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 bool rc = true; 315 316 if (!apply_ratelimit) 317 return true; 318 319 /* No rate limit on loopback */ 320 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 321 goto out; 322 323 rcu_read_lock(); 324 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 325 l3mdev_master_ifindex_rcu(dst->dev)); 326 rc = inet_peer_xrlim_allow(peer, 327 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit)); 328 rcu_read_unlock(); 329 out: 330 if (!rc) 331 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST); 332 else 333 icmp_global_consume(net); 334 return rc; 335 } 336 337 /* 338 * Maintain the counters used in the SNMP statistics for outgoing ICMP 339 */ 340 void icmp_out_count(struct net *net, unsigned char type) 341 { 342 ICMPMSGOUT_INC_STATS(net, type); 343 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 344 } 345 346 /* 347 * Checksum each fragment, and on the first include the headers and final 348 * checksum. 349 */ 350 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 351 struct sk_buff *skb) 352 { 353 struct icmp_bxm *icmp_param = from; 354 __wsum csum; 355 356 csum = skb_copy_and_csum_bits(icmp_param->skb, 357 icmp_param->offset + offset, 358 to, len); 359 360 skb->csum = csum_block_add(skb->csum, csum, odd); 361 if (icmp_pointers[icmp_param->data.icmph.type].error) 362 nf_ct_attach(skb, icmp_param->skb); 363 return 0; 364 } 365 366 static void icmp_push_reply(struct sock *sk, 367 struct icmp_bxm *icmp_param, 368 struct flowi4 *fl4, 369 struct ipcm_cookie *ipc, struct rtable **rt) 370 { 371 struct sk_buff *skb; 372 373 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 374 icmp_param->data_len+icmp_param->head_len, 375 icmp_param->head_len, 376 ipc, rt, MSG_DONTWAIT) < 0) { 377 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 378 ip_flush_pending_frames(sk); 379 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 380 struct icmphdr *icmph = icmp_hdr(skb); 381 __wsum csum; 382 struct sk_buff *skb1; 383 384 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 385 (char *)icmph, 386 icmp_param->head_len); 387 skb_queue_walk(&sk->sk_write_queue, skb1) { 388 csum = csum_add(csum, skb1->csum); 389 } 390 icmph->checksum = csum_fold(csum); 391 skb->ip_summed = CHECKSUM_NONE; 392 ip_push_pending_frames(sk, fl4); 393 } 394 } 395 396 /* 397 * Driving logic for building and sending ICMP messages. 398 */ 399 400 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 401 { 402 struct rtable *rt = skb_rtable(skb); 403 struct net *net = dev_net_rcu(rt->dst.dev); 404 bool apply_ratelimit = false; 405 struct ipcm_cookie ipc; 406 struct flowi4 fl4; 407 struct sock *sk; 408 __be32 daddr, saddr; 409 u32 mark = IP4_REPLY_MARK(net, skb->mark); 410 int type = icmp_param->data.icmph.type; 411 int code = icmp_param->data.icmph.code; 412 413 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 414 return; 415 416 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 417 local_bh_disable(); 418 419 /* is global icmp_msgs_per_sec exhausted ? */ 420 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit)) 421 goto out_bh_enable; 422 423 sk = icmp_xmit_lock(net); 424 if (!sk) 425 goto out_bh_enable; 426 427 icmp_param->data.icmph.checksum = 0; 428 429 ipcm_init(&ipc); 430 ipc.tos = ip_hdr(skb)->tos; 431 ipc.sockc.mark = mark; 432 daddr = ipc.addr = ip_hdr(skb)->saddr; 433 saddr = fib_compute_spec_dst(skb); 434 435 if (icmp_param->replyopts.opt.opt.optlen) { 436 ipc.opt = &icmp_param->replyopts.opt; 437 if (ipc.opt->opt.srr) 438 daddr = icmp_param->replyopts.opt.opt.faddr; 439 } 440 memset(&fl4, 0, sizeof(fl4)); 441 fl4.daddr = daddr; 442 fl4.saddr = saddr; 443 fl4.flowi4_mark = mark; 444 fl4.flowi4_uid = sock_net_uid(net, NULL); 445 fl4.flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb))); 446 fl4.flowi4_proto = IPPROTO_ICMP; 447 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 448 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 449 rt = ip_route_output_key(net, &fl4); 450 if (IS_ERR(rt)) 451 goto out_unlock; 452 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 453 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt); 454 ip_rt_put(rt); 455 out_unlock: 456 icmp_xmit_unlock(sk); 457 out_bh_enable: 458 local_bh_enable(); 459 } 460 461 /* 462 * The device used for looking up which routing table to use for sending an ICMP 463 * error is preferably the source whenever it is set, which should ensure the 464 * icmp error can be sent to the source host, else lookup using the routing 465 * table of the destination device, else use the main routing table (index 0). 