1 /* 2 * NET3: Implementation of the ICMP protocol layer. 3 * 4 * Alan Cox, <alan@redhat.com> 5 * 6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $ 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 * 13 * Some of the function names and the icmp unreach table for this 14 * module were derived from [icmp.c 1.0.11 06/02/93] by 15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting. 16 * Other than that this module is a complete rewrite. 17 * 18 * Fixes: 19 * Clemens Fruhwirth : introduce global icmp rate limiting 20 * with icmp type masking ability instead 21 * of broken per type icmp timeouts. 22 * Mike Shaver : RFC1122 checks. 23 * Alan Cox : Multicast ping reply as self. 24 * Alan Cox : Fix atomicity lockup in ip_build_xmit 25 * call. 26 * Alan Cox : Added 216,128 byte paths to the MTU 27 * code. 28 * Martin Mares : RFC1812 checks. 29 * Martin Mares : Can be configured to follow redirects 30 * if acting as a router _without_ a 31 * routing protocol (RFC 1812). 32 * Martin Mares : Echo requests may be configured to 33 * be ignored (RFC 1812). 34 * Martin Mares : Limitation of ICMP error message 35 * transmit rate (RFC 1812). 36 * Martin Mares : TOS and Precedence set correctly 37 * (RFC 1812). 38 * Martin Mares : Now copying as much data from the 39 * original packet as we can without 40 * exceeding 576 bytes (RFC 1812). 41 * Willy Konynenberg : Transparent proxying support. 42 * Keith Owens : RFC1191 correction for 4.2BSD based 43 * path MTU bug. 44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are 45 * valid (RFC 1812). 46 * Andi Kleen : Check all packet lengths properly 47 * and moved all kfree_skb() up to 48 * icmp_rcv. 49 * Andi Kleen : Move the rate limit bookkeeping 50 * into the dest entry and use a token 51 * bucket filter (thanks to ANK). Make 52 * the rates sysctl configurable. 53 * Yu Tianli : Fixed two ugly bugs in icmp_send 54 * - IP option length was accounted wrongly 55 * - ICMP header length was not accounted 56 * at all. 57 * Tristan Greaves : Added sysctl option to ignore bogus 58 * broadcast responses from broken routers. 59 * 60 * To Fix: 61 * 62 * - Should use skb_pull() instead of all the manual checking. 63 * This would also greatly simply some upper layer error handlers. --AK 64 * 65 */ 66 67 #include <linux/module.h> 68 #include <linux/types.h> 69 #include <linux/jiffies.h> 70 #include <linux/kernel.h> 71 #include <linux/fcntl.h> 72 #include <linux/socket.h> 73 #include <linux/in.h> 74 #include <linux/inet.h> 75 #include <linux/inetdevice.h> 76 #include <linux/netdevice.h> 77 #include <linux/string.h> 78 #include <linux/netfilter_ipv4.h> 79 #include <net/snmp.h> 80 #include <net/ip.h> 81 #include <net/route.h> 82 #include <net/protocol.h> 83 #include <net/icmp.h> 84 #include <net/tcp.h> 85 #include <net/udp.h> 86 #include <net/raw.h> 87 #include <linux/skbuff.h> 88 #include <net/sock.h> 89 #include <linux/errno.h> 90 #include <linux/timer.h> 91 #include <linux/init.h> 92 #include <asm/system.h> 93 #include <asm/uaccess.h> 94 #include <net/checksum.h> 95 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 __be32 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) __read_mostly; 118 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_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 __read_mostly; 192 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1; 193 194 /* Control parameter - ignore bogus broadcast responses? */ 195 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1; 