1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * NET3: Implementation of the ICMP protocol layer. 4 * 5 * Alan Cox, <alan@lxorguk.ukuu.org.uk> 6 * 7 * Some of the function names and the icmp unreach table for this 8 * module were derived from [icmp.c 1.0.11 06/02/93] by 9 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting. 10 * Other than that this module is a complete rewrite. 11 * 12 * Fixes: 13 * Clemens Fruhwirth : introduce global icmp rate limiting 14 * with icmp type masking ability instead 15 * of broken per type icmp timeouts. 16 * Mike Shaver : RFC1122 checks. 17 * Alan Cox : Multicast ping reply as self. 18 * Alan Cox : Fix atomicity lockup in ip_build_xmit 19 * call. 20 * Alan Cox : Added 216,128 byte paths to the MTU 21 * code. 22 * Martin Mares : RFC1812 checks. 23 * Martin Mares : Can be configured to follow redirects 24 * if acting as a router _without_ a 25 * routing protocol (RFC 1812). 26 * Martin Mares : Echo requests may be configured to 27 * be ignored (RFC 1812). 28 * Martin Mares : Limitation of ICMP error message 29 * transmit rate (RFC 1812). 30 * Martin Mares : TOS and Precedence set correctly 31 * (RFC 1812). 32 * Martin Mares : Now copying as much data from the 33 * original packet as we can without 34 * exceeding 576 bytes (RFC 1812). 35 * Willy Konynenberg : Transparent proxying support. 36 * Keith Owens : RFC1191 correction for 4.2BSD based 37 * path MTU bug. 38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are 39 * valid (RFC 1812). 40 * Andi Kleen : Check all packet lengths properly 41 * and moved all kfree_skb() up to 42 * icmp_rcv. 43 * Andi Kleen : Move the rate limit bookkeeping 44 * into the dest entry and use a token 45 * bucket filter (thanks to ANK). Make 46 * the rates sysctl configurable. 47 * Yu Tianli : Fixed two ugly bugs in icmp_send 48 * - IP option length was accounted wrongly 49 * - ICMP header length was not accounted 50 * at all. 51 * Tristan Greaves : Added sysctl option to ignore bogus 52 * broadcast responses from broken routers. 53 * 54 * To Fix: 55 * 56 * - Should use skb_pull() instead of all the manual checking. 57 * This would also greatly simply some upper layer error handlers. --AK 58 */ 59 60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 61 62 #include <linux/module.h> 63 #include <linux/types.h> 64 #include <linux/jiffies.h> 65 #include <linux/kernel.h> 66 #include <linux/fcntl.h> 67 #include <linux/socket.h> 68 #include <linux/in.h> 69 #include <linux/inet.h> 70 #include <linux/inetdevice.h> 71 #include <linux/netdevice.h> 72 #include <linux/string.h> 73 #include <linux/netfilter_ipv4.h> 74 #include <linux/slab.h> 75 #include <net/snmp.h> 76 #include <net/ip.h> 77 #include <net/route.h> 78 #include <net/protocol.h> 79 #include <net/icmp.h> 80 #include <net/tcp.h> 81 #include <net/udp.h> 82 #include <net/raw.h> 83 #include <net/ping.h> 84 #include <linux/skbuff.h> 85 #include <net/sock.h> 86 #include <linux/errno.h> 87 #include <linux/timer.h> 88 #include <linux/init.h> 89 #include <linux/uaccess.h> 90 #include <net/checksum.h> 91 #include <net/xfrm.h> 92 #include <net/inet_common.h> 93 #include <net/ip_fib.h> 94 #include <net/l3mdev.h> 95 #include <net/addrconf.h> 96 #include <net/inet_dscp.h> 97 #define CREATE_TRACE_POINTS 98 #include <trace/events/icmp.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 enum skb_drop_reason (*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 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk); 200 201 /* Called with BH disabled */ 202 static inline struct sock *icmp_xmit_lock(struct net *net) 203 { 204 struct sock *sk; 205 206 sk = this_cpu_read(ipv4_icmp_sk); 207 208 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 209 /* This can happen if the output path signals a 210 * dst_link_failure() for an outgoing ICMP packet. 211 */ 212 return NULL; 213 } 214 sock_net_set(sk, net); 215 return sk; 216 } 217 218 static inline void icmp_xmit_unlock(struct sock *sk) 219 { 220 sock_net_set(sk, &init_net); 221 spin_unlock(&sk->sk_lock.slock); 222 } 223 224 int sysctl_icmp_msgs_per_sec __read_mostly = 1000; 225 int sysctl_icmp_msgs_burst __read_mostly = 50; 226 227 static struct { 228 atomic_t credit; 229 u32 stamp; 230 } icmp_global; 231 232 /** 233 * icmp_global_allow - Are we allowed to send one more ICMP message ? 234 * 235 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec. 236 * Returns false if we reached the limit and can not send another packet. 237 * Works in tandem with icmp_global_consume(). 