466 */ 467 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 468 { 469 struct net_device *route_lookup_dev = NULL; 470 471 if (skb->dev) 472 route_lookup_dev = skb->dev; 473 else if (skb_dst(skb)) 474 route_lookup_dev = skb_dst(skb)->dev; 475 return route_lookup_dev; 476 } 477 478 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4, 479 struct sk_buff *skb_in, 480 const struct iphdr *iph, __be32 saddr, 481 dscp_t dscp, u32 mark, int type, 482 int code, struct icmp_bxm *param) 483 { 484 struct net_device *route_lookup_dev; 485 struct dst_entry *dst, *dst2; 486 struct rtable *rt, *rt2; 487 struct flowi4 fl4_dec; 488 int err; 489 490 memset(fl4, 0, sizeof(*fl4)); 491 fl4->daddr = (param->replyopts.opt.opt.srr ? 492 param->replyopts.opt.opt.faddr : iph->saddr); 493 fl4->saddr = saddr; 494 fl4->flowi4_mark = mark; 495 fl4->flowi4_uid = sock_net_uid(net, NULL); 496 fl4->flowi4_tos = inet_dscp_to_dsfield(dscp); 497 fl4->flowi4_proto = IPPROTO_ICMP; 498 fl4->fl4_icmp_type = type; 499 fl4->fl4_icmp_code = code; 500 route_lookup_dev = icmp_get_route_lookup_dev(skb_in); 501 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev); 502 503 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4)); 504 rt = ip_route_output_key_hash(net, fl4, skb_in); 505 if (IS_ERR(rt)) 506 return rt; 507 508 /* No need to clone since we're just using its address. */ 509 rt2 = rt; 510 511 dst = xfrm_lookup(net, &rt->dst, 512 flowi4_to_flowi(fl4), NULL, 0); 513 rt = dst_rtable(dst); 514 if (!IS_ERR(dst)) { 515 if (rt != rt2) 516 return rt; 517 if (inet_addr_type_dev_table(net, route_lookup_dev, 518 fl4->daddr) == RTN_LOCAL) 519 return rt; 520 } else if (PTR_ERR(dst) == -EPERM) { 521 rt = NULL; 522 } else { 523 return rt; 524 } 525 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 526 if (err) 527 goto relookup_failed; 528 529 if (inet_addr_type_dev_table(net, route_lookup_dev, 530 fl4_dec.saddr) == RTN_LOCAL) { 531 rt2 = __ip_route_output_key(net, &fl4_dec); 532 if (IS_ERR(rt2)) 533 err = PTR_ERR(rt2); 534 } else { 535 struct flowi4 fl4_2 = {}; 536 unsigned long orefdst; 537 538 fl4_2.daddr = fl4_dec.saddr; 539 rt2 = ip_route_output_key(net, &fl4_2); 540 if (IS_ERR(rt2)) { 541 err = PTR_ERR(rt2); 542 goto relookup_failed; 543 } 544 /* Ugh! */ 545 orefdst = skb_in->_skb_refdst; /* save old refdst */ 546 skb_dst_set(skb_in, NULL); 547 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 548 dscp, rt2->dst.dev) ? -EINVAL : 0; 549 550 dst_release(&rt2->dst); 551 rt2 = skb_rtable(skb_in); 552 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 553 } 554 555 if (err) 556 goto relookup_failed; 557 558 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL, 559 XFRM_LOOKUP_ICMP); 560 rt2 = dst_rtable(dst2); 561 if (!IS_ERR(dst2)) { 562 dst_release(&rt->dst); 563 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 564 rt = rt2; 565 } else if (PTR_ERR(dst2) == -EPERM) { 566 if (rt) 567 dst_release(&rt->dst); 568 return rt2; 569 } else { 570 err = PTR_ERR(dst2); 571 goto relookup_failed; 572 } 573 return rt; 574 575 relookup_failed: 576 if (rt) 577 return rt; 578 return ERR_PTR(err); 579 } 580 581 /* 582 * Send an ICMP message in response to a situation 583 * 584 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 585 * MAY send more (we do). 586 * MUST NOT change this header information. 587 * MUST NOT reply to a multicast/broadcast IP address. 588 * MUST NOT reply to a multicast/broadcast MAC address. 589 * MUST reply to only the first fragment. 590 */ 591 592 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 593 const struct ip_options *opt) 594 { 595 struct iphdr *iph; 596 int room; 597 struct icmp_bxm icmp_param; 598 struct rtable *rt = skb_rtable(skb_in); 599 bool apply_ratelimit = false; 600 struct ipcm_cookie ipc; 601 struct flowi4 fl4; 602 __be32 saddr; 603 u8 tos; 604 u32 mark; 605 struct net *net; 606 struct sock *sk; 607 608 if (!rt) 609 return; 610 611 rcu_read_lock(); 612 613 if (rt->dst.dev) 614 net = dev_net_rcu(rt->dst.dev); 615 else if (skb_in->dev) 616 net = dev_net_rcu(skb_in->dev); 617 else 618 goto out; 619 620 /* 621 * Find the original header. It is expected to be valid, of course. 