(struct net_proto_family *ops) 1104 { 1105 struct inet_sock *inet; 1106 int i; 1107 1108 for (i = 0; i < NR_CPUS; i++) { 1109 int err; 1110 1111 if (!cpu_possible(i)) 1112 continue; 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