1 /* 2 * NET3: Implementation of the ICMP protocol layer. 3 * 4 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 * 11 * Some of the function names and the icmp unreach table for this 12 * module were derived from [icmp.c 1.0.11 06/02/93] by 13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting. 14 * Other than that this module is a complete rewrite. 15 * 16 * Fixes: 17 * Clemens Fruhwirth : introduce global icmp rate limiting 18 * with icmp type masking ability instead 19 * of broken per type icmp timeouts. 20 * Mike Shaver : RFC1122 checks. 21 * Alan Cox : Multicast ping reply as self. 22 * Alan Cox : Fix atomicity lockup in ip_build_xmit 23 * call. 24 * Alan Cox : Added 216,128 byte paths to the MTU 25 * code. 26 * Martin Mares : RFC1812 checks. 27 * Martin Mares : Can be configured to follow redirects 28 * if acting as a router _without_ a 29 * routing protocol (RFC 1812). 30 * Martin Mares : Echo requests may be configured to 31 * be ignored (RFC 1812). 32 * Martin Mares : Limitation of ICMP error message 33 * transmit rate (RFC 1812). 34 * Martin Mares : TOS and Precedence set correctly 35 * (RFC 1812). 36 * Martin Mares : Now copying as much data from the 37 * original packet as we can without 38 * exceeding 576 bytes (RFC 1812). 39 * Willy Konynenberg : Transparent proxying support. 40 * Keith Owens : RFC1191 correction for 4.2BSD based 41 * path MTU bug. 42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are 43 * valid (RFC 1812). 44 * Andi Kleen : Check all packet lengths properly 45 * and moved all kfree_skb() up to 46 * icmp_rcv. 47 * Andi Kleen : Move the rate limit bookkeeping 48 * into the dest entry and use a token 49 * bucket filter (thanks to ANK). Make 50 * the rates sysctl configurable. 51 * Yu Tianli : Fixed two ugly bugs in icmp_send 52 * - IP option length was accounted wrongly 53 * - ICMP header length was not accounted 54 * at all. 55 * Tristan Greaves : Added sysctl option to ignore bogus 56 * broadcast responses from broken routers. 57 * 58 * To Fix: 59 * 60 * - Should use skb_pull() instead of all the manual checking. 61 * This would also greatly simply some upper layer error handlers. --AK 62 * 63 */ 64 65 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 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 <linux/slab.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 <net/ping.h> 89 #include <linux/skbuff.h> 90 #include <net/sock.h> 91 #include <linux/errno.h> 92 #include <linux/timer.h> 93 #include <linux/init.h> 94 #include <asm/uaccess.h> 95 #include <net/checksum.h> 96 #include <net/xfrm.h> 97 #include <net/inet_common.h> 98 #include <net/ip_fib.h> 99 100 /* 101 * Build xmit assembly blocks 102 */ 103 104 struct icmp_bxm { 105 struct sk_buff *skb; 106 int offset; 107 int data_len; 108 109 struct { 110 struct icmphdr icmph; 111 __be32 times[3]; 112 } data; 113 int head_len; 114 struct ip_options_data replyopts; 115 }; 116 117 /* An array of errno for error messages from dest unreach. */ 118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 119 120 const struct icmp_err icmp_err_convert[] = { 121 { 122 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 123 .fatal = 0, 124 }, 125 { 126 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 127 .fatal = 0, 128 }, 129 { 130 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 131 .fatal = 1, 132 }, 133 { 134 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 135 .fatal = 1, 136 }, 137 { 138 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 139 .fatal = 0, 140 }, 141 { 142 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 143 .fatal = 0, 144 }, 145 { 146 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 147 .fatal = 1, 148 }, 149 { 150 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 151 .fatal = 1, 152 }, 153 { 154 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 155 .fatal = 1, 156 }, 157 { 158 