622 * Check this, icmp_send is called from the most obscure devices 623 * sometimes. 624 */ 625 iph = ip_hdr(skb_in); 626 627 if ((u8 *)iph < skb_in->head || 628 (skb_network_header(skb_in) + sizeof(*iph)) > 629 skb_tail_pointer(skb_in)) 630 goto out; 631 632 /* 633 * No replies to physical multicast/broadcast 634 */ 635 if (skb_in->pkt_type != PACKET_HOST) 636 goto out; 637 638 /* 639 * Now check at the protocol level 640 */ 641 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 642 goto out; 643 644 /* 645 * Only reply to fragment 0. We byte re-order the constant 646 * mask for efficiency. 647 */ 648 if (iph->frag_off & htons(IP_OFFSET)) 649 goto out; 650 651 /* 652 * If we send an ICMP error to an ICMP error a mess would result.. 653 */ 654 if (icmp_pointers[type].error) { 655 /* 656 * We are an error, check if we are replying to an 657 * ICMP error 658 */ 659 if (iph->protocol == IPPROTO_ICMP) { 660 u8 _inner_type, *itp; 661 662 itp = skb_header_pointer(skb_in, 663 skb_network_header(skb_in) + 664 (iph->ihl << 2) + 665 offsetof(struct icmphdr, 666 type) - 667 skb_in->data, 668 sizeof(_inner_type), 669 &_inner_type); 670 if (!itp) 671 goto out; 672 673 /* 674 * Assume any unknown ICMP type is an error. This 675 * isn't specified by the RFC, but think about it.. 676 */ 677 if (*itp > NR_ICMP_TYPES || 678 icmp_pointers[*itp].error) 679 goto out; 680 } 681 } 682 683 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 684 local_bh_disable(); 685 686 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 687 * incoming dev is loopback. If outgoing dev change to not be 688 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 689 */ 690 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 691 !icmpv4_global_allow(net, type, code, &apply_ratelimit)) 692 goto out_bh_enable; 693 694 sk = icmp_xmit_lock(net); 695 if (!sk) 696 goto out_bh_enable; 697 698 /* 699 * Construct source address and options. 700 */ 701 702 saddr = iph->daddr; 703 if (!(rt->rt_flags & RTCF_LOCAL)) { 704 struct net_device *dev = NULL; 705 706 rcu_read_lock(); 707 if (rt_is_input_route(rt) && 708 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)) 709 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 710 711 if (dev) 712 saddr = inet_select_addr(dev, iph->saddr, 713 RT_SCOPE_LINK); 714 else 715 saddr = 0; 716 rcu_read_unlock(); 717 } 718 719 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 720 IPTOS_PREC_INTERNETCONTROL) : 721 iph->tos; 722 mark = IP4_REPLY_MARK(net, skb_in->mark); 723 724 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 725 goto out_unlock; 726 727 728 /* 729 * Prepare data for ICMP header. 730 */ 731 732 icmp_param.data.icmph.type = type; 733 icmp_param.data.icmph.code = code; 734 icmp_param.data.icmph.un.gateway = info; 735 icmp_param.data.icmph.checksum = 0; 736 icmp_param.skb = skb_in; 737 icmp_param.offset = skb_network_offset(skb_in); 738 ipcm_init(&ipc); 739 ipc.tos = tos; 740 ipc.addr = iph->saddr; 741 ipc.opt = &icmp_param.replyopts.opt; 742 ipc.sockc.mark = mark; 743 744 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, 745 inet_dsfield_to_dscp(tos), mark, type, code, 746 &icmp_param); 747 if (IS_ERR(rt)) 748 goto out_unlock; 749 750 /* peer icmp_ratelimit */ 751 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 752 goto ende; 753 754 /* RFC says return as much as we can without exceeding 576 bytes. */ 755 756 room = dst_mtu(&rt->dst); 757 if (room > 576) 758 room = 576; 759 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 760 room -= sizeof(struct icmphdr); 761 /* Guard against tiny mtu. We need to include at least one 762 * IP network header for this message to make any sense. 