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 __read_mostly = 1 * HZ; 210 int sysctl_icmp_ratemask __read_mostly = 0x1818; 211 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly; 212 213 /* 214 * ICMP control array. This specifies what to do with each ICMP. 215 */ 216 217 struct icmp_control { 218 void (*handler)(struct sk_buff *skb); 219 short error; /* This ICMP is classed as an error message */ 220 }; 221 222 static const 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 void icmp_out_count(unsigned char type) 319 { 320 ICMPMSGOUT_INC_STATS(type); 321 ICMP_INC_STATS(ICMP_MIB_OUTMSGS); 322 } 323 324 /* 325 * Checksum each fragment, and on the first include the headers and final 326 * checksum. 327 */ 328 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 329 struct sk_buff *skb) 330 { 331 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 332 __wsum csum; 333 334 csum = skb_copy_and_csum_bits(icmp_param->skb, 335 icmp_param->offset + offset, 336 to, len, 0); 337 338 skb->csum = csum_block_add(skb->csum, csum, odd); 339 if (icmp_pointers[icmp_param->data.icmph.type].error) 340 nf_ct_attach(skb, icmp_param->skb); 341 return 0; 342 } 343 344 static void icmp_push_reply(struct icmp_bxm *icmp_param, 345 struct ipcm_cookie *ipc, struct rtable *rt) 346 { 347 struct sk_buff *skb; 348 349 if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param, 350 icmp_param->data_len+icmp_param->head_len, 351 icmp_param->head_len, 352 ipc, rt, MSG_DONTWAIT) < 0) 353 ip_flush_pending_frames(icmp_socket->sk); 354 else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) { 355 struct icmphdr *icmph = icmp_hdr(skb); 356 __wsum csum = 0; 357 struct sk_buff *skb1; 358 359 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) { 360 csum = csum_add(csum, skb1->csum); 361 } 362 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 363 (char *)icmph, 364 icmp_param->head_len, csum); 365 icmph->checksum = csum_fold(csum); 366 skb->ip_summed = CHECKSUM_NONE; 367 ip_push_pending_frames(icmp_socket->sk); 368 } 369 } 370 371 /* 372 * Driving logic for building and sending ICMP messages. 373 */ 374 375 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 376 { 377 struct sock *sk = icmp_socket->sk; 378 struct inet_sock *inet = inet_sk(sk); 379 struct ipcm_cookie ipc; 380 struct rtable *rt = (struct rtable *)skb->dst; 381 __be32 daddr; 382 383 if (ip_options_echo(&icmp_param->replyopts, skb)) 384 return; 385 386 if (icmp_xmit_lock()) 387 return; 388 389 icmp_param->data.icmph.checksum = 0; 390 391 inet->tos = ip_hdr(skb)->tos; 392 daddr = ipc.addr = rt->rt_src; 393 ipc.opt = NULL; 394 if (icmp_param->replyopts.optlen) { 395 ipc.opt = &icmp_param->replyopts; 396 if (ipc.opt->srr) 397 daddr = icmp_param->replyopts.faddr; 398 } 399 { 400 struct flowi fl = { .nl_u = { .ip4_u = 401 { .daddr = daddr, 402 .saddr = rt->rt_spec_dst, 403 .tos = RT_TOS(ip_hdr(skb)->tos) } }, 404 .proto = IPPROTO_ICMP }; 405 security_skb_classify_flow(skb, &fl); 406 if (ip_route_output_key(&rt, &fl)) 407 goto out_unlock; 408 } 409 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type, 410 icmp_param->data.icmph.code)) 411 icmp_push_reply(icmp_param, &ipc, rt); 412 ip_rt_put(rt); 413 out_unlock: 414 icmp_xmit_unlock(); 415 } 416 417 418 /* 419 * Send an ICMP message in response to a situation 420 * 421 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 422 * MAY send more (we do). 423 * MUST NOT change this header information. 424 * MUST NOT reply to a multicast/broadcast IP address. 