238 */ 239 bool icmp_global_allow(void) 240 { 241 u32 delta, now, oldstamp; 242 int incr, new, old; 243 244 /* Note: many cpus could find this condition true. 245 * Then later icmp_global_consume() could consume more credits, 246 * this is an acceptable race. 247 */ 248 if (atomic_read(&icmp_global.credit) > 0) 249 return true; 250 251 now = jiffies; 252 oldstamp = READ_ONCE(icmp_global.stamp); 253 delta = min_t(u32, now - oldstamp, HZ); 254 if (delta < HZ / 50) 255 return false; 256 257 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ; 258 if (!incr) 259 return false; 260 261 if (cmpxchg(&icmp_global.stamp, oldstamp, now) == oldstamp) { 262 old = atomic_read(&icmp_global.credit); 263 do { 264 new = min(old + incr, READ_ONCE(sysctl_icmp_msgs_burst)); 265 } while (!atomic_try_cmpxchg(&icmp_global.credit, &old, new)); 266 } 267 return true; 268 } 269 EXPORT_SYMBOL(icmp_global_allow); 270 271 void icmp_global_consume(void) 272 { 273 int credits = get_random_u32_below(3); 274 275 /* Note: this might make icmp_global.credit negative. */ 276 if (credits) 277 atomic_sub(credits, &icmp_global.credit); 278 } 279 EXPORT_SYMBOL(icmp_global_consume); 280 281 static bool icmpv4_mask_allow(struct net *net, int type, int code) 282 { 283 if (type > NR_ICMP_TYPES) 284 return true; 285 286 /* Don't limit PMTU discovery. */ 287 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 288 return true; 289 290 /* Limit if icmp type is enabled in ratemask. */ 291 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask))) 292 return true; 293 294 return false; 295 } 296 297 static bool icmpv4_global_allow(struct net *net, int type, int code, 298 bool *apply_ratelimit) 299 { 300 if (icmpv4_mask_allow(net, type, code)) 301 return true; 302 303 if (icmp_global_allow()) { 304 *apply_ratelimit = true; 305 return true; 306 } 307 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL); 308 return false; 309 } 310 311 /* 312 * Send an ICMP frame. 313 */ 314 315 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 316 struct flowi4 *fl4, int type, int code, 317 bool apply_ratelimit) 318 { 319 struct dst_entry *dst = &rt->dst; 320 struct inet_peer *peer; 321 bool rc = true; 322 323 if (!apply_ratelimit) 324 return true; 325 326 /* No rate limit on loopback */ 327 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 328 goto out; 329 330 rcu_read_lock(); 331 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 332 l3mdev_master_ifindex_rcu(dst->dev)); 333 rc = inet_peer_xrlim_allow(peer, 334 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit)); 335 rcu_read_unlock(); 336 out: 337 if (!rc) 338 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST); 339 else 340 icmp_global_consume(); 341 return rc; 342 } 343 344 /* 345 * Maintain the counters used in the SNMP statistics for outgoing ICMP 346 */ 347 void icmp_out_count(struct net *net, unsigned char type) 348 { 349 ICMPMSGOUT_INC_STATS(net, type); 350 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 351 } 352 353 /* 354 * Checksum each fragment, and on the first include the headers and final 355 * checksum. 356 */ 357 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 358 struct sk_buff *skb) 359 { 360 struct icmp_bxm *icmp_param = from; 361 __wsum csum; 362 363 csum = skb_copy_and_csum_bits(icmp_param->skb, 364 icmp_param->offset + offset, 365 to, len); 366 367 skb->csum = csum_block_add(skb->csum, csum, odd); 368 if (icmp_pointers[icmp_param->data.icmph.type].error) 369 nf_ct_attach(skb, icmp_param->skb); 370 return 0; 371 } 372 373 static void icmp_push_reply(struct sock *sk, 374 struct icmp_bxm *icmp_param, 375 struct flowi4 *fl4, 376 struct ipcm_cookie *ipc, struct rtable **rt) 377 { 378 struct sk_buff *skb; 379 380 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 381 icmp_param->data_len+icmp_param->head_len, 382 icmp_param->head_len, 383 ipc, rt, MSG_DONTWAIT) < 0) { 384 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 385 ip_flush_pending_frames(sk); 386 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 387 struct icmphdr *icmph = icmp_hdr(skb); 388 __wsum csum; 389 struct sk_buff *skb1; 390 391 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 392 (char *)icmph, 393 icmp_param->head_len); 394 skb_queue_walk(&sk->sk_write_queue, skb1) { 395 csum = csum_add(csum, skb1->csum); 396 } 397 icmph->checksum = csum_fold(csum); 398 skb->ip_summed = CHECKSUM_NONE; 399 ip_push_pending_frames(sk, fl4); 400 } 401 } 402 403 /* 404 * Driving logic for building and sending ICMP messages. 405 */ 406 407 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 408 { 409 struct rtable *rt = skb_rtable(skb); 410 struct net *net = dev_net_rcu(rt->dst.dev); 411 bool apply_ratelimit = false; 412 struct ipcm_cookie ipc; 413 struct flowi4 fl4; 414 struct sock *sk; 415 struct inet_sock *inet; 416 __be32 daddr, saddr; 417 u32 mark = IP4_REPLY_MARK(net, skb->mark); 418 int type = icmp_param->data.icmph.type; 419 int code = icmp_param->data.icmph.code; 420 421 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 422 return; 423 424 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 425 local_bh_disable(); 426 427 /* is global icmp_msgs_per_sec exhausted ? */ 428 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit)) 429 goto out_bh_enable; 430 431 sk = icmp_xmit_lock(net); 432 if (!sk) 433 goto out_bh_enable; 434 inet = inet_sk(sk); 435 436 icmp_param->data.icmph.checksum = 0; 437 438 ipcm_init(&ipc); 439 inet->tos = ip_hdr(skb)->tos; 440 ipc.sockc.mark = mark; 441 daddr = ipc.addr = ip_hdr(skb)->saddr; 442 saddr = fib_compute_spec_dst(skb); 443 444 if (icmp_param->replyopts.opt.opt.optlen) { 445 ipc.opt = &icmp_param->replyopts.opt; 446 if (ipc.opt->opt.srr) 447 daddr = icmp_param->replyopts.opt.opt.faddr; 448 } 449 memset(&fl4, 0, sizeof(fl4)); 450 fl4.daddr = daddr; 451 fl4.saddr = saddr; 452 fl4.flowi4_mark = mark; 453 fl4.flowi4_uid = sock_net_uid(net, NULL); 454 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 455 fl4.flowi4_proto = IPPROTO_ICMP; 456 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 457 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 458 rt = ip_route_output_key(net, &fl4); 459 if (IS_ERR(rt)) 460 goto out_unlock; 461 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 462 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt); 463 ip_rt_put(rt); 464 out_unlock: 465 icmp_xmit_unlock(sk); 466 out_bh_enable: 467 local_bh_enable(); 468 } 469 470 /* 471 * The device used for looking up which routing table to use for sending an ICMP 472 * error is preferably the source whenever it is set, which should ensure the 473 * icmp error can be sent to the source host, else lookup using the routing 474 * table of the destination device, else use the main routing table (index 0). 475 */ 476 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 477 { 478 struct net_device *route_lookup_dev = NULL; 479 480 if (skb->dev) 481 route_lookup_dev = skb->dev; 482 else if (skb_dst(skb)) 483 route_lookup_dev = skb_dst(skb)->dev; 484 return route_lookup_dev; 485 } 486 487 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4, 488 struct sk_buff *skb_in, 489 const struct iphdr *iph, __be32 saddr, 490 dscp_t dscp, u32 mark, int type, 491 int code, struct icmp_bxm *param) 492 { 493 struct net_device *route_lookup_dev; 494 struct rtable *rt, *rt2; 495 struct flowi4 fl4_dec; 496 int err; 497 498 memset(fl4, 0, sizeof(*fl4)); 499 fl4->daddr = (param->replyopts.opt.opt.srr ? 500 param->replyopts.opt.opt.faddr : iph->saddr); 501 fl4->saddr = saddr; 502 fl4->flowi4_mark = mark; 503 fl4->flowi4_uid = sock_net_uid(net, NULL); 504 fl4->flowi4_tos = inet_dscp_to_dsfield(dscp); 505 fl4->flowi4_proto = IPPROTO_ICMP; 506 fl4->fl4_icmp_type = type; 507 fl4->fl4_icmp_code = code; 508 route_lookup_dev = icmp_get_route_lookup_dev(skb_in); 509 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev); 510 511 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4)); 512 rt = ip_route_output_key_hash(net, fl4, skb_in); 513 if (IS_ERR(rt)) 514 return rt; 515 516 /* No need to clone since we're just using its address. */ 517 rt2 = rt; 518 519 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 520 flowi4_to_flowi(fl4), NULL, 0); 521 if (!IS_ERR(rt)) { 522 if (rt != rt2) 523 return rt; 524 } else if (PTR_ERR(rt) == -EPERM) { 525 rt = NULL; 526 } else 527 return rt; 528 529 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 530 if (err) 531 goto relookup_failed; 532 533 if (inet_addr_type_dev_table(net, route_lookup_dev, 534 fl4_dec.saddr) == RTN_LOCAL) { 535 rt2 = __ip_route_output_key(net, &fl4_dec); 536 if (IS_ERR(rt2)) 537 err = PTR_ERR(rt2); 538 } else { 539 struct flowi4 fl4_2 = {}; 540 unsigned long orefdst; 541 542 fl4_2.daddr = fl4_dec.saddr; 543 rt2 = ip_route_output_key(net, &fl4_2); 544 if (IS_ERR(rt2)) { 545 err = PTR_ERR(rt2); 546 goto relookup_failed; 547 } 548 /* Ugh! */ 549 orefdst = skb_in->_skb_refdst; /* save old refdst */ 550 skb_dst_set(skb_in, NULL); 551 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 552 dscp, rt2->dst.dev); 553 554 dst_release(&rt2->dst); 555 rt2 = skb_rtable(skb_in); 556 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 557 } 558 559 if (err) 560 goto relookup_failed; 561 562 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 563 flowi4_to_flowi(&fl4_dec), NULL, 564 XFRM_LOOKUP_ICMP); 565 if (!IS_ERR(rt2)) { 566 dst_release(&rt->dst); 567 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 568 rt = rt2; 569 } else if (PTR_ERR(rt2) == -EPERM) { 570 if (rt) 571 dst_release(&rt->dst); 572 return rt2; 573 } else { 574 err = PTR_ERR(rt2); 575 goto relookup_failed; 576 } 577 return rt; 578 579 relookup_failed: 580 if (rt) 581 return rt; 582 return ERR_PTR(err); 583 } 584 585 /* 586 * Send an ICMP message in response to a situation 587 * 588 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 589 * MAY send more (we do). 590 * MUST NOT change this header information. 591 * MUST NOT reply to a multicast/broadcast IP address. 592 * MUST NOT reply to a multicast/broadcast MAC address. 593 * MUST reply to only the first fragment. 594 */ 595 596 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 597 const struct ip_options *opt) 598 { 599 struct iphdr *iph; 600 int room; 601 struct icmp_bxm icmp_param; 602 struct rtable *rt = skb_rtable(skb_in); 603 bool apply_ratelimit = false; 604 struct ipcm_cookie ipc; 605 struct flowi4 fl4; 606 __be32 saddr; 607 u8 tos; 608 u32 mark; 609 struct net *net; 610 struct sock *sk; 611 612 if (!rt) 613 return; 614 615 rcu_read_lock(); 616 617 if (rt->dst.dev) 618 net = dev_net_rcu(rt->dst.dev); 619 else if (skb_in->dev) 620 net = dev_net_rcu(skb_in->dev); 621 else 622 goto out; 623 624 /* 625 * Find the original header. It is expected to be valid, of course. 