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 /** 225 * icmp_global_allow - Are we allowed to send one more ICMP message ? 226 * @net: network namespace 227 * 228 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec. 229 * Returns false if we reached the limit and can not send another packet. 230 * Works in tandem with icmp_global_consume(). 231 */ 232 bool icmp_global_allow(struct net *net) 233 { 234 u32 delta, now, oldstamp; 235 int incr, new, old; 236 237 /* Note: many cpus could find this condition true. 238 * Then later icmp_global_consume() could consume more credits, 239 * this is an acceptable race. 240 */ 241 if (atomic_read(&net->ipv4.icmp_global_credit) > 0) 242 return true; 243 244 now = jiffies; 245 oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp); 246 delta = min_t(u32, now - oldstamp, HZ); 247 if (delta < HZ / 50) 248 return false; 249 250 incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ; 251 if (!incr) 252 return false; 253 254 if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) { 255 old = atomic_read(&net->ipv4.icmp_global_credit); 256 do { 257 new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst)); 258 } while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new)); 259 } 260 return true; 261 } 262 EXPORT_SYMBOL(icmp_global_allow); 263 264 void icmp_global_consume(struct net *net) 265 { 266 int credits = get_random_u32_below(3); 267 268 /* Note: this might make icmp_global.credit negative. */ 269 if (credits) 270 atomic_sub(credits, &net->ipv4.icmp_global_credit); 271 } 272 EXPORT_SYMBOL(icmp_global_consume); 273 274 static bool icmpv4_mask_allow(struct net *net, int type, int code) 275 { 276 if (type > NR_ICMP_TYPES) 277 return true; 278 279 /* Don't limit PMTU discovery. */ 280 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 281 return true; 282 283 /* Limit if icmp type is enabled in ratemask. */ 284 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask))) 285 return true; 286 287 return false; 288 } 289 290 static bool icmpv4_global_allow(struct net *net, int type, int code, 291 bool *apply_ratelimit) 292 { 293 if (icmpv4_mask_allow(net, type, code)) 294 return true; 295 296 if (icmp_global_allow(net)) { 297 *apply_ratelimit = true; 298 return true; 299 } 300 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL); 301 return false; 302 } 303 304 /* 305 * Send an ICMP frame. 306 */ 307 308 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 309 struct flowi4 *fl4, int type, int code, 310 bool apply_ratelimit) 311 { 312 struct dst_entry *dst = &rt->dst; 313 struct inet_peer *peer; 314 struct net_device *dev; 315 bool rc = true; 316 317 if (!apply_ratelimit) 318 return true; 319 320 /* No rate limit on loopback */ 321 dev = dst_dev(dst); 322 if (dev && (dev->flags & IFF_LOOPBACK)) 323 goto out; 324 325 rcu_read_lock(); 326 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 327 l3mdev_master_ifindex_rcu(dev)); 328 rc = inet_peer_xrlim_allow(peer, 329 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit)); 330 rcu_read_unlock(); 331 out: 332 if (!rc) 333 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST); 334 else 335 icmp_global_consume(net); 336 return rc; 337 } 338 339 /* 340 * Maintain the counters used in the SNMP statistics for outgoing ICMP 341 */ 342 void icmp_out_count(struct net *net, unsigned char type) 343 { 344 ICMPMSGOUT_INC_STATS(net, type); 345 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 346 } 347 348 /* 349 * Checksum each fragment, and on the first include the headers and final 350 * checksum. 