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 159 .fatal = 1, 160 }, 161 { 162 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 163 .fatal = 1, 164 }, 165 { 166 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 167 .fatal = 0, 168 }, 169 { 170 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 171 .fatal = 0, 172 }, 173 { 174 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 175 .fatal = 1, 176 }, 177 { 178 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 179 .fatal = 1, 180 }, 181 { 182 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 183 .fatal = 1, 184 }, 185 }; 186 EXPORT_SYMBOL(icmp_err_convert); 187 188 /* 189 * ICMP control array. This specifies what to do with each ICMP. 190 */ 191 192 struct icmp_control { 193 void (*handler)(struct sk_buff *skb); 194 short error; /* This ICMP is classed as an error message */ 195 }; 196 197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 198 199 /* 200 * The ICMP socket(s). This is the most convenient way to flow control 201 * our ICMP output as well as maintain a clean interface throughout 202 * all layers. All Socketless IP sends will soon be gone. 203 * 204 * On SMP we have one ICMP socket per-cpu. 205 */ 206 static struct sock *icmp_sk(struct net *net) 207 { 208 return net->ipv4.icmp_sk[smp_processor_id()]; 209 } 210 211 static inline struct sock *icmp_xmit_lock(struct net *net) 212 { 213 struct sock *sk; 214 215 local_bh_disable(); 216 217 sk = icmp_sk(net); 218 219 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 220 /* This can happen if the output path signals a 221 * dst_link_failure() for an outgoing ICMP packet. 222 */ 223 local_bh_enable(); 224 return NULL; 225 } 226 return sk; 227 } 228 229 static inline void icmp_xmit_unlock(struct sock *sk) 230 { 231 spin_unlock_bh(&sk->sk_lock.slock); 232 } 233 234 /* 235 * Send an ICMP frame. 236 */ 237 238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 239 struct flowi4 *fl4, int type, int code) 240 { 241 struct dst_entry *dst = &rt->dst; 242 bool rc = true; 243 244 if (type > NR_ICMP_TYPES) 245 goto out; 246 247 /* Don't limit PMTU discovery. */ 248 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 249 goto out; 250 251 /* No rate limit on loopback */ 252 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 253 goto out; 254 255 /* Limit if icmp type is enabled in ratemask. */ 256 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) { 257 struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1); 258 rc = inet_peer_xrlim_allow(peer, 259 net->ipv4.sysctl_icmp_ratelimit); 260 inet_putpeer(peer); 261 } 262 out: 263 return rc; 264 } 265 266 /* 267 * Maintain the counters used in the SNMP statistics for outgoing ICMP 268 */ 269 void icmp_out_count(struct net *net, unsigned char type) 270 { 271 ICMPMSGOUT_INC_STATS(net, type); 272 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 273 } 274 275 /* 276 * Checksum each fragment, and on the first include the headers and final 277 * checksum. 278 */ 279 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 280 struct sk_buff *skb) 281 { 282 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 283 __wsum csum; 284 285 csum = skb_copy_and_csum_bits(icmp_param->skb, 286 icmp_param->offset + offset, 287 to, len, 0); 288 289 skb->csum = csum_block_add(skb->csum, csum, odd); 290 if (icmp_pointers[icmp_param->data.icmph.type].error) 291 nf_ct_attach(skb, icmp_param->skb); 292 return 0; 293 } 294 295 static void icmp_push_reply(struct icmp_bxm *icmp_param, 296 struct flowi4 *fl4, 297 struct ipcm_cookie *ipc, struct rtable **rt) 298 { 299 struct sock *sk; 300 struct sk_buff *skb; 301 302 sk = icmp_sk(dev_net((*rt)->dst.dev)); 303 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 304 icmp_param->data_len+icmp_param->head_len, 305 icmp_param->head_len, 306 ipc, rt, MSG_DONTWAIT) < 0) { 307 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS); 308 ip_flush_pending_frames(sk); 309 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 310 struct icmphdr *icmph = icmp_hdr(skb); 311 __wsum csum = 0; 312 struct sk_buff *skb1; 313 314 skb_queue_walk(&sk->sk_write_queue, skb1) { 315 csum = csum_add(csum, skb1->csum); 316 } 317 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 318 (char *)icmph, 319 icmp_param->head_len, csum); 320 icmph->checksum = csum_fold(csum); 321 skb->ip_summed = CHECKSUM_NONE; 322 ip_push_pending_frames(sk, fl4); 323 } 324 } 325 326 /* 327 * Driving logic for building and sending ICMP messages. 