626 * Check this, icmp_send is called from the most obscure devices 627 * sometimes. 628 */ 629 iph = ip_hdr(skb_in); 630 631 if ((u8 *)iph < skb_in->head || 632 (skb_network_header(skb_in) + sizeof(*iph)) > 633 skb_tail_pointer(skb_in)) 634 goto out; 635 636 /* 637 * No replies to physical multicast/broadcast 638 */ 639 if (skb_in->pkt_type != PACKET_HOST) 640 goto out; 641 642 /* 643 * Now check at the protocol level 644 */ 645 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 646 goto out; 647 648 /* 649 * Only reply to fragment 0. We byte re-order the constant 650 * mask for efficiency. 651 */ 652 if (iph->frag_off & htons(IP_OFFSET)) 653 goto out; 654 655 /* 656 * If we send an ICMP error to an ICMP error a mess would result.. 657 */ 658 if (icmp_pointers[type].error) { 659 /* 660 * We are an error, check if we are replying to an 661 * ICMP error 662 */ 663 if (iph->protocol == IPPROTO_ICMP) { 664 u8 _inner_type, *itp; 665 666 itp = skb_header_pointer(skb_in, 667 skb_network_header(skb_in) + 668 (iph->ihl << 2) + 669 offsetof(struct icmphdr, 670 type) - 671 skb_in->data, 672 sizeof(_inner_type), 673 &_inner_type); 674 if (!itp) 675 goto out; 676 677 /* 678 * Assume any unknown ICMP type is an error. This 679 * isn't specified by the RFC, but think about it.. 680 */ 681 if (*itp > NR_ICMP_TYPES || 682 icmp_pointers[*itp].error) 683 goto out; 684 } 685 } 686 687 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 688 local_bh_disable(); 689 690 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 691 * incoming dev is loopback. If outgoing dev change to not be 692 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 693 */ 694 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 695 !icmpv4_global_allow(net, type, code, &apply_ratelimit)) 696 goto out_bh_enable; 697 698 sk = icmp_xmit_lock(net); 699 if (!sk) 700 goto out_bh_enable; 701 702 /* 703 * Construct source address and options. 704 */ 705 706 saddr = iph->daddr; 707 if (!(rt->rt_flags & RTCF_LOCAL)) { 708 struct net_device *dev = NULL; 709 710 rcu_read_lock(); 711 if (rt_is_input_route(rt) && 712 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)) 713 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 714 715 if (dev) 716 saddr = inet_select_addr(dev, iph->saddr, 717 RT_SCOPE_LINK); 718 else 719 saddr = 0; 720 rcu_read_unlock(); 721 } 722 723 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 724 IPTOS_PREC_INTERNETCONTROL) : 725 iph->tos; 726 mark = IP4_REPLY_MARK(net, skb_in->mark); 727 728 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 729 goto out_unlock; 730 731 732 /* 733 * Prepare data for ICMP header. 734 */ 735 736 icmp_param.data.icmph.type = type; 737 icmp_param.data.icmph.code = code; 738 icmp_param.data.icmph.un.gateway = info; 739 icmp_param.data.icmph.checksum = 0; 740 icmp_param.skb = skb_in; 741 icmp_param.offset = skb_network_offset(skb_in); 742 inet_sk(sk)->tos = tos; 743 ipcm_init(&ipc); 744 ipc.addr = iph->saddr; 745 ipc.opt = &icmp_param.replyopts.opt; 746 ipc.sockc.mark = mark; 747 748 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, 749 inet_dsfield_to_dscp(tos), mark, type, code, 750 &icmp_param); 751 if (IS_ERR(rt)) 752 goto out_unlock; 753 754 /* peer icmp_ratelimit */ 755 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 756 goto ende; 757 758 /* RFC says return as much as we can without exceeding 576 bytes. */ 759 760 room = dst_mtu(&rt->dst); 761 if (room > 576) 762 room = 576; 763 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 764 room -= sizeof(struct icmphdr); 765 /* Guard against tiny mtu. We need to include at least one 766 * IP network header for this message to make any sense. 767 */ 768 if (room <= (int)sizeof(struct iphdr)) 769 goto ende; 770 771 icmp_param.data_len = skb_in->len - icmp_param.offset; 772 if (icmp_param.data_len > room) 773 icmp_param.data_len = room; 774 icmp_param.head_len = sizeof(struct icmphdr); 775 776 /* if we don't have a source address at this point, fall back to the 777 * dummy address instead of sending out a packet with a source address 778 * of 0.0.0.0 779 */ 780 if (!fl4.saddr) 781 fl4.saddr = htonl(INADDR_DUMMY); 782 783 trace_icmp_send(skb_in, type, code); 784 785 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt); 786 ende: 787 ip_rt_put(rt); 788 out_unlock: 789 icmp_xmit_unlock(sk); 790 out_bh_enable: 791 local_bh_enable(); 792 out: 793 rcu_read_unlock(); 794 } 795 EXPORT_SYMBOL(__icmp_send); 796 797 #if IS_ENABLED(CONFIG_NF_NAT) 798 #include <net/netfilter/nf_conntrack.h> 799 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 800 { 801 struct sk_buff *cloned_skb = NULL; 802 struct ip_options opts = { 0 }; 803 enum ip_conntrack_info ctinfo; 804 struct nf_conn *ct; 805 __be32 orig_ip; 806 807 ct = nf_ct_get(skb_in, &ctinfo); 808 if (!ct || !