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/flow.h> 76 #include <net/snmp.h> 77 #include <net/ip.h> 78 #include <net/route.h> 79 #include <net/protocol.h> 80 #include <net/icmp.h> 81 #include <net/tcp.h> 82 #include <net/udp.h> 83 #include <net/raw.h> 84 #include <net/ping.h> 85 #include <linux/skbuff.h> 86 #include <net/sock.h> 87 #include <linux/errno.h> 88 #include <linux/timer.h> 89 #include <linux/init.h> 90 #include <linux/uaccess.h> 91 #include <net/checksum.h> 92 #include <net/xfrm.h> 93 #include <net/inet_common.h> 94 #include <net/ip_fib.h> 95 #include <net/l3mdev.h> 96 #include <net/addrconf.h> 97 #include <net/inet_dscp.h> 98 #define CREATE_TRACE_POINTS 99 #include <trace/events/icmp.h> 100 101 /* 102 * Build xmit assembly blocks 103 */ 104 105 struct icmp_bxm { 106 struct sk_buff *skb; 107 int offset; 108 int data_len; 109 110 struct { 111 struct icmphdr icmph; 112 __be32 times[3]; 113 } data; 114 int head_len; 115 struct ip_options_data replyopts; 116 }; 117 118 /* An array of errno for error messages from dest unreach. */ 119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 120 121 const struct icmp_err icmp_err_convert[] = { 122 { 123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 124 .fatal = 0, 125 }, 126 { 127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 128 .fatal = 0, 129 }, 130 { 131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 132 .fatal = 1, 133 }, 134 { 135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 136 .fatal = 1, 137 }, 138 { 139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 140 .fatal = 0, 141 }, 142 { 143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 144 .fatal = 0, 145 }, 146 { 147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 148 .fatal = 1, 149 }, 150 { 151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 152 .fatal = 1, 153 }, 154 { 155 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 156 .fatal = 1, 157 }, 158 { 159 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 160 .fatal = 1, 161 }, 162 { 163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 164 .fatal = 1, 165 }, 166 { 167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 168 .fatal = 0, 169 }, 170 { 171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 172 .fatal = 0, 173 }, 174 { 175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 176 .fatal = 1, 177 }, 178 { 179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 180 .fatal = 1, 181 }, 182 { 183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 184 .fatal = 1, 185 }, 186 }; 187 EXPORT_SYMBOL(icmp_err_convert); 188 189 /* 190 * ICMP control array. This specifies what to do with each ICMP. 191 */ 192 193 struct icmp_control { 194 enum skb_drop_reason (*handler)(struct sk_buff *skb); 195 short error; /* This ICMP is classed as an error message */ 196 }; 197 198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 199 200 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk); 201 202 /* Called with BH disabled */ 203 static inline struct sock *icmp_xmit_lock(struct net *net) 204 { 205 struct sock *sk; 206 207 sk = this_cpu_read(ipv4_icmp_sk); 208 209 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 210 /* This can happen if the output path signals a 211 * dst_link_failure() for an outgoing ICMP packet. 212 */ 213 return NULL; 214 } 215 sock_net_set(sk, net); 216 return sk; 217 } 218 219 static inline void icmp_xmit_unlock(struct sock *sk) 220 { 221 sock_net_set(sk, &init_net); 222 spin_unlock(&sk->sk_lock.slock); 223 } 224 225 /** 226 * icmp_global_allow - Are we allowed to send one more ICMP message ? 227 * @net: network namespace 228 * 229 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec. 230 * Returns false if we reached the limit and can not send another packet. 231 * Works in tandem with icmp_global_consume(). 