425 * MUST NOT reply to a multicast/broadcast MAC address. 426 * MUST reply to only the first fragment. 427 */ 428 429 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info) 430 { 431 struct iphdr *iph; 432 int room; 433 struct icmp_bxm icmp_param; 434 struct rtable *rt = (struct rtable *)skb_in->dst; 435 struct ipcm_cookie ipc; 436 __be32 saddr; 437 u8 tos; 438 439 if (!rt) 440 goto out; 441 442 /* 443 * Find the original header. It is expected to be valid, of course. 444 * Check this, icmp_send is called from the most obscure devices 445 * sometimes. 446 */ 447 iph = ip_hdr(skb_in); 448 449 if ((u8 *)iph < skb_in->head || 450 (skb_in->network_header + sizeof(*iph)) > skb_in->tail) 451 goto out; 452 453 /* 454 * No replies to physical multicast/broadcast 455 */ 456 if (skb_in->pkt_type != PACKET_HOST) 457 goto out; 458 459 /* 460 * Now check at the protocol level 461 */ 462 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 463 goto out; 464 465 /* 466 * Only reply to fragment 0. We byte re-order the constant 467 * mask for efficiency. 468 */ 469 if (iph->frag_off & htons(IP_OFFSET)) 470 goto out; 471 472 /* 473 * If we send an ICMP error to an ICMP error a mess would result.. 474 */ 475 if (icmp_pointers[type].error) { 476 /* 477 * We are an error, check if we are replying to an 478 * ICMP error 479 */ 480 if (iph->protocol == IPPROTO_ICMP) { 481 u8 _inner_type, *itp; 482 483 itp = skb_header_pointer(skb_in, 484 skb_network_header(skb_in) + 485 (iph->ihl << 2) + 486 offsetof(struct icmphdr, 487 type) - 488 skb_in->data, 489 sizeof(_inner_type), 490 &_inner_type); 491 if (itp == NULL) 492 goto out; 493 494 /* 495 * Assume any unknown ICMP type is an error. This 496 * isn't specified by the RFC, but think about it.. 497 */ 498 if (*itp > NR_ICMP_TYPES || 499 icmp_pointers[*itp].error) 500 goto out; 501 } 502 } 503 504 if (icmp_xmit_lock()) 505 return; 506 507 /* 508 * Construct source address and options. 509 */ 510 511 saddr = iph->daddr; 512 if (!(rt->rt_flags & RTCF_LOCAL)) { 513 struct net_device *dev = NULL; 514 515 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr) 516 dev = dev_get_by_index(&init_net, rt->fl.iif); 517 518 if (dev) { 519 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 520 dev_put(dev); 521 } else 522 saddr = 0; 523 } 524 525 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 526 IPTOS_PREC_INTERNETCONTROL) : 527 iph->tos; 528 529 if (ip_options_echo(&icmp_param.replyopts, skb_in)) 530 goto out_unlock; 531 532 533 /* 534 * Prepare data for ICMP header. 535 */ 536 537 icmp_param.data.icmph.type = type; 538 icmp_param.data.icmph.code = code; 539 icmp_param.data.icmph.un.gateway = info; 540 icmp_param.data.icmph.checksum = 0; 541 icmp_param.skb = skb_in; 542 icmp_param.offset = skb_network_offset(skb_in); 543 icmp_out_count(icmp_param.data.icmph.type); 544 inet_sk(icmp_socket->sk)->tos = tos; 545 ipc.addr = iph->saddr; 546 ipc.opt = &icmp_param.replyopts; 547 548 { 549 struct flowi fl = { 550 .nl_u = { 551 .ip4_u = { 552 .daddr = icmp_param.replyopts.srr ? 