(ct->status & IPS_SRC_NAT)) { 809 __icmp_send(skb_in, type, code, info, &opts); 810 return; 811 } 812 813 if (skb_shared(skb_in)) 814 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 815 816 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 817 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 818 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 819 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 820 goto out; 821 822 orig_ip = ip_hdr(skb_in)->saddr; 823 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 824 __icmp_send(skb_in, type, code, info, &opts); 825 ip_hdr(skb_in)->saddr = orig_ip; 826 out: 827 consume_skb(cloned_skb); 828 } 829 EXPORT_SYMBOL(icmp_ndo_send); 830 #endif 831 832 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 833 { 834 const struct iphdr *iph = (const struct iphdr *)skb->data; 835 const struct net_protocol *ipprot; 836 int protocol = iph->protocol; 837 838 /* Checkin full IP header plus 8 bytes of protocol to 839 * avoid additional coding at protocol handlers. 840 */ 841 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 842 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 843 return; 844 } 845 846 raw_icmp_error(skb, protocol, info); 847 848 ipprot = rcu_dereference(inet_protos[protocol]); 849 if (ipprot && ipprot->err_handler) 850 ipprot->err_handler(skb, info); 851 } 852 853 static bool icmp_tag_validation(int proto) 854 { 855 bool ok; 856 857 rcu_read_lock(); 858 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 859 rcu_read_unlock(); 860 return ok; 861 } 862 863 /* 864 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 865 * ICMP_PARAMETERPROB. 866 */ 867 868 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb) 869 { 870 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET; 871 const struct iphdr *iph; 872 struct icmphdr *icmph; 873 struct net *net; 874 u32 info = 0; 875 876 net = dev_net_rcu(skb_dst(skb)->dev); 877 878 /* 879 * Incomplete header ? 880 * Only checks for the IP header, there should be an 881 * additional check for longer headers in upper levels. 882 */ 883 884 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 885 goto out_err; 886 887 icmph = icmp_hdr(skb); 888 iph = (const struct iphdr *)skb->data; 889 890 if (iph->ihl < 5) { /* Mangled header, drop. */ 891 reason = SKB_DROP_REASON_IP_INHDR; 892 goto out_err; 893 } 894 895 switch (icmph->type) { 896 case ICMP_DEST_UNREACH: 897 switch (icmph->code & 15) { 898 case ICMP_NET_UNREACH: 899 case ICMP_HOST_UNREACH: 900 case ICMP_PROT_UNREACH: 901 case ICMP_PORT_UNREACH: 902 break; 903 case ICMP_FRAG_NEEDED: 904 /* for documentation of the ip_no_pmtu_disc 905 * values please see 906 * Documentation/networking/ip-sysctl.rst 907 */ 908 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) { 909 default: 910 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 911 &iph->daddr); 912 break; 913 case 2: 914 goto out; 915 case 3: 916 if (!icmp_tag_validation(iph->protocol)) 917 goto out; 918 fallthrough; 919 case 0: 920 info = ntohs(icmph->un.frag.mtu); 921 } 922 break; 923 case ICMP_SR_FAILED: 924 net_dbg_ratelimited("%pI4: Source Route Failed\n", 925 &iph->daddr); 926 break; 927 default: 928 break; 929 } 930 if (icmph->code > NR_ICMP_UNREACH) 931 goto out; 932 break; 933 case ICMP_PARAMETERPROB: 934 info = ntohl(icmph->un.gateway) >> 24; 935 break; 936 case ICMP_TIME_EXCEEDED: 937 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 938 if (icmph->code == ICMP_EXC_FRAGTIME) 939 goto out; 940 break; 941 } 942 943 /* 944 * Throw it at our lower layers 945 * 946 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 947 * header. 948 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 949 * transport layer. 950 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 951 * transport layer. 952 */ 953 954 /* 955 * Check the other end isn't violating RFC 1122. Some routers send 956 * bogus responses to broadcast frames. If you see this message 957 * first check your netmask matches at both ends, if it does then 958 * get the other vendor to fix their kit. 959 */ 960 961 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) && 962 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 963 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 964 &ip_hdr(skb)->saddr, 965 icmph->type, icmph->code, 966 &iph->daddr, skb->dev->name); 967 goto out; 968 } 969 970 icmp_socket_deliver(skb, info); 971 972 out: 973 return reason; 974 out_err: 975 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 976 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED; 977 } 978 979 980 /* 981 * Handle ICMP_REDIRECT. 982 */ 983 984 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb) 985 { 986 if (skb->len < sizeof(struct iphdr)) { 987 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 988 return SKB_DROP_REASON_PKT_TOO_SMALL; 989 } 990 991 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 992 /* there aught to be a stat */ 993 return SKB_DROP_REASON_NOMEM; 994 } 995 996 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 997 return SKB_NOT_DROPPED_YET; 998 } 999 1000 /* 1001 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests. 1002 * 1003 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 1004 * requests. 