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 <linux/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 #include <net/l3mdev.h> 100 101 /* 102 * Build xmit assembly blocks 103 */ 104 105 struct icmp_bxm { 106 struct sk_buff *skb; 107 int offset; 108 int data_len; 109 110 struct { 111 struct icmphdr icmph; 112 __be32 times[3]; 113 } data; 114 int head_len; 115 struct ip_options_data replyopts; 116 }; 117 118 /* An array of errno for error messages from dest unreach. */ 119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 120 121 const struct icmp_err icmp_err_convert[] = { 122 { 123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 124 .fatal = 0, 125 }, 126 { 127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 128 .fatal = 0, 129 }, 130 { 131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 132 .fatal = 1, 133 }, 134 { 135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 136 .fatal = 1, 137 }, 138 { 139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 140 .fatal = 0, 141 }, 142 { 143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 144 .fatal = 0, 145 }, 146 { 147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 148 .fatal = 1, 149 }, 150 { 151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 152 .fatal = 1, 153 }, 154 { 155 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 156 .fatal = 1, 157 }, 158 { 159 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 160 .fatal = 1, 161 }, 162 { 163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 164 .fatal = 1, 165 }, 166 { 167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 168 .fatal = 0, 169 }, 170 { 171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 172 .fatal = 0, 173 }, 174 { 175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 176 .fatal = 1, 177 }, 178 { 179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 180 .fatal = 1, 181 }, 182 { 183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 184 .fatal = 1, 185 }, 186 }; 187 EXPORT_SYMBOL(icmp_err_convert); 188 189 /* 190 * ICMP control array. This specifies what to do with each ICMP. 191 */ 192 193 struct icmp_control { 194 bool (*handler)(struct sk_buff *skb); 195 short error; /* This ICMP is classed as an error message */ 196 }; 197 198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 199 200 /* 201 * The ICMP socket(s). This is the most convenient way to flow control 202 * our ICMP output as well as maintain a clean interface throughout 203 * all layers. All Socketless IP sends will soon be gone. 204 * 205 * On SMP we have one ICMP socket per-cpu. 206 */ 207 static struct sock *icmp_sk(struct net *net) 208 { 209 return *this_cpu_ptr(net->ipv4.icmp_sk); 210 } 211 212 /* Called with BH disabled */ 213 static inline struct sock *icmp_xmit_lock(struct net *net) 214 { 215 struct sock *sk; 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 return NULL; 224 } 225 return sk; 226 } 227 228 static inline void icmp_xmit_unlock(struct sock *sk) 229 { 230 spin_unlock(&sk->sk_lock.slock); 231 } 232 233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000; 234 int sysctl_icmp_msgs_burst __read_mostly = 50; 235 236 static struct { 237 spinlock_t lock; 238 u32 credit; 239 u32 stamp; 240 } icmp_global = { 241 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock), 242 }; 243 244 /** 245 * icmp_global_allow - Are we allowed to send one more ICMP message ? 246 * 247 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec. 248 * Returns false if we reached the limit and can not send another packet. 249 * Note: called with BH disabled 250 */ 251 bool icmp_global_allow(void) 252 { 253 u32 credit, delta, incr = 0, now = (u32)jiffies; 254 bool rc = false; 255 256 /* Check if token bucket is empty and cannot be refilled 257 * without taking the spinlock. 