232 */ 233 bool icmp_global_allow(struct net *net) 234 { 235 u32 delta, now, oldstamp; 236 int incr, new, old; 237 238 /* Note: many cpus could find this condition true. 239 * Then later icmp_global_consume() could consume more credits, 240 * this is an acceptable race. 241 */ 242 if (atomic_read(&net->ipv4.icmp_global_credit) > 0) 243 return true; 244 245 now = jiffies; 246 oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp); 247 delta = min_t(u32, now - oldstamp, HZ); 248 if (delta < HZ / 50) 249 return false; 250 251 incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ; 252 if (!incr) 253 return false; 254 255 if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) { 256 old = atomic_read(&net->ipv4.icmp_global_credit); 257 do { 258 new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst)); 259 } while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new)); 260 } 261 return true; 262 } 263 EXPORT_SYMBOL(icmp_global_allow); 264 265 void icmp_global_consume(struct net *net) 266 { 267 int credits = get_random_u32_below(3); 268 269 /* Note: this might make icmp_global.credit negative. */ 270 if (credits) 271 atomic_sub(credits, &net->ipv4.icmp_global_credit); 272 } 273 EXPORT_SYMBOL(icmp_global_consume); 274 275 static bool icmpv4_mask_allow(struct net *net, int type, int code) 276 { 277 if (type > NR_ICMP_TYPES) 278 return true; 279 280 /* Don't limit PMTU discovery. */ 281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 282 return true; 283 284 /* Limit if icmp type is enabled in ratemask. */ 285 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask))) 286 return true; 287 288 return false; 289 } 290 291 static bool icmpv4_global_allow(struct net *net, int type, int code, 292 bool *apply_ratelimit) 293 { 294 if (icmpv4_mask_allow(net, type, code)) 295 return true; 296 297 if (icmp_global_allow(net)) { 298 *apply_ratelimit = true; 299 return true; 300 } 301 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL); 302 return false; 303 } 304 305 /* 306 * Send an ICMP frame. 307 */ 308 309 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 310 struct flowi4 *fl4, int type, int code, 311 bool apply_ratelimit) 312 { 313 struct dst_entry *dst = &rt->dst; 314 struct inet_peer *peer; 315 struct net_device *dev; 316 bool rc = true; 317 318 if (!apply_ratelimit) 319 return true; 320 321 /* No rate limit on loopback */ 322 rcu_read_lock(); 323 dev = dst_dev_rcu(dst); 324 if (dev && (dev->flags & IFF_LOOPBACK)) 325 goto out; 326 327 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 328 l3mdev_master_ifindex_rcu(dev)); 329 rc = inet_peer_xrlim_allow(peer, 330 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit)); 331 out: 332 rcu_read_unlock(); 333 if (!rc) 334 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST); 335 else 336 icmp_global_consume(net); 337 return rc; 338 } 339 340 /* 341 * Maintain the counters used in the SNMP statistics for outgoing ICMP 342 */ 343 void icmp_out_count(struct net *net, unsigned char type) 344 { 345 ICMPMSGOUT_INC_STATS(net, type); 346 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 347 } 348 349 /* 350 * Checksum each fragment, and on the first include the headers and final 351 * checksum. 352 */ 353 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 354 struct sk_buff *skb) 355 { 356 struct icmp_bxm *icmp_param = from; 357 __wsum csum; 358 359 csum = skb_copy_and_csum_bits(icmp_param->skb, 360 icmp_param->offset + offset, 361 to, len); 362 363 skb->csum = csum_block_add(skb->csum, csum, odd); 364 if (icmp_pointers[icmp_param->data.icmph.type].error) 365 nf_ct_attach(skb, icmp_param->skb); 366 return 0; 367 } 368 369 static void icmp_push_reply(struct sock *sk, 370 struct icmp_bxm *icmp_param, 371 struct flowi4 *fl4, 372 struct ipcm_cookie *ipc, struct rtable **rt) 373 { 374 struct sk_buff *skb; 375 376 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 377 icmp_param->data_len+icmp_param->head_len, 378 icmp_param->head_len, 379 ipc, rt, MSG_DONTWAIT) < 0) { 380 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 381 ip_flush_pending_frames(sk); 382 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 383 struct icmphdr *icmph = icmp_hdr(skb); 384 __wsum csum; 385 struct sk_buff *skb1; 386 387 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 388 (char *)icmph, 389 icmp_param->head_len); 390 skb_queue_walk(&sk->sk_write_queue, skb1) { 391 csum = csum_add(csum, skb1->csum); 392 } 393 icmph->checksum = csum_fold(csum); 394 skb->ip_summed = CHECKSUM_NONE; 395 ip_push_pending_frames(sk, fl4); 396 } 397 } 398 399 /* 400 * Driving logic for building and sending ICMP messages. 