1005 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 1006 * included in the reply. 1007 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 1008 * echo requests, MUST have default=NOT. 1009 * RFC 8335: 8 MUST have a config option to enable/disable ICMP 1010 * Extended Echo Functionality, MUST be disabled by default 1011 * See also WRT handling of options once they are done and working. 1012 */ 1013 1014 static enum skb_drop_reason icmp_echo(struct sk_buff *skb) 1015 { 1016 struct icmp_bxm icmp_param; 1017 struct net *net; 1018 1019 net = dev_net_rcu(skb_dst(skb)->dev); 1020 /* should there be an ICMP stat for ignored echos? */ 1021 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all)) 1022 return SKB_NOT_DROPPED_YET; 1023 1024 icmp_param.data.icmph = *icmp_hdr(skb); 1025 icmp_param.skb = skb; 1026 icmp_param.offset = 0; 1027 icmp_param.data_len = skb->len; 1028 icmp_param.head_len = sizeof(struct icmphdr); 1029 1030 if (icmp_param.data.icmph.type == ICMP_ECHO) 1031 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 1032 else if (!icmp_build_probe(skb, &icmp_param.data.icmph)) 1033 return SKB_NOT_DROPPED_YET; 1034 1035 icmp_reply(&icmp_param, skb); 1036 return SKB_NOT_DROPPED_YET; 1037 } 1038 1039 /* Helper for icmp_echo and icmpv6_echo_reply. 1040 * Searches for net_device that matches PROBE interface identifier 1041 * and builds PROBE reply message in icmphdr. 1042 * 1043 * Returns false if PROBE responses are disabled via sysctl 1044 */ 1045 1046 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr) 1047 { 1048 struct net *net = dev_net_rcu(skb->dev); 1049 struct icmp_ext_hdr *ext_hdr, _ext_hdr; 1050 struct icmp_ext_echo_iio *iio, _iio; 1051 struct inet6_dev *in6_dev; 1052 struct in_device *in_dev; 1053 struct net_device *dev; 1054 char buff[IFNAMSIZ]; 1055 u16 ident_len; 1056 u8 status; 1057 1058 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe)) 1059 return false; 1060 1061 /* We currently only support probing interfaces on the proxy node 1062 * Check to ensure L-bit is set 1063 */ 1064 if (!(ntohs(icmphdr->un.echo.sequence) & 1)) 1065 return false; 1066 /* Clear status bits in reply message */ 1067 icmphdr->un.echo.sequence &= htons(0xFF00); 1068 if (icmphdr->type == ICMP_EXT_ECHO) 1069 icmphdr->type = ICMP_EXT_ECHOREPLY; 1070 else 1071 icmphdr->type = ICMPV6_EXT_ECHO_REPLY; 1072 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr); 1073 /* Size of iio is class_type dependent. 1074 * Only check header here and assign length based on ctype in the switch statement 1075 */ 1076 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio); 1077 if (!ext_hdr || !iio) 1078 goto send_mal_query; 1079 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) || 1080 ntohs(iio->extobj_hdr.length) > sizeof(_iio)) 1081 goto send_mal_query; 1082 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr); 1083 iio = skb_header_pointer(skb, sizeof(_ext_hdr), 1084 sizeof(iio->extobj_hdr) + ident_len, &_iio); 1085 if (!iio) 1086 goto send_mal_query; 1087 1088 status = 0; 1089 dev = NULL; 1090 switch (iio->extobj_hdr.class_type) { 1091 case ICMP_EXT_ECHO_CTYPE_NAME: 1092 if (ident_len >= IFNAMSIZ) 1093 goto send_mal_query; 1094 memset(buff, 0, sizeof(buff)); 1095 memcpy(buff, &iio->ident.name, ident_len); 1096 dev = dev_get_by_name(net, buff); 1097 break; 1098 case ICMP_EXT_ECHO_CTYPE_INDEX: 1099 if (ident_len != sizeof(iio->ident.ifindex)) 1100 goto send_mal_query; 1101 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex)); 1102 break; 1103 case ICMP_EXT_ECHO_CTYPE_ADDR: 1104 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) || 1105 ident_len != sizeof(iio->ident.addr.ctype3_hdr) + 1106 iio->ident.addr.ctype3_hdr.addrlen) 1107 goto send_mal_query; 1108 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) { 1109 case ICMP_AFI_IP: 1110 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr)) 1111 goto send_mal_query; 1112 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr); 1113 break; 1114 #if IS_ENABLED(CONFIG_IPV6) 1115 case ICMP_AFI_IP6: 1116 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr)) 1117 goto send_mal_query; 1118 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev); 1119 dev_hold(dev); 1120 break; 1121 #endif 1122 default: 1123 goto send_mal_query; 1124 } 1125 break; 1126 default: 1127 goto send_mal_query; 1128 } 1129 if (!dev) { 1130 icmphdr->code = ICMP_EXT_CODE_NO_IF; 1131 return true; 1132 } 1133 /* Fill bits in reply message */ 1134 if (dev->flags & IFF_UP) 1135 status |= ICMP_EXT_ECHOREPLY_ACTIVE; 1136 1137 in_dev = __in_dev_get_rcu(dev); 1138 if (in_dev && rcu_access_pointer(in_dev->ifa_list)) 1139 status |= ICMP_EXT_ECHOREPLY_IPV4; 1140 1141 in6_dev = __in6_dev_get(dev); 1142 if (in6_dev && !list_empty(&in6_dev->addr_list)) 1143 status |= ICMP_EXT_ECHOREPLY_IPV6; 1144 1145 dev_put(dev); 1146 icmphdr->un.echo.sequence |= htons(status); 1147 return true; 1148 send_mal_query: 1149 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY; 1150 return true; 1151 } 1152 EXPORT_SYMBOL_GPL(icmp_build_probe); 1153 1154 /* 1155 * Handle ICMP Timestamp requests. 