258 */ 259 if (!icmp_global.credit) { 260 delta = min_t(u32, now - icmp_global.stamp, HZ); 261 if (delta < HZ / 50) 262 return false; 263 } 264 265 spin_lock(&icmp_global.lock); 266 delta = min_t(u32, now - icmp_global.stamp, HZ); 267 if (delta >= HZ / 50) { 268 incr = sysctl_icmp_msgs_per_sec * delta / HZ ; 269 if (incr) 270 icmp_global.stamp = now; 271 } 272 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst); 273 if (credit) { 274 credit--; 275 rc = true; 276 } 277 icmp_global.credit = credit; 278 spin_unlock(&icmp_global.lock); 279 return rc; 280 } 281 EXPORT_SYMBOL(icmp_global_allow); 282 283 static bool icmpv4_mask_allow(struct net *net, int type, int code) 284 { 285 if (type > NR_ICMP_TYPES) 286 return true; 287 288 /* Don't limit PMTU discovery. */ 289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 290 return true; 291 292 /* Limit if icmp type is enabled in ratemask. */ 293 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask)) 294 return true; 295 296 return false; 297 } 298 299 static bool icmpv4_global_allow(struct net *net, int type, int code) 300 { 301 if (icmpv4_mask_allow(net, type, code)) 302 return true; 303 304 if (icmp_global_allow()) 305 return true; 306 307 return false; 308 } 309 310 /* 311 * Send an ICMP frame. 312 */ 313 314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 315 struct flowi4 *fl4, int type, int code) 316 { 317 struct dst_entry *dst = &rt->dst; 318 struct inet_peer *peer; 319 bool rc = true; 320 int vif; 321 322 if (icmpv4_mask_allow(net, type, code)) 323 goto out; 324 325 /* No rate limit on loopback */ 326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 327 goto out; 328 329 vif = l3mdev_master_ifindex(dst->dev); 330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1); 331 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit); 332 if (peer) 333 inet_putpeer(peer); 334 out: 335 return rc; 336 } 337 338 /* 339 * Maintain the counters used in the SNMP statistics for outgoing ICMP 340 */ 341 void icmp_out_count(struct net *net, unsigned char type) 342 { 343 ICMPMSGOUT_INC_STATS(net, type); 344 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 345 } 346 347 /* 348 * Checksum each fragment, and on the first include the headers and final 349 * checksum. 350 */ 351 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 352 struct sk_buff *skb) 353 { 354 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 355 __wsum csum; 356 357 csum = skb_copy_and_csum_bits(icmp_param->skb, 358 icmp_param->offset + offset, 359 to, len, 0); 360 361 skb->csum = csum_block_add(skb->csum, csum, odd); 362 if (icmp_pointers[icmp_param->data.icmph.type].error) 363 nf_ct_attach(skb, icmp_param->skb); 364 return 0; 365 } 366 367 static void icmp_push_reply(struct icmp_bxm *icmp_param, 368 struct flowi4 *fl4, 369 struct ipcm_cookie *ipc, struct rtable **rt) 370 { 371 struct sock *sk; 372 struct sk_buff *skb; 373 374 sk = icmp_sk(dev_net((*rt)->dst.dev)); 375 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 376 icmp_param->data_len+icmp_param->head_len, 377 icmp_param->head_len, 378 ipc, rt, MSG_DONTWAIT) < 0) { 379 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 380 ip_flush_pending_frames(sk); 381 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 382 struct icmphdr *icmph = icmp_hdr(skb); 383 __wsum csum = 0; 384 struct sk_buff *skb1; 385 386 skb_queue_walk(&sk->sk_write_queue, skb1) { 387 csum = csum_add(csum, skb1->csum); 388 } 389 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 390 (char *)icmph, 391 icmp_param->head_len, csum); 392 icmph->checksum = csum_fold(csum); 393 skb->ip_summed = CHECKSUM_NONE; 394 ip_push_pending_frames(sk, fl4); 395 } 396 } 397 398 /* 399 * Driving logic for building and sending ICMP messages. 