553 icmp_param.replyopts.faddr : 554 iph->saddr, 555 .saddr = saddr, 556 .tos = RT_TOS(tos) 557 } 558 }, 559 .proto = IPPROTO_ICMP, 560 .uli_u = { 561 .icmpt = { 562 .type = type, 563 .code = code 564 } 565 } 566 }; 567 security_skb_classify_flow(skb_in, &fl); 568 if (ip_route_output_key(&rt, &fl)) 569 goto out_unlock; 570 } 571 572 if (!icmpv4_xrlim_allow(rt, type, code)) 573 goto ende; 574 575 /* RFC says return as much as we can without exceeding 576 bytes. */ 576 577 room = dst_mtu(&rt->u.dst); 578 if (room > 576) 579 room = 576; 580 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen; 581 room -= sizeof(struct icmphdr); 582 583 icmp_param.data_len = skb_in->len - icmp_param.offset; 584 if (icmp_param.data_len > room) 585 icmp_param.data_len = room; 586 icmp_param.head_len = sizeof(struct icmphdr); 587 588 icmp_push_reply(&icmp_param, &ipc, rt); 589 ende: 590 ip_rt_put(rt); 591 out_unlock: 592 icmp_xmit_unlock(); 593 out:; 594 } 595 596 597 /* 598 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 599 */ 600 601 static void icmp_unreach(struct sk_buff *skb) 602 { 603 struct iphdr *iph; 604 struct icmphdr *icmph; 605 int hash, protocol; 606 struct net_protocol *ipprot; 607 struct sock *raw_sk; 608 u32 info = 0; 609 610 /* 611 * Incomplete header ? 612 * Only checks for the IP header, there should be an 613 * additional check for longer headers in upper levels. 614 */ 615 616 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 617 goto out_err; 618 619 icmph = icmp_hdr(skb); 620 iph = (struct iphdr *)skb->data; 621 622 if (iph->ihl < 5) /* Mangled header, drop. */ 623 goto out_err; 624 625 if (icmph->type == ICMP_DEST_UNREACH) { 626 switch (icmph->code & 15) { 627 case ICMP_NET_UNREACH: 628 case ICMP_HOST_UNREACH: 629 case ICMP_PROT_UNREACH: 630 case ICMP_PORT_UNREACH: 631 break; 632 case ICMP_FRAG_NEEDED: 633 if (ipv4_config.no_pmtu_disc) { 634 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: " 635 "fragmentation needed " 636 "and DF set.\n", 637 NIPQUAD(iph->daddr)); 638 } else { 639 info = ip_rt_frag_needed(iph, 640 ntohs(icmph->un.frag.mtu)); 641 if (!info) 642 goto out; 643 } 644 break; 645 case ICMP_SR_FAILED: 646 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source " 647 "Route Failed.\n", 648 NIPQUAD(iph->daddr)); 649 break; 650 default: 651 break; 652 } 653 if (icmph->code > NR_ICMP_UNREACH) 654 goto out; 655 } else if (icmph->type == ICMP_PARAMETERPROB) 656 info = ntohl(icmph->un.gateway) >> 24; 657 658 /* 659 * Throw it at our lower layers 660 * 661 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 662 * header. 663 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 664 * transport layer. 665 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 666 * transport layer. 667 */ 668 669 /* 670 * Check the other end isnt violating RFC 1122. Some routers send 671 * bogus responses to broadcast frames. If you see this message 672 * first check your netmask matches at both ends, if it does then 673 * get the other vendor to fix their kit. 674 */ 675 676 if (!sysctl_icmp_ignore_bogus_error_responses && 677 inet_addr_type(iph->daddr) == RTN_BROADCAST) { 678 if (net_ratelimit()) 679 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP " 680 "type %u, code %u " 681 "error to a broadcast: %u.%u.%u.%u on %s\n", 682 NIPQUAD(ip_hdr(skb)->saddr), 683 icmph->type, icmph->code, 684 NIPQUAD(iph->daddr), 685 skb->dev->name); 686 goto out; 687 } 688 689 /* Checkin full IP header plus 8 bytes of protocol to 690 * avoid additional coding at protocol handlers. 691 */ 692 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 693 goto out; 694 695 iph = (struct iphdr *)skb->data; 696 protocol = iph->protocol; 697 698 /* 699 * Deliver ICMP message to raw sockets. Pretty useless feature? 