1156 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1157 * SHOULD be in the kernel for minimum random latency. 1158 * MUST be accurate to a few minutes. 1159 * MUST be updated at least at 15Hz. 1160 */ 1161 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb) 1162 { 1163 struct icmp_bxm icmp_param; 1164 /* 1165 * Too short. 1166 */ 1167 if (skb->len < 4) 1168 goto out_err; 1169 1170 /* 1171 * Fill in the current time as ms since midnight UT: 1172 */ 1173 icmp_param.data.times[1] = inet_current_timestamp(); 1174 icmp_param.data.times[2] = icmp_param.data.times[1]; 1175 1176 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1177 1178 icmp_param.data.icmph = *icmp_hdr(skb); 1179 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1180 icmp_param.data.icmph.code = 0; 1181 icmp_param.skb = skb; 1182 icmp_param.offset = 0; 1183 icmp_param.data_len = 0; 1184 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1185 icmp_reply(&icmp_param, skb); 1186 return SKB_NOT_DROPPED_YET; 1187 1188 out_err: 1189 __ICMP_INC_STATS(dev_net_rcu(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 1190 return SKB_DROP_REASON_PKT_TOO_SMALL; 1191 } 1192 1193 static enum skb_drop_reason icmp_discard(struct sk_buff *skb) 1194 { 1195 /* pretend it was a success */ 1196 return SKB_NOT_DROPPED_YET; 1197 } 1198 1199 /* 1200 * Deal with incoming ICMP packets. 1201 */ 1202 int icmp_rcv(struct sk_buff *skb) 1203 { 1204 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1205 struct rtable *rt = skb_rtable(skb); 1206 struct net *net = dev_net_rcu(rt->dst.dev); 1207 struct icmphdr *icmph; 1208 1209 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1210 struct sec_path *sp = skb_sec_path(skb); 1211 int nh; 1212 1213 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1214 XFRM_STATE_ICMP)) { 1215 reason = SKB_DROP_REASON_XFRM_POLICY; 1216 goto drop; 1217 } 1218 1219 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1220 goto drop; 1221 1222 nh = skb_network_offset(skb); 1223 skb_set_network_header(skb, sizeof(*icmph)); 1224 1225 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, 1226 skb)) { 1227 reason = SKB_DROP_REASON_XFRM_POLICY; 1228 goto drop; 1229 } 1230 1231 skb_set_network_header(skb, nh); 1232 } 1233 1234 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1235 1236 if (skb_checksum_simple_validate(skb)) 1237 goto csum_error; 1238 1239 if (!pskb_pull(skb, sizeof(*icmph))) 1240 goto error; 1241 1242 icmph = icmp_hdr(skb); 1243 1244 ICMPMSGIN_INC_STATS(net, icmph->type); 1245 1246 /* Check for ICMP Extended Echo (PROBE) messages */ 1247 if (icmph->type == ICMP_EXT_ECHO) { 1248 /* We can't use icmp_pointers[].handler() because it is an array of 1249 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42. 1250 */ 1251 reason = icmp_echo(skb); 1252 goto reason_check; 1253 } 1254 1255 if (icmph->type == ICMP_EXT_ECHOREPLY) { 1256 reason = ping_rcv(skb); 1257 goto reason_check; 1258 } 1259 1260 /* 1261 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1262 * 1263 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1264 * discarded. 1265 */ 1266 if (icmph->type > NR_ICMP_TYPES) { 1267 reason = SKB_DROP_REASON_UNHANDLED_PROTO; 1268 goto error; 1269 } 1270 1271 /* 1272 * Parse the ICMP message 1273 */ 1274 1275 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1276 /* 1277 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1278 * silently ignored (we let user decide with a sysctl). 1279 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1280 * discarded if to broadcast/multicast. 1281 */ 1282 if ((icmph->type == ICMP_ECHO || 1283 icmph->type == ICMP_TIMESTAMP) && 1284 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) { 1285 reason = SKB_DROP_REASON_INVALID_PROTO; 1286 goto error; 1287 } 1288 if (icmph->type != ICMP_ECHO && 1289 icmph->type != ICMP_TIMESTAMP && 1290 icmph->type != ICMP_ADDRESS && 1291 icmph->type != ICMP_ADDRESSREPLY) { 1292 reason = SKB_DROP_REASON_INVALID_PROTO; 1293 goto error; 1294 } 1295 } 1296 1297 reason = icmp_pointers[icmph->type].handler(skb); 1298 reason_check: 1299 if (!reason) { 1300 consume_skb(skb); 1301 return NET_RX_SUCCESS; 1302 } 1303 1304 drop: 1305 kfree_skb_reason(skb, reason); 1306 return NET_RX_DROP; 1307 csum_error: 1308 reason = SKB_DROP_REASON_ICMP_CSUM; 1309 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1310 error: 1311 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1312 goto drop; 1313 } 1314 1315 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1316 { 1317 struct icmp_extobj_hdr *objh, _objh; 1318 struct icmp_ext_hdr *exth, _exth; 1319 u16 olen; 1320 1321 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1322 if (!exth) 1323 return false; 1324 if (exth->version != 2) 1325 return true; 1326 1327 if (exth->checksum && 1328 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1329 return false; 1330 1331 off += sizeof(_exth); 1332 while (off < skb->len) { 1333 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1334 