763 */ 764 if (room <= (int)sizeof(struct iphdr)) 765 goto ende; 766 767 icmp_param.data_len = skb_in->len - icmp_param.offset; 768 if (icmp_param.data_len > room) 769 icmp_param.data_len = room; 770 icmp_param.head_len = sizeof(struct icmphdr); 771 772 /* if we don't have a source address at this point, fall back to the 773 * dummy address instead of sending out a packet with a source address 774 * of 0.0.0.0 775 */ 776 if (!fl4.saddr) 777 fl4.saddr = htonl(INADDR_DUMMY); 778 779 trace_icmp_send(skb_in, type, code); 780 781 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt); 782 ende: 783 ip_rt_put(rt); 784 out_unlock: 785 icmp_xmit_unlock(sk); 786 out_bh_enable: 787 local_bh_enable(); 788 out: 789 rcu_read_unlock(); 790 } 791 EXPORT_SYMBOL(__icmp_send); 792 793 #if IS_ENABLED(CONFIG_NF_NAT) 794 #include <net/netfilter/nf_conntrack.h> 795 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 796 { 797 struct sk_buff *cloned_skb = NULL; 798 struct ip_options opts = { 0 }; 799 enum ip_conntrack_info ctinfo; 800 struct nf_conn *ct; 801 __be32 orig_ip; 802 803 ct = nf_ct_get(skb_in, &ctinfo); 804 if (!ct || !(ct->status & IPS_SRC_NAT)) { 805 __icmp_send(skb_in, type, code, info, &opts); 806 return; 807 } 808 809 if (skb_shared(skb_in)) 810 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 811 812 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 813 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 814 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 815 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 816 goto out; 817 818 orig_ip = ip_hdr(skb_in)->saddr; 819 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 820 __icmp_send(skb_in, type, code, info, &opts); 821 ip_hdr(skb_in)->saddr = orig_ip; 822 out: 823 consume_skb(cloned_skb); 824 } 825 EXPORT_SYMBOL(icmp_ndo_send); 826 #endif 827 828 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 829 { 830 const struct iphdr *iph = (const struct iphdr *)skb->data; 831 const struct net_protocol *ipprot; 832 int protocol = iph->protocol; 833 834 /* Checkin full IP header plus 8 bytes of protocol to 835 * avoid additional coding at protocol handlers. 836 */ 837 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 838 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 839 return; 840 } 841 842 raw_icmp_error(skb, protocol, info); 843 844 ipprot = rcu_dereference(inet_protos[protocol]); 845 if (ipprot && ipprot->err_handler) 846 ipprot->err_handler(skb, info); 847 } 848 849 static bool icmp_tag_validation(int proto) 850 { 851 bool ok; 852 853 rcu_read_lock(); 854 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 855 rcu_read_unlock(); 856 return ok; 857 } 858 859 /* 860 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 861 * ICMP_PARAMETERPROB. 862 */ 863 864 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb) 865 { 866 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET; 867 const struct iphdr *iph; 868 struct icmphdr *icmph; 869 struct net *net; 870 u32 info = 0; 871 872 net = dev_net_rcu(skb_dst(skb)->dev); 873 874 /* 875 * Incomplete header ? 876 * Only checks for the IP header, there should be an 877 * additional check for longer headers in upper levels. 878 */ 879 880 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 881 goto out_err; 882 883 icmph = icmp_hdr(skb); 884 iph = (const struct iphdr *)skb->data; 885 886 if (iph->ihl < 5) { /* Mangled header, drop. */ 887 reason = SKB_DROP_REASON_IP_INHDR; 888 goto out_err; 889 } 890 891 switch (icmph->type) { 892 case ICMP_DEST_UNREACH: 893 switch (icmph->code & 15) { 894 case ICMP_NET_UNREACH: 895 case ICMP_HOST_UNREACH: 896 case ICMP_PROT_UNREACH: 897 case ICMP_PORT_UNREACH: 898 break; 899 case ICMP_FRAG_NEEDED: 900 /* for documentation of the ip_no_pmtu_disc 901 * values please see 902 * Documentation/networking/ip-sysctl.rst 903 */ 904 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) { 905 default: 906 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 907 &iph->daddr); 908 break; 909 case 2: 910 goto out; 911 case 3: 912 if (!icmp_tag_validation(iph->protocol)) 913 goto out; 914 fallthrough; 915 case 0: 916 info = ntohs(icmph->un.frag.mtu); 917 } 918 break; 919 case ICMP_SR_FAILED: 920 net_dbg_ratelimited("%pI4: Source Route Failed\n", 921 &iph->daddr); 922 break; 923 default: 924 break; 925 } 926 if (icmph->code > NR_ICMP_UNREACH) 927 goto out; 928 break; 929 case ICMP_PARAMETERPROB: 930 info = ntohl(icmph->un.gateway) >> 24; 931 break; 932 case ICMP_TIME_EXCEEDED: 933 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 934 if (icmph->code == ICMP_EXC_FRAGTIME) 935 goto out; 936 break; 937 } 938 939 /* 940 * Throw it at our lower layers 941 * 942 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 943 * header. 