401 */ 402 403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 404 { 405 struct rtable *rt = skb_rtable(skb); 406 struct net *net = dev_net_rcu(rt->dst.dev); 407 bool apply_ratelimit = false; 408 struct ipcm_cookie ipc; 409 struct flowi4 fl4; 410 struct sock *sk; 411 __be32 daddr, saddr; 412 u32 mark = IP4_REPLY_MARK(net, skb->mark); 413 int type = icmp_param->data.icmph.type; 414 int code = icmp_param->data.icmph.code; 415 416 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 417 return; 418 419 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */ 420 local_bh_disable(); 421 422 /* is global icmp_msgs_per_sec exhausted ? */ 423 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit)) 424 goto out_bh_enable; 425 426 sk = icmp_xmit_lock(net); 427 if (!sk) 428 goto out_bh_enable; 429 430 icmp_param->data.icmph.checksum = 0; 431 432 ipcm_init(&ipc); 433 ipc.tos = ip_hdr(skb)->tos; 434 ipc.sockc.mark = mark; 435 daddr = ipc.addr = ip_hdr(skb)->saddr; 436 saddr = fib_compute_spec_dst(skb); 437 438 if (icmp_param->replyopts.opt.opt.optlen) { 439 ipc.opt = &icmp_param->replyopts.opt; 440 if (ipc.opt->opt.srr) 441 daddr = icmp_param->replyopts.opt.opt.faddr; 442 } 443 memset(&fl4, 0, sizeof(fl4)); 444 fl4.daddr = daddr; 445 fl4.saddr = saddr; 446 fl4.flowi4_mark = mark; 447 fl4.flowi4_uid = sock_net_uid(net, NULL); 448 fl4.flowi4_dscp = ip4h_dscp(ip_hdr(skb)); 449 fl4.flowi4_proto = IPPROTO_ICMP; 450 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 451 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 452 rt = ip_route_output_key(net, &fl4); 453 if (IS_ERR(rt)) 454 goto out_unlock; 455 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit)) 456 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt); 457 ip_rt_put(rt); 458 out_unlock: 459 icmp_xmit_unlock(sk); 460 out_bh_enable: 461 local_bh_enable(); 462 } 463 464 /* 465 * The device used for looking up which routing table to use for sending an ICMP 466 * error is preferably the source whenever it is set, which should ensure the 467 * icmp error can be sent to the source host, else lookup using the routing 468 * table of the destination device, else use the main routing table (index 0). 469 */ 470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 471 { 472 struct net_device *dev = skb->dev; 473 const struct dst_entry *dst; 474 475 if (dev) 476 return dev; 477 dst = skb_dst(skb); 478 return dst ? dst_dev(dst) : NULL; 479 } 480 481 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4, 482 struct sk_buff *skb_in, 483 const struct iphdr *iph, __be32 saddr, 484 dscp_t dscp, u32 mark, int type, 485 int code, struct icmp_bxm *param) 486 { 487 struct net_device *route_lookup_dev; 488 struct dst_entry *dst, *dst2; 489 struct rtable *rt, *rt2; 490 struct flowi4 fl4_dec; 491 int err; 492 493 memset(fl4, 0, sizeof(*fl4)); 494 fl4->daddr = (param->replyopts.opt.opt.srr ? 495 param->replyopts.opt.opt.faddr : iph->saddr); 496 fl4->saddr = saddr; 497 fl4->flowi4_mark = mark; 498 fl4->flowi4_uid = sock_net_uid(net, NULL); 499 fl4->flowi4_dscp = dscp; 500 fl4->flowi4_proto = IPPROTO_ICMP; 501 fl4->fl4_icmp_type = type; 502 fl4->fl4_icmp_code = code; 503 route_lookup_dev = icmp_get_route_lookup_dev(skb_in); 504 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev); 505 506 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4)); 507 rt = ip_route_output_key_hash(net, fl4, skb_in); 508 if (IS_ERR(rt)) 509 return rt; 510 511 /* No need to clone since we're just using its address. */ 512 rt2 = rt; 513 514 dst = xfrm_lookup(net, &rt->dst, 515 flowi4_to_flowi(fl4), NULL, 0); 516 rt = dst_rtable(dst); 517 if (!IS_ERR(dst)) { 518 if (rt != rt2) 519 return rt; 520 if (inet_addr_type_dev_table(net, route_lookup_dev, 521 fl4->daddr) == RTN_LOCAL) 522 return rt; 523 } else if (PTR_ERR(dst) == -EPERM) { 524 rt = NULL; 525 } else { 526 return rt; 527 } 528 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 529 if (err) 530 goto relookup_failed; 531 532 if (inet_addr_type_dev_table(net, route_lookup_dev, 533 fl4_dec.saddr) == RTN_LOCAL) { 534 rt2 = __ip_route_output_key(net, &fl4_dec); 535 if (IS_ERR(rt2)) 536 err = PTR_ERR(rt2); 537 } else { 538 struct flowi4 fl4_2 = {}; 539 unsigned long orefdst; 540 541 fl4_2.daddr = fl4_dec.saddr; 542 rt2 = ip_route_output_key(net, &fl4_2); 543 if (IS_ERR(rt2)) { 544 err = PTR_ERR(rt2); 