700 */ 701 702 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */ 703 hash = protocol & (MAX_INET_PROTOS - 1); 704 read_lock(&raw_v4_lock); 705 if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) { 706 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr, 707 iph->saddr, 708 skb->dev->ifindex)) != NULL) { 709 raw_err(raw_sk, skb, info); 710 raw_sk = sk_next(raw_sk); 711 iph = (struct iphdr *)skb->data; 712 } 713 } 714 read_unlock(&raw_v4_lock); 715 716 rcu_read_lock(); 717 ipprot = rcu_dereference(inet_protos[hash]); 718 if (ipprot && ipprot->err_handler) 719 ipprot->err_handler(skb, info); 720 rcu_read_unlock(); 721 722 out: 723 return; 724 out_err: 725 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 726 goto out; 727 } 728 729 730 /* 731 * Handle ICMP_REDIRECT. 732 */ 733 734 static void icmp_redirect(struct sk_buff *skb) 735 { 736 struct iphdr *iph; 737 738 if (skb->len < sizeof(struct iphdr)) 739 goto out_err; 740 741 /* 742 * Get the copied header of the packet that caused the redirect 743 */ 744 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 745 goto out; 746 747 iph = (struct iphdr *)skb->data; 748 749 switch (icmp_hdr(skb)->code & 7) { 750 case ICMP_REDIR_NET: 751 case ICMP_REDIR_NETTOS: 752 /* 753 * As per RFC recommendations now handle it as a host redirect. 754 */ 755 case ICMP_REDIR_HOST: 756 case ICMP_REDIR_HOSTTOS: 757 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr, 758 icmp_hdr(skb)->un.gateway, 759 iph->saddr, skb->dev); 760 break; 761 } 762 out: 763 return; 764 out_err: 765 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 766 goto out; 767 } 768 769 /* 770 * Handle ICMP_ECHO ("ping") requests. 771 * 772 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 773 * requests. 774 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 775 * included in the reply. 776 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 777 * echo requests, MUST have default=NOT. 778 * See also WRT handling of options once they are done and working. 779 */ 780 781 static void icmp_echo(struct sk_buff *skb) 782 { 783 if (!sysctl_icmp_echo_ignore_all) { 784 struct icmp_bxm icmp_param; 785 786 icmp_param.data.icmph = *icmp_hdr(skb); 787 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 788 icmp_param.skb = skb; 789 icmp_param.offset = 0; 790 icmp_param.data_len = skb->len; 791 icmp_param.head_len = sizeof(struct icmphdr); 792 icmp_reply(&icmp_param, skb); 793 } 794 } 795 796 /* 797 * Handle ICMP Timestamp requests. 798 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 799 * SHOULD be in the kernel for minimum random latency. 800 * MUST be accurate to a few minutes. 801 * MUST be updated at least at 15Hz. 802 */ 803 static void icmp_timestamp(struct sk_buff *skb) 804 { 805 struct timeval tv; 806 struct icmp_bxm icmp_param; 807 /* 808 * Too short. 809 */ 810 if (skb->len < 4) 811 goto out_err; 812 813 /* 814 * Fill in the current time as ms since midnight UT: 815 */ 816 do_gettimeofday(&tv); 817 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 + 818 tv.tv_usec / 1000); 819 icmp_param.data.times[2] = icmp_param.data.times[1]; 820 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 821 BUG(); 822 icmp_param.data.icmph = *icmp_hdr(skb); 823 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 824 icmp_param.data.icmph.code = 0; 825 icmp_param.skb = skb; 826 icmp_param.offset = 0; 827 icmp_param.data_len = 0; 828 icmp_param.head_len = sizeof(struct icmphdr) + 12; 829 icmp_reply(&icmp_param, skb); 830 out: 831 return; 832 out_err: 833 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 834 goto out; 835 } 836 837 838 /* 839 * Handle ICMP_ADDRESS_MASK requests. (RFC950) 840 * 841 * RFC1122 (3.2.2.9). A host MUST only send replies to 842 * ADDRESS_MASK requests if it's been configured as an address mask 843 * agent. Receiving a request doesn't constitute implicit permission to 844 * act as one. Of course, implementing this correctly requires (SHOULD) 845 * a way to turn the functionality on and off. Another one for sysctl(), 846 * I guess. -- MS 847 * 848 * RFC1812 (4.3.3.9). A router MUST implement it. 849 * A router SHOULD have switch turning it on/off. 850 * This switch MUST be ON by default. 