328 */ 329 330 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 331 { 332 struct ipcm_cookie ipc; 333 struct rtable *rt = skb_rtable(skb); 334 struct net *net = dev_net(rt->dst.dev); 335 struct flowi4 fl4; 336 struct sock *sk; 337 struct inet_sock *inet; 338 __be32 daddr, saddr; 339 340 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb)) 341 return; 342 343 sk = icmp_xmit_lock(net); 344 if (sk == NULL) 345 return; 346 inet = inet_sk(sk); 347 348 icmp_param->data.icmph.checksum = 0; 349 350 inet->tos = ip_hdr(skb)->tos; 351 daddr = ipc.addr = ip_hdr(skb)->saddr; 352 saddr = fib_compute_spec_dst(skb); 353 ipc.opt = NULL; 354 ipc.tx_flags = 0; 355 if (icmp_param->replyopts.opt.opt.optlen) { 356 ipc.opt = &icmp_param->replyopts.opt; 357 if (ipc.opt->opt.srr) 358 daddr = icmp_param->replyopts.opt.opt.faddr; 359 } 360 memset(&fl4, 0, sizeof(fl4)); 361 fl4.daddr = daddr; 362 fl4.saddr = saddr; 363 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 364 fl4.flowi4_proto = IPPROTO_ICMP; 365 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 366 rt = ip_route_output_key(net, &fl4); 367 if (IS_ERR(rt)) 368 goto out_unlock; 369 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type, 370 icmp_param->data.icmph.code)) 371 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 372 ip_rt_put(rt); 373 out_unlock: 374 icmp_xmit_unlock(sk); 375 } 376 377 static struct rtable *icmp_route_lookup(struct net *net, 378 struct flowi4 *fl4, 379 struct sk_buff *skb_in, 380 const struct iphdr *iph, 381 __be32 saddr, u8 tos, 382 int type, int code, 383 struct icmp_bxm *param) 384 { 385 struct rtable *rt, *rt2; 386 struct flowi4 fl4_dec; 387 int err; 388 389 memset(fl4, 0, sizeof(*fl4)); 390 fl4->daddr = (param->replyopts.opt.opt.srr ? 391 param->replyopts.opt.opt.faddr : iph->saddr); 392 fl4->saddr = saddr; 393 fl4->flowi4_tos = RT_TOS(tos); 394 fl4->flowi4_proto = IPPROTO_ICMP; 395 fl4->fl4_icmp_type = type; 396 fl4->fl4_icmp_code = code; 397 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4)); 398 rt = __ip_route_output_key(net, fl4); 399 if (IS_ERR(rt)) 400 return rt; 401 402 /* No need to clone since we're just using its address. */ 403 rt2 = rt; 404 405 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 406 flowi4_to_flowi(fl4), NULL, 0); 407 if (!IS_ERR(rt)) { 408 if (rt != rt2) 409 return rt; 410 } else if (PTR_ERR(rt) == -EPERM) { 411 rt = NULL; 412 } else 413 return rt; 414 415 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 416 if (err) 417 goto relookup_failed; 418 419 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) { 420 rt2 = __ip_route_output_key(net, &fl4_dec); 421 if (IS_ERR(rt2)) 422 err = PTR_ERR(rt2); 423 } else { 424 struct flowi4 fl4_2 = {}; 425 unsigned long orefdst; 426 427 fl4_2.daddr = fl4_dec.saddr; 428 rt2 = ip_route_output_key(net, &fl4_2); 429 if (IS_ERR(rt2)) { 430 err = PTR_ERR(rt2); 431 goto relookup_failed; 432 } 433 /* Ugh! */ 434 orefdst = skb_in->_skb_refdst; /* save old refdst */ 435 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 436 RT_TOS(tos), rt2->dst.dev); 437 438 dst_release(&rt2->dst); 439 rt2 = skb_rtable(skb_in); 440 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 441 } 442 443 if (err) 444 goto relookup_failed; 445 446 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 447 flowi4_to_flowi(&fl4_dec), NULL, 448 XFRM_LOOKUP_ICMP); 449 if (!IS_ERR(rt2)) { 450 dst_release(&rt->dst); 451 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 452 rt = rt2; 453 } else if (PTR_ERR(rt2) == -EPERM) { 454 if (rt) 455 dst_release(&rt->dst); 456 return rt2; 457 } else { 458 err = PTR_ERR(rt2); 459 goto relookup_failed; 460 } 461 return rt; 462 463 relookup_failed: 464 if (rt) 465 return rt; 466 return ERR_PTR(err); 467 } 468 469 /* 470 * Send an ICMP message in response to a situation 471 * 472 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 473 * MAY send more (we do). 