944 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 945 * transport layer. 946 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 947 * transport layer. 948 */ 949 950 /* 951 * Check the other end isn't violating RFC 1122. Some routers send 952 * bogus responses to broadcast frames. If you see this message 953 * first check your netmask matches at both ends, if it does then 954 * get the other vendor to fix their kit. 955 */ 956 957 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) && 958 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 959 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 960 &ip_hdr(skb)->saddr, 961 icmph->type, icmph->code, 962 &iph->daddr, skb->dev->name); 963 goto out; 964 } 965 966 icmp_socket_deliver(skb, info); 967 968 out: 969 return reason; 970 out_err: 971 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 972 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED; 973 } 974 975 976 /* 977 * Handle ICMP_REDIRECT. 978 */ 979 980 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb) 981 { 982 if (skb->len < sizeof(struct iphdr)) { 983 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS); 984 return SKB_DROP_REASON_PKT_TOO_SMALL; 985 } 986 987 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 988 /* there aught to be a stat */ 989 return SKB_DROP_REASON_NOMEM; 990 } 991 992 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 993 return SKB_NOT_DROPPED_YET; 994 } 995 996 /* 997 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests. 998 * 999 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 1000 * requests. 1001 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 1002 * included in the reply. 1003 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 1004 * echo requests, MUST have default=NOT. 1005 * RFC 8335: 8 MUST have a config option to enable/disable ICMP 1006 * Extended Echo Functionality, MUST be disabled by default 1007 * See also WRT handling of options once they are done and working. 1008 */ 1009 1010 static enum skb_drop_reason icmp_echo(struct sk_buff *skb) 1011 { 1012 struct icmp_bxm icmp_param; 1013 struct net *net; 1014 1015 net = dev_net_rcu(skb_dst(skb)->dev); 1016 /* should there be an ICMP stat for ignored echos? */ 1017 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all)) 1018 return SKB_NOT_DROPPED_YET; 1019 1020 icmp_param.data.icmph = *icmp_hdr(skb); 1021 icmp_param.skb = skb; 1022 icmp_param.offset = 0; 1023 icmp_param.data_len = skb->len; 1024 icmp_param.head_len = sizeof(struct icmphdr); 1025 1026 if (icmp_param.data.icmph.type == ICMP_ECHO) 1027 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 1028 else if (!icmp_build_probe(skb, &icmp_param.data.icmph)) 1029 return SKB_NOT_DROPPED_YET; 1030 1031 icmp_reply(&icmp_param, skb); 1032 return SKB_NOT_DROPPED_YET; 1033 } 1034 1035 /* Helper for icmp_echo and icmpv6_echo_reply. 1036 * Searches for net_device that matches PROBE interface identifier 1037 * and builds PROBE reply message in icmphdr. 1038 * 1039 * Returns false if PROBE responses are disabled via sysctl 1040 */ 1041 1042 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr) 1043 { 1044 struct net *net = dev_net_rcu(skb->dev); 1045 struct icmp_ext_hdr *ext_hdr, _ext_hdr; 1046 struct icmp_ext_echo_iio *iio, _iio; 1047 struct inet6_dev *in6_dev; 1048 struct in_device *in_dev; 1049 struct net_device *dev; 1050 char buff[IFNAMSIZ]; 1051 u16 ident_len; 1052 u8 status; 1053 1054 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe)) 1055 return false; 1056 1057 /* We currently only support probing interfaces on the proxy node 1058 * Check to ensure L-bit is set 1059 */ 1060 if (!(ntohs(icmphdr->un.echo.sequence) & 1)) 1061 return false; 1062 /* Clear status bits in reply message */ 1063 icmphdr->un.echo.sequence &= htons(0xFF00); 1064 if (icmphdr->type == ICMP_EXT_ECHO) 1065 icmphdr->type = ICMP_EXT_ECHOREPLY; 1066 else 1067 icmphdr->type = ICMPV6_EXT_ECHO_REPLY; 1068 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr); 1069 /* Size of iio is class_type dependent. 