400 */ 401 402 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 403 { 404 struct ipcm_cookie ipc; 405 struct rtable *rt = skb_rtable(skb); 406 struct net *net = dev_net(rt->dst.dev); 407 struct flowi4 fl4; 408 struct sock *sk; 409 struct inet_sock *inet; 410 __be32 daddr, saddr; 411 u32 mark = IP4_REPLY_MARK(net, skb->mark); 412 int type = icmp_param->data.icmph.type; 413 int code = icmp_param->data.icmph.code; 414 415 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 416 return; 417 418 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 419 local_bh_disable(); 420 421 /* global icmp_msgs_per_sec */ 422 if (!icmpv4_global_allow(net, type, code)) 423 goto out_bh_enable; 424 425 sk = icmp_xmit_lock(net); 426 if (!sk) 427 goto out_bh_enable; 428 inet = inet_sk(sk); 429 430 icmp_param->data.icmph.checksum = 0; 431 432 ipcm_init(&ipc); 433 inet->tos = ip_hdr(skb)->tos; 434 sk->sk_mark = mark; 435 daddr = ipc.addr = ip_hdr(skb)->saddr; 436 saddr = fib_compute_spec_dst(skb); 437 438 if (icmp_param->replyopts.opt.opt.optlen) { 439 ipc.opt = &icmp_param->replyopts.opt; 440 if (ipc.opt->opt.srr) 441 daddr = icmp_param->replyopts.opt.opt.faddr; 442 } 443 memset(&fl4, 0, sizeof(fl4)); 444 fl4.daddr = daddr; 445 fl4.saddr = saddr; 446 fl4.flowi4_mark = mark; 447 fl4.flowi4_uid = sock_net_uid(net, NULL); 448 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 449 fl4.flowi4_proto = IPPROTO_ICMP; 450 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 451 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 452 rt = ip_route_output_key(net, &fl4); 453 if (IS_ERR(rt)) 454 goto out_unlock; 455 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 456 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 457 ip_rt_put(rt); 458 out_unlock: 459 icmp_xmit_unlock(sk); 460 out_bh_enable: 461 local_bh_enable(); 462 } 463 464 static struct rtable *icmp_route_lookup(struct net *net, 465 struct flowi4 *fl4, 466 struct sk_buff *skb_in, 467 const struct iphdr *iph, 468 __be32 saddr, u8 tos, u32 mark, 469 int type, int code, 470 struct icmp_bxm *param) 471 { 472 struct rtable *rt, *rt2; 473 struct flowi4 fl4_dec; 474 int err; 475 476 memset(fl4, 0, sizeof(*fl4)); 477 fl4->daddr = (param->replyopts.opt.opt.srr ? 478 param->replyopts.opt.opt.faddr : iph->saddr); 479 fl4->saddr = saddr; 480 fl4->flowi4_mark = mark; 481 fl4->flowi4_uid = sock_net_uid(net, NULL); 482 fl4->flowi4_tos = RT_TOS(tos); 483 fl4->flowi4_proto = IPPROTO_ICMP; 484 fl4->fl4_icmp_type = type; 485 fl4->fl4_icmp_code = code; 486 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev); 487 488 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4)); 489 rt = ip_route_output_key_hash(net, fl4, skb_in); 490 if (IS_ERR(rt)) 491 return rt; 492 493 /* No need to clone since we're just using its address. */ 494 rt2 = rt; 495 496 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 497 flowi4_to_flowi(fl4), NULL, 0); 498 if (!IS_ERR(rt)) { 499 if (rt != rt2) 500 return rt; 501 } else if (PTR_ERR(rt) == -EPERM) { 502 rt = NULL; 503 } else 504 return rt; 505 506 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 507 if (err) 508 goto relookup_failed; 509 510 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev, 511 fl4_dec.saddr) == RTN_LOCAL) { 512 rt2 = __ip_route_output_key(net, &fl4_dec); 513 if (IS_ERR(rt2)) 514 err = PTR_ERR(rt2); 515 } else { 516 struct flowi4 fl4_2 = {}; 517 unsigned long orefdst; 518 519 fl4_2.daddr = fl4_dec.saddr; 520 rt2 = ip_route_output_key(net, &fl4_2); 521 if (IS_ERR(rt2)) { 522 err = PTR_ERR(rt2); 523 goto relookup_failed; 524 } 525 /* Ugh! */ 526 orefdst = skb_in->_skb_refdst; /* save old refdst */ 527 