351 */ 352 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 353 struct sk_buff *skb) 354 { 355 struct icmp_bxm *icmp_param = from; 356 __wsum csum; 357 358 csum = skb_copy_and_csum_bits(icmp_param->skb, 359 icmp_param->offset + offset, 360 to, len); 361 362 skb->csum = csum_block_add(skb->csum, csum, odd); 363 if (icmp_pointers[icmp_param->data.icmph.type].error) 364 nf_ct_attach(skb, icmp_param->skb); 365 return 0; 366 } 367 368 static void icmp_push_reply(struct sock *sk, 369 struct icmp_bxm *icmp_param, 370 struct flowi4 *fl4, 371 struct ipcm_cookie *ipc, struct rtable **rt) 372 { 373 struct sk_buff *skb; 374 375 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; 384 struct sk_buff *skb1; 385 386 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 387 (char *)icmph, 388 icmp_param->head_len); 389 skb_queue_walk(&sk->sk_write_queue, skb1) { 390 csum = csum_add(csum, skb1->csum); 391 } 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 rtable *rt = skb_rtable(skb); 405 struct net *net = dev_net_rcu(rt->dst.dev); 406 bool apply_ratelimit = false; 407 struct ipcm_cookie ipc; 408 struct flowi4 fl4; 409 struct sock *sk; 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 icmpv4_global_allow and icmp_xmit_lock */ 419 local_bh_disable(); 420 421 /* is global icmp_msgs_per_sec exhausted ? */ 422 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit)) 423 goto out_bh_enable; 424 425 sk = icmp_xmit_lock(net); 426 if (!sk) 427 goto out_bh_enable; 428 429 icmp_param->data.icmph.checksum = 0; 430 431 ipcm_init(&ipc); 432 ipc.tos = ip_hdr(skb)->tos; 433 ipc.sockc.mark = mark; 434 daddr = ipc.addr = ip_hdr(skb)->saddr; 435 saddr = fib_compute_spec_dst(skb); 436 437 if (icmp_param->replyopts.opt.opt.optlen) { 438 ipc.opt = &icmp_param->replyopts.opt; 439 if (ipc.opt->opt.srr) 440 daddr = icmp_param->replyopts.opt.opt.faddr; 441 } 442 memset(&fl4, 0, sizeof(fl4)); 443 fl4.daddr = daddr; 444 fl4.saddr = saddr; 445 fl4.flowi4_mark = mark; 446 fl4.flowi4_uid = sock_net_uid(net, NULL); 447 fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb))); 448 fl4.flowi4_proto = IPPROTO_ICMP; 449 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 450 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 451 rt = ip_route_output_key(net, &fl4); 452 if (IS_ERR(rt)) 453 goto out_unlock; 454 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 455 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt); 456 ip_rt_put(rt); 457 out_unlock: 458 icmp_xmit_unlock(sk); 459 out_bh_enable: 460 local_bh_enable(); 461 } 462 463 /* 464 * The device used for looking up which routing table to use for sending an ICMP 465 * error is preferably the source whenever it is set, which should ensure the 466 * icmp error can be sent to the source host, else lookup using the routing 467 * table of the destination device, else use the main routing table (index 0). 468 */ 469 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 470 { 471 struct net_device *dev = skb->dev; 472 const struct dst_entry *dst; 473 474 if (dev) 475 return dev; 476 dst = skb_dst(skb); 477 return dst ? dst_dev(dst) : NULL; 478 } 479 480 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4, 481 struct sk_buff *skb_in, 482 const struct iphdr *iph, __be32 saddr, 483 dscp_t dscp, u32 mark, int type, 484 int code, struct icmp_bxm *param) 485 { 486 struct net_device *route_lookup_dev; 487 struct dst_entry *dst, *dst2; 488 struct rtable *rt, *rt2; 489 struct flowi4 fl4_dec; 490 int err; 491 492 memset(fl4, 0, sizeof(*fl4)); 493 fl4->daddr = (param->replyopts.opt.opt.srr ? 