1070 * Only check header here and assign length based on ctype in the switch statement 1071 */ 1072 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio); 1073 if (!ext_hdr || !iio) 1074 goto send_mal_query; 1075 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) || 1076 ntohs(iio->extobj_hdr.length) > sizeof(_iio)) 1077 goto send_mal_query; 1078 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr); 1079 iio = skb_header_pointer(skb, sizeof(_ext_hdr), 1080 sizeof(iio->extobj_hdr) + ident_len, &_iio); 1081 if (!iio) 1082 goto send_mal_query; 1083 1084 status = 0; 1085 dev = NULL; 1086 switch (iio->extobj_hdr.class_type) { 1087 case ICMP_EXT_ECHO_CTYPE_NAME: 1088 if (ident_len >= IFNAMSIZ) 1089 goto send_mal_query; 1090 memset(buff, 0, sizeof(buff)); 1091 memcpy(buff, &iio->ident.name, ident_len); 1092 dev = dev_get_by_name(net, buff); 1093 break; 1094 case ICMP_EXT_ECHO_CTYPE_INDEX: 1095 if (ident_len != sizeof(iio->ident.ifindex)) 1096 goto send_mal_query; 1097 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex)); 1098 break; 1099 case ICMP_EXT_ECHO_CTYPE_ADDR: 1100 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) || 1101 ident_len != sizeof(iio->ident.addr.ctype3_hdr) + 1102 iio->ident.addr.ctype3_hdr.addrlen) 1103 goto send_mal_query; 1104 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) { 1105 case ICMP_AFI_IP: 1106 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr)) 1107 goto send_mal_query; 1108 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr); 1109 break; 1110 #if IS_ENABLED(CONFIG_IPV6) 1111 case ICMP_AFI_IP6: 1112 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr)) 1113 goto send_mal_query; 1114 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev); 1115 dev_hold(dev); 1116 break; 1117 #endif 1118 default: 1119 goto send_mal_query; 1120 } 1121 break; 1122 default: 1123 goto send_mal_query; 1124 } 1125 if (!dev) { 1126 icmphdr->code = ICMP_EXT_CODE_NO_IF; 1127 return true; 1128 } 1129 /* Fill bits in reply message */ 1130 if (dev->flags & IFF_UP) 1131 status |= ICMP_EXT_ECHOREPLY_ACTIVE; 1132 1133 in_dev = __in_dev_get_rcu(dev); 1134 if (in_dev && rcu_access_pointer(in_dev->ifa_list)) 1135 status |= ICMP_EXT_ECHOREPLY_IPV4; 1136 1137 in6_dev = __in6_dev_get(dev); 1138 if (in6_dev && !list_empty(&in6_dev->addr_list)) 1139 status |= ICMP_EXT_ECHOREPLY_IPV6; 1140 1141 dev_put(dev); 1142 icmphdr->un.echo.sequence |= htons(status); 1143 return true; 1144 send_mal_query: 1145 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY; 1146 return true; 1147 } 1148 EXPORT_SYMBOL_GPL(icmp_build_probe); 1149 1150 /* 1151 * Handle ICMP Timestamp requests. 1152 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1153 * SHOULD be in the kernel for minimum random latency. 1154 * MUST be accurate to a few minutes. 1155 * MUST be updated at least at 15Hz. 1156 */ 1157 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb) 1158 { 1159 struct icmp_bxm icmp_param; 1160 /* 1161 * Too short. 1162 */ 1163 if (skb->len < 4) 1164 goto out_err; 1165 1166 /* 1167 * Fill in the current time as ms since midnight UT: 1168 */ 1169 icmp_param.data.times[1] = inet_current_timestamp(); 1170 icmp_param.data.times[2] = icmp_param.data.times[1]; 1171 1172 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1173 1174 icmp_param.data.icmph = *icmp_hdr(skb); 1175 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1176 icmp_param.data.icmph.code = 0; 1177 icmp_param.skb = skb; 1178 icmp_param.offset = 0; 1179 icmp_param.data_len = 0; 1180 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1181 icmp_reply(&icmp_param, skb); 1182 return SKB_NOT_DROPPED_YET; 1183 1184 out_err: 1185 __ICMP_INC_STATS(dev_net_rcu(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 1186 return SKB_DROP_REASON_PKT_TOO_SMALL; 1187 } 1188 1189 static enum skb_drop_reason icmp_discard(struct sk_buff *skb) 1190 { 1191 /* pretend it was a success */ 1192 return SKB_NOT_DROPPED_YET; 1193 } 1194 1195 /* 1196 * Deal with incoming ICMP packets. 