851 * 852 * Gratuitous replies, zero-source replies are not implemented, 853 * that complies with RFC. DO NOT implement them!!! All the idea 854 * of broadcast addrmask replies as specified in RFC950 is broken. 855 * The problem is that it is not uncommon to have several prefixes 856 * on one physical interface. Moreover, addrmask agent can even be 857 * not aware of existing another prefixes. 858 * If source is zero, addrmask agent cannot choose correct prefix. 859 * Gratuitous mask announcements suffer from the same problem. 860 * RFC1812 explains it, but still allows to use ADDRMASK, 861 * that is pretty silly. --ANK 862 * 863 * All these rules are so bizarre, that I removed kernel addrmask 864 * support at all. It is wrong, it is obsolete, nobody uses it in 865 * any case. --ANK 866 * 867 * Furthermore you can do it with a usermode address agent program 868 * anyway... 869 */ 870 871 static void icmp_address(struct sk_buff *skb) 872 { 873 #if 0 874 if (net_ratelimit()) 875 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n"); 876 #endif 877 } 878 879 /* 880 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain 881 * loudly if an inconsistency is found. 882 */ 883 884 static void icmp_address_reply(struct sk_buff *skb) 885 { 886 struct rtable *rt = (struct rtable *)skb->dst; 887 struct net_device *dev = skb->dev; 888 struct in_device *in_dev; 889 struct in_ifaddr *ifa; 890 891 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC)) 892 goto out; 893 894 in_dev = in_dev_get(dev); 895 if (!in_dev) 896 goto out; 897 rcu_read_lock(); 898 if (in_dev->ifa_list && 899 IN_DEV_LOG_MARTIANS(in_dev) && 900 IN_DEV_FORWARD(in_dev)) { 901 __be32 _mask, *mp; 902 903 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask); 904 BUG_ON(mp == NULL); 905 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) { 906 if (*mp == ifa->ifa_mask && 907 inet_ifa_match(rt->rt_src, ifa)) 908 break; 909 } 910 if (!ifa && net_ratelimit()) { 911 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from " 912 "%s/%u.%u.%u.%u\n", 913 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src)); 914 } 915 } 916 rcu_read_unlock(); 917 in_dev_put(in_dev); 918 out:; 919 } 920 921 static void icmp_discard(struct sk_buff *skb) 922 { 923 } 924 925 /* 926 * Deal with incoming ICMP packets. 927 */ 928 int icmp_rcv(struct sk_buff *skb) 929 { 930 struct icmphdr *icmph; 931 struct rtable *rt = (struct rtable *)skb->dst; 932 933 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS); 934 935 switch (skb->ip_summed) { 936 case CHECKSUM_COMPLETE: 937 if (!csum_fold(skb->csum)) 938 break; 939 /* fall through */ 940 case CHECKSUM_NONE: 941 skb->csum = 0; 942 if (__skb_checksum_complete(skb)) 943 goto error; 944 } 945 946 if (!pskb_pull(skb, sizeof(struct icmphdr))) 947 goto error; 948 949 icmph = icmp_hdr(skb); 950 951 ICMPMSGIN_INC_STATS_BH(icmph->type); 952 /* 953 * 18 is the highest 'known' ICMP type. Anything else is a mystery 954 * 955 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 956 * discarded. 