skb_dst_set(skb_in, NULL); 528 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 529 RT_TOS(tos), rt2->dst.dev); 530 531 dst_release(&rt2->dst); 532 rt2 = skb_rtable(skb_in); 533 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 534 } 535 536 if (err) 537 goto relookup_failed; 538 539 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 540 flowi4_to_flowi(&fl4_dec), NULL, 541 XFRM_LOOKUP_ICMP); 542 if (!IS_ERR(rt2)) { 543 dst_release(&rt->dst); 544 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 545 rt = rt2; 546 } else if (PTR_ERR(rt2) == -EPERM) { 547 if (rt) 548 dst_release(&rt->dst); 549 return rt2; 550 } else { 551 err = PTR_ERR(rt2); 552 goto relookup_failed; 553 } 554 return rt; 555 556 relookup_failed: 557 if (rt) 558 return rt; 559 return ERR_PTR(err); 560 } 561 562 /* 563 * Send an ICMP message in response to a situation 564 * 565 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 566 * MAY send more (we do). 567 * MUST NOT change this header information. 568 * MUST NOT reply to a multicast/broadcast IP address. 569 * MUST NOT reply to a multicast/broadcast MAC address. 570 * MUST reply to only the first fragment. 571 */ 572 573 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 574 const struct ip_options *opt) 575 { 576 struct iphdr *iph; 577 int room; 578 struct icmp_bxm icmp_param; 579 struct rtable *rt = skb_rtable(skb_in); 580 struct ipcm_cookie ipc; 581 struct flowi4 fl4; 582 __be32 saddr; 583 u8 tos; 584 u32 mark; 585 struct net *net; 586 struct sock *sk; 587 588 if (!rt) 589 goto out; 590 net = dev_net(rt->dst.dev); 591 592 /* 593 * Find the original header. It is expected to be valid, of course. 594 * Check this, icmp_send is called from the most obscure devices 595 * sometimes. 596 */ 597 iph = ip_hdr(skb_in); 598 599 if ((u8 *)iph < skb_in->head || 600 (skb_network_header(skb_in) + sizeof(*iph)) > 601 skb_tail_pointer(skb_in)) 602 goto out; 603 604 /* 605 * No replies to physical multicast/broadcast 606 */ 607 if (skb_in->pkt_type != PACKET_HOST) 608 goto out; 609 610 /* 611 * Now check at the protocol level 612 */ 613 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 614 goto out; 615 616 /* 617 * Only reply to fragment 0. We byte re-order the constant 618 * mask for efficiency. 619 */ 620 if (iph->frag_off & htons(IP_OFFSET)) 621 goto out; 622 623 /* 624 * If we send an ICMP error to an ICMP error a mess would result.. 625 */ 626 if (icmp_pointers[type].error) { 627 /* 628 * We are an error, check if we are replying to an 629 * ICMP error 630 */ 631 if (iph->protocol == IPPROTO_ICMP) { 632 u8 _inner_type, *itp; 633 634 itp = skb_header_pointer(skb_in, 635 skb_network_header(skb_in) + 636 (iph->ihl << 2) + 637 offsetof(struct icmphdr, 638 type) - 639 skb_in->data, 640 sizeof(_inner_type), 641 &_inner_type); 642 if (!itp) 643 goto out; 644 645 /* 646 * Assume any unknown ICMP type is an error. This 647 * isn't specified by the RFC, but think about it.. 648 */ 649 if (*itp > NR_ICMP_TYPES || 650 icmp_pointers[*itp].error) 651 goto out; 652 } 653 } 654 655 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 656 local_bh_disable(); 657 658 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 659 * incoming dev is loopback. If outgoing dev change to not be 660 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 661 */ 662 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 663 !icmpv4_global_allow(net, type, code)) 664 goto out_bh_enable; 665 666 sk = icmp_xmit_lock(net); 667 if (!sk) 668 goto out_bh_enable; 669 670 /* 671 * Construct source address and options. 672 */ 673 674 saddr = iph->daddr; 675 if (!