474 * MUST NOT change this header information. 475 * MUST NOT reply to a multicast/broadcast IP address. 476 * MUST NOT reply to a multicast/broadcast MAC address. 477 * MUST reply to only the first fragment. 478 */ 479 480 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info) 481 { 482 struct iphdr *iph; 483 int room; 484 struct icmp_bxm icmp_param; 485 struct rtable *rt = skb_rtable(skb_in); 486 struct ipcm_cookie ipc; 487 struct flowi4 fl4; 488 __be32 saddr; 489 u8 tos; 490 struct net *net; 491 struct sock *sk; 492 493 if (!rt) 494 goto out; 495 net = dev_net(rt->dst.dev); 496 497 /* 498 * Find the original header. It is expected to be valid, of course. 499 * Check this, icmp_send is called from the most obscure devices 500 * sometimes. 501 */ 502 iph = ip_hdr(skb_in); 503 504 if ((u8 *)iph < skb_in->head || 505 (skb_in->network_header + sizeof(*iph)) > skb_in->tail) 506 goto out; 507 508 /* 509 * No replies to physical multicast/broadcast 510 */ 511 if (skb_in->pkt_type != PACKET_HOST) 512 goto out; 513 514 /* 515 * Now check at the protocol level 516 */ 517 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 518 goto out; 519 520 /* 521 * Only reply to fragment 0. We byte re-order the constant 522 * mask for efficiency. 523 */ 524 if (iph->frag_off & htons(IP_OFFSET)) 525 goto out; 526 527 /* 528 * If we send an ICMP error to an ICMP error a mess would result.. 529 */ 530 if (icmp_pointers[type].error) { 531 /* 532 * We are an error, check if we are replying to an 533 * ICMP error 534 */ 535 if (iph->protocol == IPPROTO_ICMP) { 536 u8 _inner_type, *itp; 537 538 itp = skb_header_pointer(skb_in, 539 skb_network_header(skb_in) + 540 (iph->ihl << 2) + 541 offsetof(struct icmphdr, 542 type) - 543 skb_in->data, 544 sizeof(_inner_type), 545 &_inner_type); 546 if (itp == NULL) 547 goto out; 548 549 /* 550 * Assume any unknown ICMP type is an error. This 551 * isn't specified by the RFC, but think about it.. 552 */ 553 if (*itp > NR_ICMP_TYPES || 554 icmp_pointers[*itp].error) 555 goto out; 556 } 557 } 558 559 sk = icmp_xmit_lock(net); 560 if (sk == NULL) 561 return; 562 563 /* 564 * Construct source address and options. 565 */ 566 567 saddr = iph->daddr; 568 if (!(rt->rt_flags & RTCF_LOCAL)) { 569 struct net_device *dev = NULL; 570 571 rcu_read_lock(); 572 if (rt_is_input_route(rt) && 573 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 574 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 575 576 if (dev) 577 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 578 else 579 saddr = 0; 580 rcu_read_unlock(); 581 } 582 583 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 584 IPTOS_PREC_INTERNETCONTROL) : 585 iph->tos; 586 587 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in)) 588 goto out_unlock; 589 590 591 /* 592 * Prepare data for ICMP header. 