957 */ 958 if (icmph->type > NR_ICMP_TYPES) 959 goto error; 960 961 962 /* 963 * Parse the ICMP message 964 */ 965 966 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 967 /* 968 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 969 * silently ignored (we let user decide with a sysctl). 970 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 971 * discarded if to broadcast/multicast. 972 */ 973 if ((icmph->type == ICMP_ECHO || 974 icmph->type == ICMP_TIMESTAMP) && 975 sysctl_icmp_echo_ignore_broadcasts) { 976 goto error; 977 } 978 if (icmph->type != ICMP_ECHO && 979 icmph->type != ICMP_TIMESTAMP && 980 icmph->type != ICMP_ADDRESS && 981 icmph->type != ICMP_ADDRESSREPLY) { 982 goto error; 983 } 984 } 985 986 icmp_pointers[icmph->type].handler(skb); 987 988 drop: 989 kfree_skb(skb); 990 return 0; 991 error: 992 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 993 goto drop; 994 } 995 996 /* 997 * This table is the definition of how we handle ICMP. 998 */ 999 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1000 [ICMP_ECHOREPLY] = { 1001 .handler = icmp_discard, 1002 }, 1003 [1] = { 1004 .handler = icmp_discard, 1005 .error = 1, 1006 }, 1007 [2] = { 1008 .handler = icmp_discard, 1009 .error = 1, 1010 }, 1011 [ICMP_DEST_UNREACH] = { 1012 .handler = icmp_unreach, 1013 .error = 1, 1014 }, 1015 [ICMP_SOURCE_QUENCH] = { 1016 .handler = icmp_unreach, 1017 .error = 1, 1018 }, 1019 [ICMP_REDIRECT] = { 1020 .handler = icmp_redirect, 1021 .error = 1, 1022 }, 1023 [6] = { 1024 .handler = icmp_discard, 1025 .error = 1, 1026 }, 1027 [7] = { 1028 .handler = icmp_discard, 1029 .error = 1, 1030 }, 1031 [ICMP_ECHO] = { 1032 .handler = icmp_echo, 1033 }, 1034 [9] = { 1035 .handler = icmp_discard, 1036 .error = 1, 1037 }, 1038 [10] = { 1039 .handler = icmp_discard, 1040 .error = 1, 1041 }, 1042 [ICMP_TIME_EXCEEDED] = { 1043 .handler = icmp_unreach, 1044 .error = 1, 1045 }, 1046 [ICMP_PARAMETERPROB] = { 1047 .handler = icmp_unreach, 1048 .error = 1, 1049 }, 1050 [ICMP_TIMESTAMP] = { 1051 .handler = icmp_timestamp, 1052 }, 1053 [ICMP_TIMESTAMPREPLY] = { 1054 .handler = icmp_discard, 1055 }, 1056 [ICMP_INFO_REQUEST] = { 1057 .handler = icmp_discard, 1058 }, 1059 [ICMP_INFO_REPLY] = { 1060 .handler = icmp_discard, 1061 }, 1062 [ICMP_ADDRESS] = { 1063 .handler = icmp_address, 1064 }, 1065 [ICMP_ADDRESSREPLY] = { 1066 .handler = icmp_address_reply, 1067 }, 1068 }; 1069 1070 void __init icmp_init(struct net_proto_family *ops) 1071 { 1072 struct inet_sock *inet; 1073 int i; 1074 1075 for_each_possible_cpu(i) { 1076 int err; 1077 1078 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, 1079 &per_cpu(__icmp_socket, i)); 1080 1081 if (err < 0) 1082 panic("Failed to create the ICMP control socket.\n"); 1083 1084 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC; 1085 1086 /* Enough space for 2 64K ICMP packets, including 1087 * sk_buff struct overhead. 1088 */ 1089 per_cpu(__icmp_socket, i)->sk->sk_sndbuf = 1090 (2 * ((64 * 1024) + sizeof(struct sk_buff))); 1091 1092 inet = inet_sk(per_cpu(__icmp_socket, i)->sk); 1093 inet->uc_ttl = -1; 1094 inet->pmtudisc = IP_PMTUDISC_DONT; 1095 1096 /* Unhash it so that IP input processing does not even 1097 * see it, we do not wish this socket to see incoming 1098 * packets. 1099 */ 1100 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk); 1101 } 1102 } 1103 1104 EXPORT_SYMBOL(icmp_err_convert); 1105 EXPORT_SYMBOL(icmp_send); 1106 EXPORT_SYMBOL(icmp_statistics); 1107 EXPORT_SYMBOL(icmpmsg_statistics); 1108 EXPORT_SYMBOL(xrlim_allow); 1109