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