1197 */ 1198 int icmp_rcv(struct sk_buff *skb) 1199 { 1200 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 1201 struct rtable *rt = skb_rtable(skb); 1202 struct net *net = dev_net_rcu(rt->dst.dev); 1203 struct icmphdr *icmph; 1204 1205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1206 struct sec_path *sp = skb_sec_path(skb); 1207 int nh; 1208 1209 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1210 XFRM_STATE_ICMP)) { 1211 reason = SKB_DROP_REASON_XFRM_POLICY; 1212 goto drop; 1213 } 1214 1215 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1216 goto drop; 1217 1218 nh = skb_network_offset(skb); 1219 skb_set_network_header(skb, sizeof(*icmph)); 1220 1221 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, 1222 skb)) { 1223 reason = SKB_DROP_REASON_XFRM_POLICY; 1224 goto drop; 1225 } 1226 1227 skb_set_network_header(skb, nh); 1228 } 1229 1230 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1231 1232 if (skb_checksum_simple_validate(skb)) 1233 goto csum_error; 1234 1235 if (!pskb_pull(skb, sizeof(*icmph))) 1236 goto error; 1237 1238 icmph = icmp_hdr(skb); 1239 1240 ICMPMSGIN_INC_STATS(net, icmph->type); 1241 1242 /* Check for ICMP Extended Echo (PROBE) messages */ 1243 if (icmph->type == ICMP_EXT_ECHO) { 1244 /* We can't use icmp_pointers[].handler() because it is an array of 1245 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42. 1246 */ 1247 reason = icmp_echo(skb); 1248 goto reason_check; 1249 } 1250 1251 /* 1252 * Parse the ICMP message 1253 */ 1254 1255 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1256 /* 1257 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1258 * silently ignored (we let user decide with a sysctl). 1259 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1260 * discarded if to broadcast/multicast. 1261 */ 1262 if ((icmph->type == ICMP_ECHO || 1263 icmph->type == ICMP_TIMESTAMP) && 1264 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) { 1265 reason = SKB_DROP_REASON_INVALID_PROTO; 1266 goto error; 1267 } 1268 if (icmph->type != ICMP_ECHO && 1269 icmph->type != ICMP_TIMESTAMP && 1270 icmph->type != ICMP_ADDRESS && 1271 icmph->type != ICMP_ADDRESSREPLY) { 1272 reason = SKB_DROP_REASON_INVALID_PROTO; 1273 goto error; 1274 } 1275 } 1276 1277 if (icmph->type == ICMP_EXT_ECHOREPLY || 1278 icmph->type == ICMP_ECHOREPLY) { 1279 reason = ping_rcv(skb); 1280 return reason ? NET_RX_DROP : NET_RX_SUCCESS; 1281 } 1282 1283 /* 1284 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1285 * 1286 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1287 * discarded. 1288 */ 1289 if (icmph->type > NR_ICMP_TYPES) { 1290 reason = SKB_DROP_REASON_UNHANDLED_PROTO; 1291 goto error; 1292 } 1293 1294 reason = icmp_pointers[icmph->type].handler(skb); 1295 reason_check: 1296 if (!reason) { 1297 consume_skb(skb); 1298 return NET_RX_SUCCESS; 1299 } 1300 1301 drop: 1302 kfree_skb_reason(skb, reason); 1303 return NET_RX_DROP; 1304 csum_error: 1305 reason = SKB_DROP_REASON_ICMP_CSUM; 1306 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1307 error: 1308 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1309 goto drop; 1310 } 1311 1312 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1313 { 1314 struct icmp_extobj_hdr *objh, _objh; 1315 struct icmp_ext_hdr *exth, _exth; 1316 u16 olen; 1317 1318 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1319 if (!exth) 1320 return false; 1321 if (exth->version != 2) 1322 return true; 1323 1324 if (exth->checksum && 1325 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1326 return false; 1327 1328 off += sizeof(_exth); 1329 while (off < skb->len) { 1330 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1331 if (!objh) 1332 return false; 1333 1334 olen = ntohs(objh->length); 1335 if (olen < sizeof(_objh)) 1336 return false; 1337 1338 off += olen; 1339 if (off > skb->len) 1340 return false; 1341 } 1342 1343 return true; 1344 } 1345 1346 void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1347 struct sock_ee_data_rfc4884 *out, 1348 int thlen, int off) 1349 { 1350 int hlen; 1351 1352 /* original datagram headers: end of icmph to payload (skb->data) */ 1353 hlen = -skb_transport_offset(skb) - thlen; 1354 1355 /* per rfc 4884: minimal datagram length of 128 bytes */ 1356 if (off < 128 || off < hlen) 1357 return; 1358 1359 /* kernel has stripped headers: return payload offset in bytes */ 1360 off -= hlen; 1361 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1362 return; 1363 1364 out->len = off; 1365 1366 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1367 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1368 } 1369 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1370 1371 int icmp_err(struct sk_buff *skb, u32 info) 1372 { 1373 struct iphdr *iph = (struct iphdr *)skb->data; 1374 int offset = iph->ihl<<2; 1375 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1376 struct net *net = dev_net_rcu(skb->dev); 1377 int type = icmp_hdr(skb)->type; 1378 int code = icmp_hdr(skb)->code; 1379 1380 /* 1381 * Use ping_err to handle all icmp errors except those 1382 * triggered by ICMP_ECHOREPLY which sent from kernel. 