494 param->replyopts.opt.opt.faddr : iph->saddr); 495 fl4->saddr = saddr; 496 fl4->flowi4_mark = mark; 497 fl4->flowi4_uid = sock_net_uid(net, NULL); 498 fl4->flowi4_tos = inet_dscp_to_dsfield(dscp); 499 fl4->flowi4_proto = IPPROTO_ICMP; 500 fl4->fl4_icmp_type = type; 501 fl4->fl4_icmp_code = code; 502 route_lookup_dev = icmp_get_route_lookup_dev(skb_in); 503 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev); 504 505 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4)); 506 rt = ip_route_output_key_hash(net, fl4, skb_in); 507 if (IS_ERR(rt)) 508 return rt; 509 510 /* No need to clone since we're just using its address. */ 511 rt2 = rt; 512 513 dst = xfrm_lookup(net, &rt->dst, 514 flowi4_to_flowi(fl4), NULL, 0); 515 rt = dst_rtable(dst); 516 if (!IS_ERR(dst)) { 517 if (rt != rt2) 518 return rt; 519 if (inet_addr_type_dev_table(net, route_lookup_dev, 520 fl4->daddr) == RTN_LOCAL) 521 return rt; 522 } else if (PTR_ERR(dst) == -EPERM) { 523 rt = NULL; 524 } else { 525 return rt; 526 } 527 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 528 if (err) 529 goto relookup_failed; 530 531 if (inet_addr_type_dev_table(net, route_lookup_dev, 532 fl4_dec.saddr) == RTN_LOCAL) { 533 rt2 = __ip_route_output_key(net, &fl4_dec); 534 if (IS_ERR(rt2)) 535 err = PTR_ERR(rt2); 536 } else { 537 struct flowi4 fl4_2 = {}; 538 unsigned long orefdst; 539 540 fl4_2.daddr = fl4_dec.saddr; 541 rt2 = ip_route_output_key(net, &fl4_2); 542 if (IS_ERR(rt2)) { 543 err = PTR_ERR(rt2); 544 goto relookup_failed; 545 } 546 /* Ugh! */ 547 orefdst = skb_dstref_steal(skb_in); 548 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 549 dscp, rt2->dst.dev) ? -EINVAL : 0; 550 551 dst_release(&rt2->dst); 552 rt2 = skb_rtable(skb_in); 553 /* steal dst entry from skb_in, don't drop refcnt */ 554 skb_dstref_steal(skb_in); 555 skb_dstref_restore(skb_in, orefdst); 556 } 557 558 if (err) 559 goto relookup_failed; 560 561 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL, 562 XFRM_LOOKUP_ICMP); 563 rt2 = dst_rtable(dst2); 564 if (!IS_ERR(dst2)) { 565 dst_release(&rt->dst); 566 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 567 rt = rt2; 568 } else if (PTR_ERR(dst2) == -EPERM) { 569 if (rt) 570 dst_release(&rt->dst); 571 return rt2; 572 } else { 573 err = PTR_ERR(dst2); 574 goto relookup_failed; 575 } 576 return rt; 577 578 relookup_failed: 579 if (rt) 580 return rt; 581 return ERR_PTR(err); 582 } 583 584 /* 585 * Send an ICMP message in response to a situation 586 * 587 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 588 * MAY send more (we do). 589 * MUST NOT change this header information. 590 * MUST NOT reply to a multicast/broadcast IP address. 591 * MUST NOT reply to a multicast/broadcast MAC address. 592 * MUST reply to only the first fragment. 593 */ 594 595 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 596 const struct ip_options *opt) 597 { 598 struct iphdr *iph; 599 int room; 600 struct icmp_bxm icmp_param; 601 struct rtable *rt = skb_rtable(skb_in); 602 bool apply_ratelimit = false; 603 struct ipcm_cookie ipc; 604 struct flowi4 fl4; 605 __be32 saddr; 606 u8 tos; 607 u32 mark; 608 struct net *net; 609 struct sock *sk; 610 611 if (!rt) 612 return; 613 614 rcu_read_lock(); 615 616 if (rt->dst.dev) 617 net = dev_net_rcu(rt->dst.dev); 618 else if (skb_in->dev) 619 net = dev_net_rcu(skb_in->dev); 620 else 621 goto out; 622 623 /* 624 * Find the original header. It is expected to be valid, of course. 625 * Check this, icmp_send is called from the most obscure devices 626 * sometimes. 627 */ 628 iph = ip_hdr(skb_in); 629 630 if ((u8 *)iph < skb_in->head || 631 (skb_network_header(skb_in) + sizeof(*iph)) > 632 skb_tail_pointer(skb_in)) 633 goto out; 634 635 /* 636 * No replies to physical multicast/broadcast 637 */ 638 if (skb_in->pkt_type != PACKET_HOST) 639 goto out; 640 641 /* 642 * Now check at the protocol level 643 */ 644 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 645 goto out; 646 647 /* 648 * Only reply to fragment 0. We byte re-order the constant 649 * mask for efficiency. 650 */ 651 if (iph->frag_off & htons(IP_OFFSET)) 652 goto out; 653 654 /* 655 * If we send an ICMP error to an ICMP error a mess would result.. 