if (!objh) 1335 return false; 1336 1337 olen = ntohs(objh->length); 1338 if (olen < sizeof(_objh)) 1339 return false; 1340 1341 off += olen; 1342 if (off > skb->len) 1343 return false; 1344 } 1345 1346 return true; 1347 } 1348 1349 void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1350 struct sock_ee_data_rfc4884 *out, 1351 int thlen, int off) 1352 { 1353 int hlen; 1354 1355 /* original datagram headers: end of icmph to payload (skb->data) */ 1356 hlen = -skb_transport_offset(skb) - thlen; 1357 1358 /* per rfc 4884: minimal datagram length of 128 bytes */ 1359 if (off < 128 || off < hlen) 1360 return; 1361 1362 /* kernel has stripped headers: return payload offset in bytes */ 1363 off -= hlen; 1364 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1365 return; 1366 1367 out->len = off; 1368 1369 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1370 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1371 } 1372 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1373 1374 int icmp_err(struct sk_buff *skb, u32 info) 1375 { 1376 struct iphdr *iph = (struct iphdr *)skb->data; 1377 int offset = iph->ihl<<2; 1378 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1379 struct net *net = dev_net_rcu(skb->dev); 1380 int type = icmp_hdr(skb)->type; 1381 int code = icmp_hdr(skb)->code; 1382 1383 /* 1384 * Use ping_err to handle all icmp errors except those 1385 * triggered by ICMP_ECHOREPLY which sent from kernel. 1386 */ 1387 if (icmph->type != ICMP_ECHOREPLY) { 1388 ping_err(skb, offset, info); 1389 return 0; 1390 } 1391 1392 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1393 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1394 else if (type == ICMP_REDIRECT) 1395 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1396 1397 return 0; 1398 } 1399 1400 /* 1401 * This table is the definition of how we handle ICMP. 1402 */ 1403 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1404 [ICMP_ECHOREPLY] = { 1405 .handler = ping_rcv, 1406 }, 1407 [1] = { 1408 .handler = icmp_discard, 1409 .error = 1, 1410 }, 1411 [2] = { 1412 .handler = icmp_discard, 1413 .error = 1, 1414 }, 1415 [ICMP_DEST_UNREACH] = { 1416 .handler = icmp_unreach, 1417 .error = 1, 1418 }, 1419 [ICMP_SOURCE_QUENCH] = { 1420 .handler = icmp_unreach, 1421 .error = 1, 1422 }, 1423 [ICMP_REDIRECT] = { 1424 .handler = icmp_redirect, 1425 .error = 1, 1426 }, 1427 [6] = { 1428 .handler = icmp_discard, 1429 .error = 1, 1430 }, 1431 [7] = { 1432 .handler = icmp_discard, 1433 .error = 1, 1434 }, 1435 [ICMP_ECHO] = { 1436 .handler = icmp_echo, 1437 }, 1438 [9] = { 1439 .handler = icmp_discard, 1440 .error = 1, 1441 }, 1442 [10] = { 1443 .handler = icmp_discard, 1444 .error = 1, 1445 }, 1446 [ICMP_TIME_EXCEEDED] = { 1447 .handler = icmp_unreach, 1448 .error = 1, 1449 }, 1450 [ICMP_PARAMETERPROB] = { 1451 .handler = icmp_unreach, 1452 .error = 1, 1453 }, 1454 [ICMP_TIMESTAMP] = { 1455 .handler = icmp_timestamp, 1456 }, 1457 [ICMP_TIMESTAMPREPLY] = { 1458 .handler = icmp_discard, 1459 }, 1460 [ICMP_INFO_REQUEST] = { 1461 .handler = icmp_discard, 1462 }, 1463 [ICMP_INFO_REPLY] = { 1464 .handler = icmp_discard, 1465 }, 1466 [ICMP_ADDRESS] = { 1467 .handler = icmp_discard, 1468 }, 1469 [ICMP_ADDRESSREPLY] = { 1470 .handler = icmp_discard, 1471 }, 1472 }; 1473 1474 static int __net_init icmp_sk_init(struct net *net) 1475 { 1476 /* Control parameters for ECHO replies. */ 1477 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1478 net->ipv4.sysctl_icmp_echo_enable_probe = 0; 1479 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1480 1481 /* Control parameter - ignore bogus broadcast responses? */ 1482 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1483 1484 /* 1485 * Configurable global rate limit. 1486 * 1487 * ratelimit defines tokens/packet consumed for dst->rate_token 1488 * bucket ratemask defines which icmp types are ratelimited by 1489 * setting it's bit position. 1490 * 1491 * default: 1492 * dest unreachable (3), source quench (4), 1493 * time exceeded (11), parameter problem (12) 1494 */ 1495 1496 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1497 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1498 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1499 1500 return 0; 1501 } 1502 1503 static struct pernet_operations __net_initdata icmp_sk_ops = { 1504 .init = icmp_sk_init, 1505 }; 1506 1507 int __init icmp_init(void) 1508 { 1509 int err, i; 1510 1511 for_each_possible_cpu(i) { 1512 struct sock *sk; 1513 1514 err = inet_ctl_sock_create(&sk, PF_INET, 1515 SOCK_RAW, IPPROTO_ICMP, &init_net); 1516 if (err < 0) 1517 return err; 1518 1519 per_cpu(ipv4_icmp_sk, i) = sk; 1520 1521 /* Enough space for 2 64K ICMP packets, including 1522 * sk_buff/skb_shared_info struct overhead. 1523 */ 1524 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1525 1526 /* 1527 * Speedup sock_wfree() 1528 */ 1529 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1530 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1531 } 1532 return register_pernet_subsys(&icmp_sk_ops); 1533 } 1534