593 */ 594 595 icmp_param.data.icmph.type = type; 596 icmp_param.data.icmph.code = code; 597 icmp_param.data.icmph.un.gateway = info; 598 icmp_param.data.icmph.checksum = 0; 599 icmp_param.skb = skb_in; 600 icmp_param.offset = skb_network_offset(skb_in); 601 inet_sk(sk)->tos = tos; 602 ipc.addr = iph->saddr; 603 ipc.opt = &icmp_param.replyopts.opt; 604 ipc.tx_flags = 0; 605 606 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, 607 type, code, &icmp_param); 608 if (IS_ERR(rt)) 609 goto out_unlock; 610 611 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 612 goto ende; 613 614 /* RFC says return as much as we can without exceeding 576 bytes. */ 615 616 room = dst_mtu(&rt->dst); 617 if (room > 576) 618 room = 576; 619 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 620 room -= sizeof(struct icmphdr); 621 622 icmp_param.data_len = skb_in->len - icmp_param.offset; 623 if (icmp_param.data_len > room) 624 icmp_param.data_len = room; 625 icmp_param.head_len = sizeof(struct icmphdr); 626 627 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 628 ende: 629 ip_rt_put(rt); 630 out_unlock: 631 icmp_xmit_unlock(sk); 632 out:; 633 } 634 EXPORT_SYMBOL(icmp_send); 635 636 637 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 638 { 639 const struct iphdr *iph = (const struct iphdr *) skb->data; 640 const struct net_protocol *ipprot; 641 int protocol = iph->protocol; 642 643 /* Checkin full IP header plus 8 bytes of protocol to 644 * avoid additional coding at protocol handlers. 645 */ 646 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) 647 return; 648 649 raw_icmp_error(skb, protocol, info); 650 651 rcu_read_lock(); 652 ipprot = rcu_dereference(inet_protos[protocol]); 653 if (ipprot && ipprot->err_handler) 654 ipprot->err_handler(skb, info); 655 rcu_read_unlock(); 656 } 657 658 /* 659 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH. 660 */ 661 662 static void icmp_unreach(struct sk_buff *skb) 663 { 664 const struct iphdr *iph; 665 struct icmphdr *icmph; 666 struct net *net; 667 u32 info = 0; 668 669 net = dev_net(skb_dst(skb)->dev); 670 671 /* 672 * Incomplete header ? 673 * Only checks for the IP header, there should be an 674 * additional check for longer headers in upper levels. 675 */ 676 677 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 678 goto out_err; 679 680 icmph = icmp_hdr(skb); 681 iph = (const struct iphdr *)skb->data; 682 683 if (iph->ihl < 5) /* Mangled header, drop. */ 684 goto out_err; 685 686 if (icmph->type == ICMP_DEST_UNREACH) { 687 switch (icmph->code & 15) { 688 case ICMP_NET_UNREACH: 689 case ICMP_HOST_UNREACH: 690 case ICMP_PROT_UNREACH: 691 case ICMP_PORT_UNREACH: 692 break; 693 case ICMP_FRAG_NEEDED: 694 if (ipv4_config.no_pmtu_disc) { 695 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"), 696 &iph->daddr); 697 } else { 698 info = ntohs(icmph->un.frag.mtu); 699 if (!info) 700 goto out; 701 } 702 break; 703 case ICMP_SR_FAILED: 704 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"), 705 &iph->daddr); 706 break; 707 default: 708 break; 709 } 710 if (icmph->code > NR_ICMP_UNREACH) 711 goto out; 712 } else if (icmph->type == ICMP_PARAMETERPROB) 713 info = ntohl(icmph->un.gateway) >> 24; 714 715 /* 716 * Throw it at our lower layers 717 * 718 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 719 * header. 720 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 721 * transport layer. 722 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 723 * transport layer. 724 */ 725 726 /* 727 * Check the other end isn't violating RFC 1122. Some routers send 728 * bogus responses to broadcast frames. If you see this message 729 * first check your netmask matches at both ends, if it does then 730 * get the other vendor to fix their kit. 731 */ 732 733 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 734 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) { 735 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 736 &ip_hdr(skb)->saddr, 737 icmph->type, icmph->code, 738 &iph->daddr, skb->dev->name); 739 goto out; 740 } 741 742 icmp_socket_deliver(skb, info); 743 744 out: 745 return; 746 out_err: 747 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 748 goto out; 749 } 750 751 752 /* 753 * Handle ICMP_REDIRECT. 