656 */ 657 if (icmp_pointers[type].error) { 658 /* 659 * We are an error, check if we are replying to an 660 * ICMP error 661 */ 662 if (iph->protocol == IPPROTO_ICMP) { 663 u8 _inner_type, *itp; 664 665 itp = skb_header_pointer(skb_in, 666 skb_network_header(skb_in) + 667 (iph->ihl << 2) + 668 offsetof(struct icmphdr, 669 type) - 670 skb_in->data, 671 sizeof(_inner_type), 672 &_inner_type); 673 if (!itp) 674 goto out; 675 676 /* 677 * Assume any unknown ICMP type is an error. This 678 * isn't specified by the RFC, but think about it.. 679 */ 680 if (*itp > NR_ICMP_TYPES || 681 icmp_pointers[*itp].error) 682 goto out; 683 } 684 } 685 686 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 687 local_bh_disable(); 688 689 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 690 * incoming dev is loopback. If outgoing dev change to not be 691 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 692 */ 693 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 694 !icmpv4_global_allow(net, type, code, &apply_ratelimit)) 695 goto out_bh_enable; 696 697 sk = icmp_xmit_lock(net); 698 if (!sk) 699 goto out_bh_enable; 700 701 /* 702 * Construct source address and options. 703 */ 704 705 saddr = iph->daddr; 706 if (!(rt->rt_flags & RTCF_LOCAL)) { 707 struct net_device *dev = NULL; 708 709 rcu_read_lock(); 710 if (rt_is_input_route(rt) && 711 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)) 712 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 713 714 if (dev) 715 saddr = inet_select_addr(dev, iph->saddr, 716 RT_SCOPE_LINK); 717 else 718 saddr = 0; 719 rcu_read_unlock(); 720 } 721 722 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 723 IPTOS_PREC_INTERNETCONTROL) : 724 iph->tos; 725 mark = IP4_REPLY_MARK(net, skb_in->mark); 726 727 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 728 goto out_unlock; 729 730 731 /* 732 * Prepare data for ICMP header. 733 */ 734 735 icmp_param.data.icmph.type = type; 736 icmp_param.data.icmph.code = code; 737 icmp_param.data.icmph.un.gateway = info; 738 icmp_param.data.icmph.checksum = 0; 739 icmp_param.skb = skb_in; 740 icmp_param.offset = skb_network_offset(skb_in); 741 ipcm_init(&ipc); 742 ipc.tos = tos; 743 ipc.addr = iph->saddr; 744 ipc.opt = &icmp_param.replyopts.opt; 745 ipc.sockc.mark = mark; 746 747 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, 748 inet_dsfield_to_dscp(tos), mark, type, code, 749 &icmp_param); 750 if (IS_ERR(rt)) 751 goto out_unlock; 752 753 /* peer icmp_ratelimit */ 754 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 755 goto ende; 756 757 /* RFC says return as much as we can without exceeding 576 bytes. */ 758 759 room = dst_mtu(&rt->dst); 760 if (room > 576) 761 room = 576; 762 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 763 room -= sizeof(struct icmphdr); 764 /* Guard against tiny mtu. We need to include at least one 765 * IP network header for this message to make any sense. 766 */ 767 if (room <= (int)sizeof(struct iphdr)) 768 goto ende; 769 770 icmp_param.data_len = skb_in->len - icmp_param.offset; 771 if (icmp_param.data_len > room) 772 icmp_param.data_len = room; 773 icmp_param.head_len = sizeof(struct icmphdr); 774 775 /* if we don't have a source address at this point, fall back to the 776 * dummy address instead of sending out a packet with a source address 777 * of 0.0.0.0 778 */ 779 if (!fl4.saddr) 780 fl4.saddr = htonl(INADDR_DUMMY); 781 782 trace_icmp_send(skb_in, type, code); 783 784 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt); 785 ende: 786 ip_rt_put(rt); 787 out_unlock: 788 icmp_xmit_unlock(sk); 789 out_bh_enable: 790 local_bh_enable(); 791 out: 792 rcu_read_unlock(); 793 } 794 EXPORT_SYMBOL(__icmp_send); 795 796 #if IS_ENABLED(CONFIG_NF_NAT) 797 #include <net/netfilter/nf_conntrack.h> 798 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 799 { 800 struct sk_buff *cloned_skb = NULL; 801 struct ip_options opts = { 0 }; 802 enum ip_conntrack_info ctinfo; 803 struct nf_conn *ct; 804 __be32 orig_ip; 805 806 ct = nf_ct_get(skb_in, &ctinfo); 807 if (!ct || !