1383 */ 1384 if (icmph->type != ICMP_ECHOREPLY) { 1385 ping_err(skb, offset, info); 1386 return 0; 1387 } 1388 1389 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1390 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1391 else if (type == ICMP_REDIRECT) 1392 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1393 1394 return 0; 1395 } 1396 1397 /* 1398 * This table is the definition of how we handle ICMP. 1399 */ 1400 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1401 [ICMP_ECHOREPLY] = { 1402 .handler = ping_rcv, 1403 }, 1404 [1] = { 1405 .handler = icmp_discard, 1406 .error = 1, 1407 }, 1408 [2] = { 1409 .handler = icmp_discard, 1410 .error = 1, 1411 }, 1412 [ICMP_DEST_UNREACH] = { 1413 .handler = icmp_unreach, 1414 .error = 1, 1415 }, 1416 [ICMP_SOURCE_QUENCH] = { 1417 .handler = icmp_unreach, 1418 .error = 1, 1419 }, 1420 [ICMP_REDIRECT] = { 1421 .handler = icmp_redirect, 1422 .error = 1, 1423 }, 1424 [6] = { 1425 .handler = icmp_discard, 1426 .error = 1, 1427 }, 1428 [7] = { 1429 .handler = icmp_discard, 1430 .error = 1, 1431 }, 1432 [ICMP_ECHO] = { 1433 .handler = icmp_echo, 1434 }, 1435 [9] = { 1436 .handler = icmp_discard, 1437 .error = 1, 1438 }, 1439 [10] = { 1440 .handler = icmp_discard, 1441 .error = 1, 1442 }, 1443 [ICMP_TIME_EXCEEDED] = { 1444 .handler = icmp_unreach, 1445 .error = 1, 1446 }, 1447 [ICMP_PARAMETERPROB] = { 1448 .handler = icmp_unreach, 1449 .error = 1, 1450 }, 1451 [ICMP_TIMESTAMP] = { 1452 .handler = icmp_timestamp, 1453 }, 1454 [ICMP_TIMESTAMPREPLY] = { 1455 .handler = icmp_discard, 1456 }, 1457 [ICMP_INFO_REQUEST] = { 1458 .handler = icmp_discard, 1459 }, 1460 [ICMP_INFO_REPLY] = { 1461 .handler = icmp_discard, 1462 }, 1463 [ICMP_ADDRESS] = { 1464 .handler = icmp_discard, 1465 }, 1466 [ICMP_ADDRESSREPLY] = { 1467 .handler = icmp_discard, 1468 }, 1469 }; 1470 1471 static int __net_init icmp_sk_init(struct net *net) 1472 { 1473 /* Control parameters for ECHO replies. */ 1474 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1475 net->ipv4.sysctl_icmp_echo_enable_probe = 0; 1476 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1477 1478 /* Control parameter - ignore bogus broadcast responses? */ 1479 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1480 1481 /* 1482 * Configurable global rate limit. 1483 * 1484 * ratelimit defines tokens/packet consumed for dst->rate_token 1485 * bucket ratemask defines which icmp types are ratelimited by 1486 * setting it's bit position. 1487 * 1488 * default: 1489 * dest unreachable (3), source quench (4), 1490 * time exceeded (11), parameter problem (12) 1491 */ 1492 1493 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1494 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1495 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1496 net->ipv4.sysctl_icmp_msgs_per_sec = 1000; 1497 net->ipv4.sysctl_icmp_msgs_burst = 50; 1498 1499 return 0; 1500 } 1501 1502 static struct pernet_operations __net_initdata icmp_sk_ops = { 1503 .init = icmp_sk_init, 1504 }; 1505 1506 int __init icmp_init(void) 1507 { 1508 int err, i; 1509 1510 for_each_possible_cpu(i) { 1511 struct sock *sk; 1512 1513 err = inet_ctl_sock_create(&sk, PF_INET, 1514 SOCK_RAW, IPPROTO_ICMP, &init_net); 1515 if (err < 0) 1516 return err; 1517 1518 per_cpu(ipv4_icmp_sk, i) = sk; 1519 1520 /* Enough space for 2 64K ICMP packets, including 1521 * sk_buff/skb_shared_info struct overhead. 1522 */ 1523 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1524 1525 /* 1526 * Speedup sock_wfree() 1527 */ 1528 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1529 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1530 } 1531 return register_pernet_subsys(&icmp_sk_ops); 1532 } 1533