(rt->rt_flags & RTCF_LOCAL)) { 676 struct net_device *dev = NULL; 677 678 rcu_read_lock(); 679 if (rt_is_input_route(rt) && 680 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 681 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 682 683 if (dev) 684 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK); 685 else 686 saddr = 0; 687 rcu_read_unlock(); 688 } 689 690 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) | 691 IPTOS_PREC_INTERNETCONTROL) : 692 iph->tos; 693 mark = IP4_REPLY_MARK(net, skb_in->mark); 694 695 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 696 goto out_unlock; 697 698 699 /* 700 * Prepare data for ICMP header. 701 */ 702 703 icmp_param.data.icmph.type = type; 704 icmp_param.data.icmph.code = code; 705 icmp_param.data.icmph.un.gateway = info; 706 icmp_param.data.icmph.checksum = 0; 707 icmp_param.skb = skb_in; 708 icmp_param.offset = skb_network_offset(skb_in); 709 inet_sk(sk)->tos = tos; 710 sk->sk_mark = mark; 711 ipcm_init(&ipc); 712 ipc.addr = iph->saddr; 713 ipc.opt = &icmp_param.replyopts.opt; 714 715 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark, 716 type, code, &icmp_param); 717 if (IS_ERR(rt)) 718 goto out_unlock; 719 720 /* peer icmp_ratelimit */ 721 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 722 goto ende; 723 724 /* RFC says return as much as we can without exceeding 576 bytes. */ 725 726 room = dst_mtu(&rt->dst); 727 if (room > 576) 728 room = 576; 729 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 730 room -= sizeof(struct icmphdr); 731 732 icmp_param.data_len = skb_in->len - icmp_param.offset; 733 if (icmp_param.data_len > room) 734 icmp_param.data_len = room; 735 icmp_param.head_len = sizeof(struct icmphdr); 736 737 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 738 ende: 739 ip_rt_put(rt); 740 out_unlock: 741 icmp_xmit_unlock(sk); 742 out_bh_enable: 743 local_bh_enable(); 744 out:; 745 } 746 EXPORT_SYMBOL(__icmp_send); 747 748 749 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 750 { 751 const struct iphdr *iph = (const struct iphdr *) skb->data; 752 const struct net_protocol *ipprot; 753 int protocol = iph->protocol; 754 755 /* Checkin full IP header plus 8 bytes of protocol to 756 * avoid additional coding at protocol handlers. 757 */ 758 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 759 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 760 return; 761 } 762 763 raw_icmp_error(skb, protocol, info); 764 765 ipprot = rcu_dereference(inet_protos[protocol]); 766 if (ipprot && ipprot->err_handler) 767 ipprot->err_handler(skb, info); 768 } 769 770 static bool icmp_tag_validation(int proto) 771 { 772 bool ok; 773 774 rcu_read_lock(); 775 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 776 rcu_read_unlock(); 777 return ok; 778 } 779 780 /* 781 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 782 * ICMP_PARAMETERPROB. 783 */ 784 785 static bool icmp_unreach(struct sk_buff *skb) 786 { 787 const struct iphdr *iph; 788 struct icmphdr *icmph; 789 struct net *net; 790 u32 info = 0; 791 792 net = dev_net(skb_dst(skb)->dev); 793 794 /* 795 * Incomplete header ? 796 * Only checks for the IP header, there should be an 797 * additional check for longer headers in upper levels. 798 */ 799 800 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 801 goto out_err; 802 803 icmph = icmp_hdr(skb); 804 iph = (const struct iphdr *)skb->data; 805 806 if (iph->ihl < 5) /* Mangled header, drop. */ 807 goto out_err; 808 809 switch (icmph->type) { 810 case ICMP_DEST_UNREACH: 811 switch (icmph->code & 15) { 812 case ICMP_NET_UNREACH: 813 case ICMP_HOST_UNREACH: 814 case ICMP_PROT_UNREACH: 815 case ICMP_PORT_UNREACH: 816 break; 817 case ICMP_FRAG_NEEDED: 818 /* for documentation of the ip_no_pmtu_disc 819 * values please see 820 * Documentation/networking/ip-sysctl.txt 821 */ 822 switch (net->ipv4.sysctl_ip_no_pmtu_disc) { 823 default: 824 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 825 &iph->daddr); 826 break; 827 case 2: 828 goto out; 829 case 3: 830 if (!icmp_tag_validation(iph->protocol)) 831 goto out; 832 /* fall through */ 833 case 0: 834 info = ntohs(icmph->un.frag.mtu); 835 } 836 break; 837 case ICMP_SR_FAILED: 838 net_dbg_ratelimited("%pI4: Source Route Failed\n", 839 &iph->daddr); 840 break; 841 default: 842 break; 843 } 844 if (icmph->code > NR_ICMP_UNREACH) 845 goto out; 846 break; 847 case ICMP_PARAMETERPROB: 848 info = ntohl(icmph->un.gateway) >> 24; 849 break; 850 case ICMP_TIME_EXCEEDED: 851 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 852 if (icmph->code == ICMP_EXC_FRAGTIME) 853 goto out; 854 break; 855 } 856 857 /* 858 * Throw it at our lower layers 859 * 860 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 861 * header. 862 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 863 * transport layer. 864 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 865 * transport layer. 866 */ 867 868 /* 869 * Check the other end isn't violating RFC 1122. Some routers send 870 * bogus responses to broadcast frames. If you see this message 871 * first check your netmask matches at both ends, if it does then 872 * get the other vendor to fix their kit. 873 */ 874 875 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 876 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 877 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 878 &ip_hdr(skb)->saddr, 879 icmph->type, icmph->code, 880 &iph->daddr, skb->dev->name); 881 goto out; 882 } 883 884 icmp_socket_deliver(skb, info); 885 886 out: 887 return true; 888 out_err: 889 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 890 return false; 891 } 892 893 894 /* 895 * Handle ICMP_REDIRECT. 896 */ 897 898 static bool icmp_redirect(struct sk_buff *skb) 899 { 900 if (skb->len < sizeof(struct iphdr)) { 901 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 902 return false; 903 } 904 905 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 906 /* there aught to be a stat */ 907 return false; 908 } 909 910 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway); 911 return true; 912 } 913 914 /* 915 * Handle ICMP_ECHO ("ping") requests. 916 * 917 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 918 * requests. 919 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 920 * included in the reply. 921 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 922 * echo requests, MUST have default=NOT. 923 * See also WRT handling of options once they are done and working. 924 */ 925 926 static bool icmp_echo(struct sk_buff *skb) 927 { 928 struct net *net; 929 930 net = dev_net(skb_dst(skb)->dev); 931 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 932 struct icmp_bxm icmp_param; 933 934 icmp_param.data.icmph = *icmp_hdr(skb); 935 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 936 icmp_param.skb = skb; 937 icmp_param.offset = 0; 938 icmp_param.data_len = skb->len; 939 icmp_param.head_len = sizeof(struct icmphdr); 940 icmp_reply(&icmp_param, skb); 941 } 942 /* should there be an ICMP stat for ignored echos? */ 943 return true; 944 } 945 946 /* 947 * Handle ICMP Timestamp requests. 948 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 949 * SHOULD be in the kernel for minimum random latency. 950 * MUST be accurate to a few minutes. 951 * MUST be updated at least at 15Hz. 952 */ 953 static bool icmp_timestamp(struct sk_buff *skb) 954 { 955 struct icmp_bxm icmp_param; 956 /* 957 * Too short. 