(ct->status & IPS_SRC_NAT)) { 808 __icmp_send(skb_in, type, code, info, &opts); 809 return; 810 } 811 812 if (skb_shared(skb_in)) 813 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 814 815 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 816 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 817 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 818 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 819 goto out; 820 821 orig_ip = ip_hdr(skb_in)->saddr; 822 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 823 __icmp_send(skb_in, type, code, info, &opts); 824 ip_hdr(skb_in)->saddr = orig_ip; 825 out: 826 consume_skb(cloned_skb); 827 } 828 EXPORT_SYMBOL(icmp_ndo_send); 829 #endif 830 831 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 832 { 833 const struct iphdr *iph = (const struct iphdr *)skb->data; 834 const struct net_protocol *ipprot; 835 int protocol = iph->protocol; 836 837 /* Checkin full IP header plus 8 bytes of protocol to 838 * avoid additional coding at protocol handlers. 839 */ 840 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 841 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 842 return; 843 } 844 845 raw_icmp_error(skb, protocol, info); 846 847 ipprot = rcu_dereference(inet_protos[protocol]); 848 if (ipprot && ipprot->err_handler) 849 ipprot->err_handler(skb, info); 850 } 851 852 static bool icmp_tag_validation(int proto) 853 { 854 bool ok; 855 856 rcu_read_lock(); 857 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 858 rcu_read_unlock(); 859 return ok; 860 } 861 862 /* 863 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 864 * ICMP_PARAMETERPROB. 865 */ 866 867 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb) 868 { 869 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET; 870 const struct iphdr *iph; 871 struct icmphdr *icmph; 872 struct net *net; 873 u32 info = 0; 874 875 net = skb_dst_dev_net_rcu(skb); 876 877 /* 878 * Incomplete header ? 879 * Only checks for the IP header, there should be an 880 * additional check for longer headers in upper levels. 881 */ 882 883 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 884 goto out_err; 885 886 icmph = icmp_hdr(skb); 887 iph = (const struct iphdr *)skb->data; 888 889 if (iph->ihl < 5) { /* Mangled header, drop. */ 890 reason = SKB_DROP_REASON_IP_INHDR; 891 goto out_err; 892 } 893 894 switch (icmph->type) { 895 case ICMP_DEST_UNREACH: 896 switch (icmph->code & 15) { 897 case ICMP_NET_UNREACH: 898 case ICMP_HOST_UNREACH: 899 case ICMP_PROT_UNREACH: 900 case ICMP_PORT_UNREACH: 901 break; 902 case ICMP_FRAG_NEEDED: 903 /* for documentation of the ip_no_pmtu_disc 904 * values please see 905 * Documentation/networking/ip-sysctl.rst 906 */ 907 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) { 908 default: 909 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 910 &iph->daddr); 911 break; 912 case 2: 913 goto out; 914 case 3: 915 if (!icmp_tag_validation(iph->protocol)) 916 goto out; 917 fallthrough; 918 case 0: 919 info = ntohs(icmph->un.frag.mtu); 920 } 921 break; 922 case ICMP_SR_FAILED: 923 net_dbg_ratelimited("%pI4: Source Route Failed\n", 924 &iph->daddr); 925 break; 926 default: 927 break; 928 } 929 if (icmph->code > NR_ICMP_UNREACH) 930 goto out; 931 break; 932 case ICMP_PARAMETERPROB: 933 info = ntohl(icmph->un.gateway) >> 24; 934 break; 935 case ICMP_TIME_EXCEEDED: 936 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 937 if (icmph->code == ICMP_EXC_FRAGTIME) 938 goto out; 939 break; 940 } 941 942 /* 943 * Throw it at our lower layers 944 * 945 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 946 * header. 947 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 948 * transport layer. 949 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 950 * transport layer. 