754 */ 755 756 static void icmp_redirect(struct sk_buff *skb) 757 { 758 if (skb->len < sizeof(struct iphdr)) { 759 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS); 760 return; 761 } 762 763 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 764 return; 765 766 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway); 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 struct net *net; 784 785 net = dev_net(skb_dst(skb)->dev); 786 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 787 struct icmp_bxm icmp_param; 788 789 icmp_param.data.icmph = *icmp_hdr(skb); 790 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 791 icmp_param.skb = skb; 792 icmp_param.offset = 0; 793 icmp_param.data_len = skb->len; 794 icmp_param.head_len = sizeof(struct icmphdr); 795 icmp_reply(&icmp_param, skb); 796 } 797 } 798 799 /* 800 * Handle ICMP Timestamp requests. 801 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 802 * SHOULD be in the kernel for minimum random latency. 803 * MUST be accurate to a few minutes. 804 * MUST be updated at least at 15Hz. 805 */ 806 static void icmp_timestamp(struct sk_buff *skb) 807 { 808 struct timespec tv; 809 struct icmp_bxm icmp_param; 810 /* 811 * Too short. 812 */ 813 if (skb->len < 4) 814 goto out_err; 815 816 /* 817 * Fill in the current time as ms since midnight UT: 818 */ 819 getnstimeofday(&tv); 820 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC + 821 tv.tv_nsec / NSEC_PER_MSEC); 822 icmp_param.data.times[2] = icmp_param.data.times[1]; 823 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)) 824 BUG(); 825 icmp_param.data.icmph = *icmp_hdr(skb); 826 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 827 icmp_param.data.icmph.code = 0; 828 icmp_param.skb = skb; 829 icmp_param.offset = 0; 830 icmp_param.data_len = 0; 831 icmp_param.head_len = sizeof(struct icmphdr) + 12; 832 icmp_reply(&icmp_param, skb); 833 out: 834 return; 835 out_err: 836 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 837 goto out; 838 } 839 840 static void icmp_discard(struct sk_buff *skb) 841 { 842 } 843 844 /* 845 * Deal with incoming ICMP packets. 846 */ 847 int icmp_rcv(struct sk_buff *skb) 848 { 849 struct icmphdr *icmph; 850 struct rtable *rt = skb_rtable(skb); 851 struct net *net = dev_net(rt->dst.dev); 852 853 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 854 struct sec_path *sp = skb_sec_path(skb); 855 int nh; 856 857 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 858 XFRM_STATE_ICMP)) 859 goto drop; 860 861 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 862 goto drop; 863 864 nh = skb_network_offset(skb); 865 skb_set_network_header(skb, sizeof(*icmph)); 866 867 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 868 goto drop; 869 870 skb_set_network_header(skb, nh); 871 } 872 873 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS); 874 875 switch (skb->ip_summed) { 876 case CHECKSUM_COMPLETE: 877 if (!csum_fold(skb->csum)) 878 break; 879 /* fall through */ 880 case CHECKSUM_NONE: 881 skb->csum = 0; 882 if (__skb_checksum_complete(skb)) 883 goto error; 884 } 885 886 if (!pskb_pull(skb, sizeof(*icmph))) 887 goto error; 888 889 icmph = icmp_hdr(skb); 890 891 ICMPMSGIN_INC_STATS_BH(net, icmph->type); 892 /* 893 * 18 is the highest 'known' ICMP type. Anything else is a mystery 894 * 895 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 896 * discarded. 897 */ 898 if (icmph->type > NR_ICMP_TYPES) 899 goto error; 900 901 902 /* 903 * Parse the ICMP message 904 */ 905 906 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 907 /* 908 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 909 * silently ignored (we let user decide with a sysctl). 910 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 911 * discarded if to broadcast/multicast. 