958 */ 959 if (skb->len < 4) 960 goto out_err; 961 962 /* 963 * Fill in the current time as ms since midnight UT: 964 */ 965 icmp_param.data.times[1] = inet_current_timestamp(); 966 icmp_param.data.times[2] = icmp_param.data.times[1]; 967 968 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 969 970 icmp_param.data.icmph = *icmp_hdr(skb); 971 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 972 icmp_param.data.icmph.code = 0; 973 icmp_param.skb = skb; 974 icmp_param.offset = 0; 975 icmp_param.data_len = 0; 976 icmp_param.head_len = sizeof(struct icmphdr) + 12; 977 icmp_reply(&icmp_param, skb); 978 return true; 979 980 out_err: 981 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 982 return false; 983 } 984 985 static bool icmp_discard(struct sk_buff *skb) 986 { 987 /* pretend it was a success */ 988 return true; 989 } 990 991 /* 992 * Deal with incoming ICMP packets. 993 */ 994 int icmp_rcv(struct sk_buff *skb) 995 { 996 struct icmphdr *icmph; 997 struct rtable *rt = skb_rtable(skb); 998 struct net *net = dev_net(rt->dst.dev); 999 bool success; 1000 1001 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1002 struct sec_path *sp = skb_sec_path(skb); 1003 int nh; 1004 1005 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1006 XFRM_STATE_ICMP)) 1007 goto drop; 1008 1009 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1010 goto drop; 1011 1012 nh = skb_network_offset(skb); 1013 skb_set_network_header(skb, sizeof(*icmph)); 1014 1015 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 1016 goto drop; 1017 1018 skb_set_network_header(skb, nh); 1019 } 1020 1021 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1022 1023 if (skb_checksum_simple_validate(skb)) 1024 goto csum_error; 1025 1026 if (!pskb_pull(skb, sizeof(*icmph))) 1027 goto error; 1028 1029 icmph = icmp_hdr(skb); 1030 1031 ICMPMSGIN_INC_STATS(net, icmph->type); 1032 /* 1033 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1034 * 1035 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1036 * discarded. 1037 */ 1038 if (icmph->type > NR_ICMP_TYPES) 1039 goto error; 1040 1041 1042 /* 1043 * Parse the ICMP message 1044 */ 1045 1046 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1047 /* 1048 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1049 * silently ignored (we let user decide with a sysctl). 1050 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1051 * discarded if to broadcast/multicast. 1052 */ 1053 if ((icmph->type == ICMP_ECHO || 1054 icmph->type == ICMP_TIMESTAMP) && 1055 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1056 goto error; 1057 } 1058 if (icmph->type != ICMP_ECHO && 1059 icmph->type != ICMP_TIMESTAMP && 1060 icmph->type != ICMP_ADDRESS && 1061 icmph->type != ICMP_ADDRESSREPLY) { 1062 goto error; 1063 } 1064 } 1065 1066 success = icmp_pointers[icmph->type].handler(skb); 1067 1068 if (success) { 1069 consume_skb(skb); 1070 return NET_RX_SUCCESS; 1071 } 1072 1073 drop: 1074 kfree_skb(skb); 1075 return NET_RX_DROP; 1076 csum_error: 1077 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1078 error: 1079 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1080 goto drop; 1081 } 1082 1083 int icmp_err(struct sk_buff *skb, u32 info) 1084 { 1085 struct iphdr *iph = (struct iphdr *)skb->data; 1086 int offset = iph->ihl<<2; 1087 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1088 int type = icmp_hdr(skb)->type; 1089 int code = icmp_hdr(skb)->code; 1090 struct net *net = dev_net(skb->dev); 1091 1092 /* 1093 * Use ping_err to handle all icmp errors except those 1094 * triggered by ICMP_ECHOREPLY which sent from kernel. 