951 */ 952 953 /* 954 * Check the other end isn't violating RFC 1122. Some routers send 955 * bogus responses to broadcast frames. If you see this message 956 * first check your netmask matches at both ends, if it does then 957 * get the other vendor to fix their kit. 958 */ 959 960 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) && 961 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 962 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 963 &ip_hdr(skb)->saddr, 964 icmph->type, icmph->code, 965 &iph->daddr, skb->dev->name); 966 goto out; 967 } 968 969 icmp_socket_deliver(skb, info); 970 971 out: 972 return reason; 973 out_err: 974 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 975 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED; 976 } 977 978 979 /* 980 * Handle ICMP_REDIRECT. 981 */ 982 983 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb) 984 { 985 if (skb->len < sizeof(struct iphdr)) { 986 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 987 return SKB_DROP_REASON_PKT_TOO_SMALL; 988 } 989 990 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 991 /* there aught to be a stat */ 992 return SKB_DROP_REASON_NOMEM; 993 } 994 995 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 996 return SKB_NOT_DROPPED_YET; 997 } 998 999 /* 1000 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests. 1001 * 1002 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 1003 * requests. 1004 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 1005 * included in the reply. 1006 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 1007 * echo requests, MUST have default=NOT. 1008 * RFC 8335: 8 MUST have a config option to enable/disable ICMP 1009 * Extended Echo Functionality, MUST be disabled by default 1010 * See also WRT handling of options once they are done and working. 1011 */ 1012 1013 static enum skb_drop_reason icmp_echo(struct sk_buff *skb) 1014 { 1015 struct icmp_bxm icmp_param; 1016 struct net *net; 1017 1018 net = skb_dst_dev_net_rcu(skb); 1019 /* should there be an ICMP stat for ignored echos? */ 1020 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all)) 1021 return SKB_NOT_DROPPED_YET; 1022 1023 icmp_param.data.icmph = *icmp_hdr(skb); 1024 icmp_param.skb = skb; 1025 icmp_param.offset = 0; 1026 icmp_param.data_len = skb->len; 1027 icmp_param.head_len = sizeof(struct icmphdr); 1028 1029 if (icmp_param.data.icmph.type == ICMP_ECHO) 1030 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 1031 else if (!icmp_build_probe(skb, &icmp_param.data.icmph)) 1032 return SKB_NOT_DROPPED_YET; 1033 1034 icmp_reply(&icmp_param, skb); 1035 return SKB_NOT_DROPPED_YET; 1036 } 1037 1038 /* Helper for icmp_echo and icmpv6_echo_reply. 1039 * Searches for net_device that matches PROBE interface identifier 1040 * and builds PROBE reply message in icmphdr. 1041 * 1042 * Returns false if PROBE responses are disabled via sysctl 1043 */ 1044 1045 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr) 1046 { 1047 struct net *net = dev_net_rcu(skb->dev); 1048 struct icmp_ext_hdr *ext_hdr, _ext_hdr; 1049 struct icmp_ext_echo_iio *iio, _iio; 1050 struct inet6_dev *in6_dev; 1051 struct in_device *in_dev; 1052 struct net_device *dev; 1053 char buff[IFNAMSIZ]; 1054 u16 ident_len; 1055 u8 status; 1056 1057 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe)) 1058 return false; 1059 1060 /* We currently only support probing interfaces on the proxy node 1061 * Check to ensure L-bit is set 1062 */ 1063 if (!(ntohs(icmphdr->un.echo.sequence) & 1)) 1064 return false; 1065 /* Clear status bits in reply message */ 1066 icmphdr->un.echo.sequence &= htons(0xFF00); 1067 if (icmphdr->type == ICMP_EXT_ECHO) 1068 icmphdr->type = ICMP_EXT_ECHOREPLY; 1069 else 1070 icmphdr->type = ICMPV6_EXT_ECHO_REPLY; 1071 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr); 1072 /* Size of iio is class_type dependent. 