545 goto relookup_failed; 546 } 547 /* Ugh! */ 548 orefdst = skb_dstref_steal(skb_in); 549 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 550 dscp, rt2->dst.dev) ? -EINVAL : 0; 551 552 dst_release(&rt2->dst); 553 rt2 = skb_rtable(skb_in); 554 /* steal dst entry from skb_in, don't drop refcnt */ 555 skb_dstref_steal(skb_in); 556 skb_dstref_restore(skb_in, orefdst); 557 } 558 559 if (err) 560 goto relookup_failed; 561 562 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL, 563 XFRM_LOOKUP_ICMP); 564 rt2 = dst_rtable(dst2); 565 if (!IS_ERR(dst2)) { 566 dst_release(&rt->dst); 567 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 568 rt = rt2; 569 } else if (PTR_ERR(dst2) == -EPERM) { 570 if (rt) 571 dst_release(&rt->dst); 572 return rt2; 573 } else { 574 err = PTR_ERR(dst2); 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 ipcm_init(&ipc); 743 ipc.tos = tos; 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 = skb_dst_dev_net_rcu(skb); 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 = skb_dst_dev_net_rcu(skb); 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(skb_dst_dev_net_rcu(skb), 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 /* 1256 * Parse the ICMP message 1257 */ 1258 1259 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1260 /* 1261 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1262 * silently ignored (we let user decide with a sysctl). 1263 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1264 * discarded if to broadcast/multicast. 1265 */ 1266 if ((icmph->type == ICMP_ECHO || 1267 icmph->type == ICMP_TIMESTAMP) && 1268 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) { 1269 reason = SKB_DROP_REASON_INVALID_PROTO; 1270 goto error; 1271 } 1272 if (icmph->type != ICMP_ECHO && 1273 icmph->type != ICMP_TIMESTAMP && 1274 icmph->type != ICMP_ADDRESS && 1275 icmph->type != ICMP_ADDRESSREPLY) { 1276 reason = SKB_DROP_REASON_INVALID_PROTO; 1277 goto error; 1278 } 1279 } 1280 1281 if (icmph->type == ICMP_EXT_ECHOREPLY || 1282 icmph->type == ICMP_ECHOREPLY) { 1283 reason = ping_rcv(skb); 1284 return reason ? NET_RX_DROP : NET_RX_SUCCESS; 1285 } 1286 1287 /* 1288 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1289 * 1290 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1291 * discarded. 1292 */ 1293 if (icmph->type > NR_ICMP_TYPES) { 1294 reason = SKB_DROP_REASON_UNHANDLED_PROTO; 1295 goto error; 1296 } 1297 1298 reason = icmp_pointers[icmph->type].handler(skb); 1299 reason_check: 1300 if (!reason) { 1301 consume_skb(skb); 1302 return NET_RX_SUCCESS; 1303 } 1304 1305 drop: 1306 kfree_skb_reason(skb, reason); 1307 return NET_RX_DROP; 1308 csum_error: 1309 reason = SKB_DROP_REASON_ICMP_CSUM; 1310 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1311 error: 1312 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1313 goto drop; 1314 } 1315 1316 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1317 { 1318 struct icmp_extobj_hdr *objh, _objh; 1319 struct icmp_ext_hdr *exth, _exth; 1320 u16 olen; 1321 1322 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1323 if (!exth) 1324 return false; 1325 if (exth->version != 2) 1326 return true; 1327 1328 if (exth->checksum && 1329 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1330 return false; 1331 1332 off += sizeof(_exth); 1333 while (off < skb->len) { 1334 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1335 if (!objh) 1336 return false; 1337 1338 olen = ntohs(objh->length); 1339 if (olen < sizeof(_objh)) 1340 return false; 1341 1342 off += olen; 1343 if (off > skb->len) 1344 return false; 1345 } 1346 1347 return true; 1348 } 1349 1350 void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1351 struct sock_ee_data_rfc4884 *out, 1352 int thlen, int off) 1353 { 1354 int hlen; 1355 1356 /* original datagram headers: end of icmph to payload (skb->data) */ 1357 hlen = -skb_transport_offset(skb) - thlen; 1358 1359 /* per rfc 4884: minimal datagram length of 128 bytes */ 1360 if (off < 128 || off < hlen) 1361 return; 1362 1363 /* kernel has stripped headers: return payload offset in bytes */ 1364 off -= hlen; 1365 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1366 return; 1367 1368 out->len = off; 1369 1370 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1371 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1372 } 1373 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1374 1375 int icmp_err(struct sk_buff *skb, u32 info) 1376 { 1377 struct iphdr *iph = (struct iphdr *)skb->data; 1378 int offset = iph->ihl<<2; 1379 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1380 struct net *net = dev_net_rcu(skb->dev); 1381 int type = icmp_hdr(skb)->type; 1382 int code = icmp_hdr(skb)->code; 1383 1384 /* 1385 * Use ping_err to handle all icmp errors except those 1386 * triggered by ICMP_ECHOREPLY which sent from kernel. 