912 */ 913 if ((icmph->type == ICMP_ECHO || 914 icmph->type == ICMP_TIMESTAMP) && 915 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 916 goto error; 917 } 918 if (icmph->type != ICMP_ECHO && 919 icmph->type != ICMP_TIMESTAMP && 920 icmph->type != ICMP_ADDRESS && 921 icmph->type != ICMP_ADDRESSREPLY) { 922 goto error; 923 } 924 } 925 926 icmp_pointers[icmph->type].handler(skb); 927 928 drop: 929 kfree_skb(skb); 930 return 0; 931 error: 932 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 933 goto drop; 934 } 935 936 /* 937 * This table is the definition of how we handle ICMP. 938 */ 939 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 940 [ICMP_ECHOREPLY] = { 941 .handler = ping_rcv, 942 }, 943 [1] = { 944 .handler = icmp_discard, 945 .error = 1, 946 }, 947 [2] = { 948 .handler = icmp_discard, 949 .error = 1, 950 }, 951 [ICMP_DEST_UNREACH] = { 952 .handler = icmp_unreach, 953 .error = 1, 954 }, 955 [ICMP_SOURCE_QUENCH] = { 956 .handler = icmp_unreach, 957 .error = 1, 958 }, 959 [ICMP_REDIRECT] = { 960 .handler = icmp_redirect, 961 .error = 1, 962 }, 963 [6] = { 964 .handler = icmp_discard, 965 .error = 1, 966 }, 967 [7] = { 968 .handler = icmp_discard, 969 .error = 1, 970 }, 971 [ICMP_ECHO] = { 972 .handler = icmp_echo, 973 }, 974 [9] = { 975 .handler = icmp_discard, 976 .error = 1, 977 }, 978 [10] = { 979 .handler = icmp_discard, 980 .error = 1, 981 }, 982 [ICMP_TIME_EXCEEDED] = { 983 .handler = icmp_unreach, 984 .error = 1, 985 }, 986 [ICMP_PARAMETERPROB] = { 987 .handler = icmp_unreach, 988 .error = 1, 989 }, 990 [ICMP_TIMESTAMP] = { 991 .handler = icmp_timestamp, 992 }, 993 [ICMP_TIMESTAMPREPLY] = { 994 .handler = icmp_discard, 995 }, 996 [ICMP_INFO_REQUEST] = { 997 .handler = icmp_discard, 998 }, 999 [ICMP_INFO_REPLY] = { 1000 .handler = icmp_discard, 1001 }, 1002 [ICMP_ADDRESS] = { 1003 .handler = icmp_discard, 1004 }, 1005 [ICMP_ADDRESSREPLY] = { 1006 .handler = icmp_discard, 1007 }, 1008 }; 1009 1010 static void __net_exit icmp_sk_exit(struct net *net) 1011 { 1012 int i; 1013 1014 for_each_possible_cpu(i) 1015 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1016 kfree(net->ipv4.icmp_sk); 1017 net->ipv4.icmp_sk = NULL; 1018 } 1019 1020 static int __net_init icmp_sk_init(struct net *net) 1021 { 1022 int i, err; 1023 1024 net->ipv4.icmp_sk = 1025 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL); 1026 if (net->ipv4.icmp_sk == NULL) 1027 return -ENOMEM; 1028 1029 for_each_possible_cpu(i) { 1030 struct sock *sk; 1031 1032 err = inet_ctl_sock_create(&sk, PF_INET, 1033 SOCK_RAW, IPPROTO_ICMP, net); 1034 if (err < 0) 1035 goto fail; 1036 1037 net->ipv4.icmp_sk[i] = sk; 1038 1039 /* Enough space for 2 64K ICMP packets, including 1040 * sk_buff/skb_shared_info struct overhead. 1041 */ 1042 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1043 1044 /* 1045 * Speedup sock_wfree() 1046 */ 1047 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1048 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1049 } 1050 1051 /* Control parameters for ECHO replies. */ 1052 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1053 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1054 1055 /* Control parameter - ignore bogus broadcast responses? */ 1056 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1057 1058 /* 1059 * Configurable global rate limit. 1060 * 1061 * ratelimit defines tokens/packet consumed for dst->rate_token 1062 * bucket ratemask defines which icmp types are ratelimited by 1063 * setting it's bit position. 1064 * 1065 * default: 1066 * dest unreachable (3), source quench (4), 1067 * time exceeded (11), parameter problem (12) 1068 */ 1069 1070 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1071 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1072 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1073 1074 return 0; 1075 1076 fail: 1077 for_each_possible_cpu(i) 1078 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]); 1079 kfree(net->ipv4.icmp_sk); 1080 return err; 1081 } 1082 1083 static struct pernet_operations __net_initdata icmp_sk_ops = { 1084 .init = icmp_sk_init, 1085 .exit = icmp_sk_exit, 1086 }; 1087 1088 int __init icmp_init(void) 1089 { 1090 return register_pernet_subsys(&icmp_sk_ops); 1091 } 1092