1073 * Only check header here and assign length based on ctype in the switch statement 1074 */ 1075 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio); 1076 if (!ext_hdr || !iio) 1077 goto send_mal_query; 1078 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) || 1079 ntohs(iio->extobj_hdr.length) > sizeof(_iio)) 1080 goto send_mal_query; 1081 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr); 1082 iio = skb_header_pointer(skb, sizeof(_ext_hdr), 1083 sizeof(iio->extobj_hdr) + ident_len, &_iio); 1084 if (!iio) 1085 goto send_mal_query; 1086 1087 status = 0; 1088 dev = NULL; 1089 switch (iio->extobj_hdr.class_type) { 1090 case ICMP_EXT_ECHO_CTYPE_NAME: 1091 if (ident_len >= IFNAMSIZ) 1092 goto send_mal_query; 1093 memset(buff, 0, sizeof(buff)); 1094 memcpy(buff, &iio->ident.name, ident_len); 1095 dev = dev_get_by_name(net, buff); 1096 break; 1097 case ICMP_EXT_ECHO_CTYPE_INDEX: 1098 if (ident_len != sizeof(iio->ident.ifindex)) 1099 goto send_mal_query; 1100 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex)); 1101 break; 1102 case ICMP_EXT_ECHO_CTYPE_ADDR: 1103 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) || 1104 ident_len != sizeof(iio->ident.addr.ctype3_hdr) + 1105 iio->ident.addr.ctype3_hdr.addrlen) 1106 goto send_mal_query; 1107 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) { 1108 case ICMP_AFI_IP: 1109 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr)) 1110 goto send_mal_query; 1111 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr); 1112 break; 1113 #if IS_ENABLED(CONFIG_IPV6) 1114 case ICMP_AFI_IP6: 1115 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr)) 1116 goto send_mal_query; 1117 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev); 1118 dev_hold(dev); 1119 break; 1120 #endif 1121 default: 1122 goto send_mal_query; 1123 } 1124 break; 1125 default: 1126 goto send_mal_query; 1127 } 1128 if (!dev) { 1129 icmphdr->code = ICMP_EXT_CODE_NO_IF; 1130 return true; 1131 } 1132 /* Fill bits in reply message */ 1133 if (dev->flags & IFF_UP) 1134 status |= ICMP_EXT_ECHOREPLY_ACTIVE; 1135 1136 in_dev = __in_dev_get_rcu(dev); 1137 if (in_dev && rcu_access_pointer(in_dev->ifa_list)) 1138 status |= ICMP_EXT_ECHOREPLY_IPV4; 1139 1140 in6_dev = __in6_dev_get(dev); 1141 if (in6_dev && !list_empty(&in6_dev->addr_list)) 1142 status |= ICMP_EXT_ECHOREPLY_IPV6; 1143 1144 dev_put(dev); 1145 icmphdr->un.echo.sequence |= htons(status); 1146 return true; 1147 send_mal_query: 1148 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY; 1149 return true; 1150 } 1151 EXPORT_SYMBOL_GPL(icmp_build_probe); 1152 1153 /* 1154 * Handle ICMP Timestamp requests. 1155 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1156 * SHOULD be in the kernel for minimum random latency. 1157 * MUST be accurate to a few minutes. 1158 * MUST be updated at least at 15Hz. 1159 */ 1160 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb) 1161 { 1162 struct icmp_bxm icmp_param; 1163 /* 1164 * Too short. 1165 */ 1166 if (skb->len < 4) 1167 goto out_err; 1168 1169 /* 1170 * Fill in the current time as ms since midnight UT: 1171 */ 1172 icmp_param.data.times[1] = inet_current_timestamp(); 1173 icmp_param.data.times[2] = icmp_param.data.times[1]; 1174 1175 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1176 1177 icmp_param.data.icmph = *icmp_hdr(skb); 1178 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1179 icmp_param.data.icmph.code = 0; 1180 icmp_param.skb = skb; 1181 icmp_param.offset = 0; 1182 icmp_param.data_len = 0; 1183 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1184 icmp_reply(&icmp_param, skb); 1185 return SKB_NOT_DROPPED_YET; 1186 1187 out_err: 1188 __ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS); 1189 return SKB_DROP_REASON_PKT_TOO_SMALL; 1190 } 1191 1192 static enum skb_drop_reason icmp_discard(struct sk_buff *skb) 1193 { 1194 /* pretend it was a success */ 1195 return SKB_NOT_DROPPED_YET; 1196 } 1197 1198 /* 1199 * Deal with incoming ICMP packets. 