1387 */ 1388 if (icmph->type != ICMP_ECHOREPLY) { 1389 ping_err(skb, offset, info); 1390 return 0; 1391 } 1392 1393 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1394 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1395 else if (type == ICMP_REDIRECT) 1396 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1397 1398 return 0; 1399 } 1400 1401 /* 1402 * This table is the definition of how we handle ICMP. 1403 */ 1404 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1405 [ICMP_ECHOREPLY] = { 1406 .handler = ping_rcv, 1407 }, 1408 [1] = { 1409 .handler = icmp_discard, 1410 .error = 1, 1411 }, 1412 [2] = { 1413 .handler = icmp_discard, 1414 .error = 1, 1415 }, 1416 [ICMP_DEST_UNREACH] = { 1417 .handler = icmp_unreach, 1418 .error = 1, 1419 }, 1420 [ICMP_SOURCE_QUENCH] = { 1421 .handler = icmp_unreach, 1422 .error = 1, 1423 }, 1424 [ICMP_REDIRECT] = { 1425 .handler = icmp_redirect, 1426 .error = 1, 1427 }, 1428 [6] = { 1429 .handler = icmp_discard, 1430 .error = 1, 1431 }, 1432 [7] = { 1433 .handler = icmp_discard, 1434 .error = 1, 1435 }, 1436 [ICMP_ECHO] = { 1437 .handler = icmp_echo, 1438 }, 1439 [9] = { 1440 .handler = icmp_discard, 1441 .error = 1, 1442 }, 1443 [10] = { 1444 .handler = icmp_discard, 1445 .error = 1, 1446 }, 1447 [ICMP_TIME_EXCEEDED] = { 1448 .handler = icmp_unreach, 1449 .error = 1, 1450 }, 1451 [ICMP_PARAMETERPROB] = { 1452 .handler = icmp_unreach, 1453 .error = 1, 1454 }, 1455 [ICMP_TIMESTAMP] = { 1456 .handler = icmp_timestamp, 1457 }, 1458 [ICMP_TIMESTAMPREPLY] = { 1459 .handler = icmp_discard, 1460 }, 1461 [ICMP_INFO_REQUEST] = { 1462 .handler = icmp_discard, 1463 }, 1464 [ICMP_INFO_REPLY] = { 1465 .handler = icmp_discard, 1466 }, 1467 [ICMP_ADDRESS] = { 1468 .handler = icmp_discard, 1469 }, 1470 [ICMP_ADDRESSREPLY] = { 1471 .handler = icmp_discard, 1472 }, 1473 }; 1474 1475 static int __net_init icmp_sk_init(struct net *net) 1476 { 1477 /* Control parameters for ECHO replies. */ 1478 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1479 net->ipv4.sysctl_icmp_echo_enable_probe = 0; 1480 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1481 1482 /* Control parameter - ignore bogus broadcast responses? */ 1483 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1484 1485 /* 1486 * Configurable global rate limit. 1487 * 1488 * ratelimit defines tokens/packet consumed for dst->rate_token 1489 * bucket ratemask defines which icmp types are ratelimited by 1490 * setting it's bit position. 1491 * 1492 * default: 1493 * dest unreachable (3), source quench (4), 1494 * time exceeded (11), parameter problem (12) 1495 */ 1496 1497 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1498 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1499 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1500 net->ipv4.sysctl_icmp_msgs_per_sec = 1000; 1501 net->ipv4.sysctl_icmp_msgs_burst = 50; 1502 1503 return 0; 1504 } 1505 1506 static struct pernet_operations __net_initdata icmp_sk_ops = { 1507 .init = icmp_sk_init, 1508 }; 1509 1510 int __init icmp_init(void) 1511 { 1512 int err, i; 1513 1514 for_each_possible_cpu(i) { 1515 struct sock *sk; 1516 1517 err = inet_ctl_sock_create(&sk, PF_INET, 1518 SOCK_RAW, IPPROTO_ICMP, &init_net); 1519 if (err < 0) 1520 return err; 1521 1522 per_cpu(ipv4_icmp_sk, i) = sk; 1523 1524 /* Enough space for 2 64K ICMP packets, including 1525 * sk_buff/skb_shared_info struct overhead. 1526 */ 1527 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1528 1529 /* 1530 * Speedup sock_wfree() 1531 */ 1532 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1533 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1534 } 1535 return register_pernet_subsys(&icmp_sk_ops); 1536 } 1537