1095 */ 1096 if (icmph->type != ICMP_ECHOREPLY) { 1097 ping_err(skb, offset, info); 1098 return 0; 1099 } 1100 1101 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1102 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1103 else if (type == ICMP_REDIRECT) 1104 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1105 1106 return 0; 1107 } 1108 1109 /* 1110 * This table is the definition of how we handle ICMP. 1111 */ 1112 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1113 [ICMP_ECHOREPLY] = { 1114 .handler = ping_rcv, 1115 }, 1116 [1] = { 1117 .handler = icmp_discard, 1118 .error = 1, 1119 }, 1120 [2] = { 1121 .handler = icmp_discard, 1122 .error = 1, 1123 }, 1124 [ICMP_DEST_UNREACH] = { 1125 .handler = icmp_unreach, 1126 .error = 1, 1127 }, 1128 [ICMP_SOURCE_QUENCH] = { 1129 .handler = icmp_unreach, 1130 .error = 1, 1131 }, 1132 [ICMP_REDIRECT] = { 1133 .handler = icmp_redirect, 1134 .error = 1, 1135 }, 1136 [6] = { 1137 .handler = icmp_discard, 1138 .error = 1, 1139 }, 1140 [7] = { 1141 .handler = icmp_discard, 1142 .error = 1, 1143 }, 1144 [ICMP_ECHO] = { 1145 .handler = icmp_echo, 1146 }, 1147 [9] = { 1148 .handler = icmp_discard, 1149 .error = 1, 1150 }, 1151 [10] = { 1152 .handler = icmp_discard, 1153 .error = 1, 1154 }, 1155 [ICMP_TIME_EXCEEDED] = { 1156 .handler = icmp_unreach, 1157 .error = 1, 1158 }, 1159 [ICMP_PARAMETERPROB] = { 1160 .handler = icmp_unreach, 1161 .error = 1, 1162 }, 1163 [ICMP_TIMESTAMP] = { 1164 .handler = icmp_timestamp, 1165 }, 1166 [ICMP_TIMESTAMPREPLY] = { 1167 .handler = icmp_discard, 1168 }, 1169 [ICMP_INFO_REQUEST] = { 1170 .handler = icmp_discard, 1171 }, 1172 [ICMP_INFO_REPLY] = { 1173 .handler = icmp_discard, 1174 }, 1175 [ICMP_ADDRESS] = { 1176 .handler = icmp_discard, 1177 }, 1178 [ICMP_ADDRESSREPLY] = { 1179 .handler = icmp_discard, 1180 }, 1181 }; 1182 1183 static void __net_exit icmp_sk_exit(struct net *net) 1184 { 1185 int i; 1186 1187 for_each_possible_cpu(i) 1188 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i)); 1189 free_percpu(net->ipv4.icmp_sk); 1190 net->ipv4.icmp_sk = NULL; 1191 } 1192 1193 static int __net_init icmp_sk_init(struct net *net) 1194 { 1195 int i, err; 1196 1197 net->ipv4.icmp_sk = alloc_percpu(struct sock *); 1198 if (!net->ipv4.icmp_sk) 1199 return -ENOMEM; 1200 1201 for_each_possible_cpu(i) { 1202 struct sock *sk; 1203 1204 err = inet_ctl_sock_create(&sk, PF_INET, 1205 SOCK_RAW, IPPROTO_ICMP, net); 1206 if (err < 0) 1207 goto fail; 1208 1209 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk; 1210 1211 /* Enough space for 2 64K ICMP packets, including 1212 * sk_buff/skb_shared_info struct overhead. 1213 */ 1214 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1215 1216 /* 1217 * Speedup sock_wfree() 1218 */ 1219 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1220 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1221 } 1222 1223 /* Control parameters for ECHO replies. */ 1224 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1225 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1226 1227 /* Control parameter - ignore bogus broadcast responses? */ 1228 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1229 1230 /* 1231 * Configurable global rate limit. 1232 * 1233 * ratelimit defines tokens/packet consumed for dst->rate_token 1234 * bucket ratemask defines which icmp types are ratelimited by 1235 * setting it's bit position. 1236 * 1237 * default: 1238 * dest unreachable (3), source quench (4), 1239 * time exceeded (11), parameter problem (12) 1240 */ 1241 1242 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1243 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1244 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1245 1246 return 0; 1247 1248 fail: 1249 icmp_sk_exit(net); 1250 return err; 1251 } 1252 1253 static struct pernet_operations __net_initdata icmp_sk_ops = { 1254 .init = icmp_sk_init, 1255 .exit = icmp_sk_exit, 1256 }; 1257 1258 int __init icmp_init(void) 1259 { 1260 return register_pernet_subsys(&icmp_sk_ops); 1261 } 1262