1200 */ 1201 int icmp_rcv(struct sk_buff *skb) 1202 { 1203 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1204 struct rtable *rt = skb_rtable(skb); 1205 struct net *net = dev_net_rcu(rt->dst.dev); 1206 struct icmphdr *icmph; 1207 1208 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1209 struct sec_path *sp = skb_sec_path(skb); 1210 int nh; 1211 1212 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1213 XFRM_STATE_ICMP)) { 1214 reason = SKB_DROP_REASON_XFRM_POLICY; 1215 goto drop; 1216 } 1217 1218 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1219 goto drop; 1220 1221 nh = skb_network_offset(skb); 1222 skb_set_network_header(skb, sizeof(*icmph)); 1223 1224 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, 1225 skb)) { 1226 reason = SKB_DROP_REASON_XFRM_POLICY; 1227 goto drop; 1228 } 1229 1230 skb_set_network_header(skb, nh); 1231 } 1232 1233 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1234 1235 if (skb_checksum_simple_validate(skb)) 1236 goto csum_error; 1237 1238 if (!pskb_pull(skb, sizeof(*icmph))) 1239 goto error; 1240 1241 icmph = icmp_hdr(skb); 1242 1243 ICMPMSGIN_INC_STATS(net, icmph->type); 1244 1245 /* Check for ICMP Extended Echo (PROBE) messages */ 1246 if (icmph->type == ICMP_EXT_ECHO) { 1247 /* We can't use icmp_pointers[].handler() because it is an array of 1248 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42. 1249 */ 1250 reason = icmp_echo(skb); 1251 goto reason_check; 1252 } 1253 1254 /* 1255 * Parse the ICMP message 1256 */ 1257 1258 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1259 /* 1260 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1261 * silently ignored (we let user decide with a sysctl). 1262 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1263 * discarded if to broadcast/multicast. 1264 */ 1265 if ((icmph->type == ICMP_ECHO || 1266 icmph->type == ICMP_TIMESTAMP) && 1267 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) { 1268 reason = SKB_DROP_REASON_INVALID_PROTO; 1269 goto error; 1270 } 1271 if (icmph->type != ICMP_ECHO && 1272 icmph->type != ICMP_TIMESTAMP && 1273 icmph->type != ICMP_ADDRESS && 1274 icmph->type != ICMP_ADDRESSREPLY) { 1275 reason = SKB_DROP_REASON_INVALID_PROTO; 1276 goto error; 1277 } 1278 } 1279 1280 if (icmph->type == ICMP_EXT_ECHOREPLY || 1281 icmph->type == ICMP_ECHOREPLY) { 1282 reason = ping_rcv(skb); 1283 return reason ? NET_RX_DROP : NET_RX_SUCCESS; 1284 } 1285 1286 /* 1287 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1288 * 1289 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1290 * discarded. 1291 */ 1292 if (icmph->type > NR_ICMP_TYPES) { 1293 reason = SKB_DROP_REASON_UNHANDLED_PROTO; 1294 goto error; 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 net->ipv4.sysctl_icmp_msgs_per_sec = 1000; 1500 net->ipv4.sysctl_icmp_msgs_burst = 50; 1501 1502 return 0; 1503 } 1504 1505 static struct pernet_operations __net_initdata icmp_sk_ops = { 1506 .init = icmp_sk_init, 1507 }; 1508 1509 int __init icmp_init(void) 1510 { 1511 int err, i; 1512 1513 for_each_possible_cpu(i) { 1514 struct sock *sk; 1515 1516 err = inet_ctl_sock_create(&sk, PF_INET, 1517 SOCK_RAW, IPPROTO_ICMP, &init_net); 1518 if (err < 0) 1519 return err; 1520 1521 per_cpu(ipv4_icmp_sk, i) = sk; 1522 1523 /* Enough space for 2 64K ICMP packets, including 1524 * sk_buff/skb_shared_info struct overhead. 1525 */ 1526 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1527 1528 /* 1529 * Speedup sock_wfree() 1530 */ 1531 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1532 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1533 } 1534 return register_pernet_subsys(&icmp_sk_ops); 1535 } 1536