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 96 /* 97 * Build xmit assembly blocks 98 */ 99 100 struct icmp_bxm { 101 struct sk_buff *skb; 102 int offset; 103 int data_len; 104 105 struct { 106 struct icmphdr icmph; 107 __be32 times[3]; 108 } data; 109 int head_len; 110 struct ip_options_data replyopts; 111 }; 112 113 /* An array of errno for error messages from dest unreach. */ 114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */ 115 116 const struct icmp_err icmp_err_convert[] = { 117 { 118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */ 119 .fatal = 0, 120 }, 121 { 122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */ 123 .fatal = 0, 124 }, 125 { 126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */, 127 .fatal = 1, 128 }, 129 { 130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */ 131 .fatal = 1, 132 }, 133 { 134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */ 135 .fatal = 0, 136 }, 137 { 138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */ 139 .fatal = 0, 140 }, 141 { 142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */ 143 .fatal = 1, 144 }, 145 { 146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */ 147 .fatal = 1, 148 }, 149 { 150 .errno = ENONET, /* ICMP_HOST_ISOLATED */ 151 .fatal = 1, 152 }, 153 { 154 .errno = ENETUNREACH, /* ICMP_NET_ANO */ 155 .fatal = 1, 156 }, 157 { 158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */ 159 .fatal = 1, 160 }, 161 { 162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */ 163 .fatal = 0, 164 }, 165 { 166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */ 167 .fatal = 0, 168 }, 169 { 170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */ 171 .fatal = 1, 172 }, 173 { 174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */ 175 .fatal = 1, 176 }, 177 { 178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */ 179 .fatal = 1, 180 }, 181 }; 182 EXPORT_SYMBOL(icmp_err_convert); 183 184 /* 185 * ICMP control array. This specifies what to do with each ICMP. 186 */ 187 188 struct icmp_control { 189 bool (*handler)(struct sk_buff *skb); 190 short error; /* This ICMP is classed as an error message */ 191 }; 192 193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1]; 194 195 /* 196 * The ICMP socket(s). This is the most convenient way to flow control 197 * our ICMP output as well as maintain a clean interface throughout 198 * all layers. All Socketless IP sends will soon be gone. 199 * 200 * On SMP we have one ICMP socket per-cpu. 201 */ 202 static struct sock *icmp_sk(struct net *net) 203 { 204 return this_cpu_read(*net->ipv4.icmp_sk); 205 } 206 207 /* Called with BH disabled */ 208 static inline struct sock *icmp_xmit_lock(struct net *net) 209 { 210 struct sock *sk; 211 212 sk = icmp_sk(net); 213 214 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) { 215 /* This can happen if the output path signals a 216 * dst_link_failure() for an outgoing ICMP packet. 217 */ 218 return NULL; 219 } 220 return sk; 221 } 222 223 static inline void icmp_xmit_unlock(struct sock *sk) 224 { 225 spin_unlock(&sk->sk_lock.slock); 226 } 227 228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000; 229 int sysctl_icmp_msgs_burst __read_mostly = 50; 230 231 static struct { 232 spinlock_t lock; 233 u32 credit; 234 u32 stamp; 235 } icmp_global = { 236 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock), 237 }; 238 239 /** 240 * icmp_global_allow - Are we allowed to send one more ICMP message ? 241 * 242 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec. 243 * Returns false if we reached the limit and can not send another packet. 244 * Note: called with BH disabled 245 */ 246 bool icmp_global_allow(void) 247 { 248 u32 credit, delta, incr = 0, now = (u32)jiffies; 249 bool rc = false; 250 251 /* Check if token bucket is empty and cannot be refilled 252 * without taking the spinlock. The READ_ONCE() are paired 253 * with the following WRITE_ONCE() in this same function. 254 */ 255 if (!READ_ONCE(icmp_global.credit)) { 256 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ); 257 if (delta < HZ / 50) 258 return false; 259 } 260 261 spin_lock(&icmp_global.lock); 262 delta = min_t(u32, now - icmp_global.stamp, HZ); 263 if (delta >= HZ / 50) { 264 incr = sysctl_icmp_msgs_per_sec * delta / HZ ; 265 if (incr) 266 WRITE_ONCE(icmp_global.stamp, now); 267 } 268 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst); 269 if (credit) { 270 /* We want to use a credit of one in average, but need to randomize 271 * it for security reasons. 272 */ 273 credit = max_t(int, credit - prandom_u32_max(3), 0); 274 rc = true; 275 } 276 WRITE_ONCE(icmp_global.credit, credit); 277 spin_unlock(&icmp_global.lock); 278 return rc; 279 } 280 EXPORT_SYMBOL(icmp_global_allow); 281 282 static bool icmpv4_mask_allow(struct net *net, int type, int code) 283 { 284 if (type > NR_ICMP_TYPES) 285 return true; 286 287 /* Don't limit PMTU discovery. */ 288 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 289 return true; 290 291 /* Limit if icmp type is enabled in ratemask. */ 292 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask)) 293 return true; 294 295 return false; 296 } 297 298 static bool icmpv4_global_allow(struct net *net, int type, int code) 299 { 300 if (icmpv4_mask_allow(net, type, code)) 301 return true; 302 303 if (icmp_global_allow()) 304 return true; 305 306 return false; 307 } 308 309 /* 310 * Send an ICMP frame. 311 */ 312 313 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 314 struct flowi4 *fl4, int type, int code) 315 { 316 struct dst_entry *dst = &rt->dst; 317 struct inet_peer *peer; 318 bool rc = true; 319 int vif; 320 321 if (icmpv4_mask_allow(net, type, code)) 322 goto out; 323 324 /* No rate limit on loopback */ 325 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 326 goto out; 327 328 vif = l3mdev_master_ifindex(dst->dev); 329 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1); 330 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit); 331 if (peer) 332 inet_putpeer(peer); 333 out: 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 = (struct icmp_bxm *)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 icmp_bxm *icmp_param, 367 struct flowi4 *fl4, 368 struct ipcm_cookie *ipc, struct rtable **rt) 369 { 370 struct sock *sk; 371 struct sk_buff *skb; 372 373 sk = icmp_sk(dev_net((*rt)->dst.dev)); 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 struct flowi4 fl4; 407 struct sock *sk; 408 struct inet_sock *inet; 409 __be32 daddr, saddr; 410 u32 mark = IP4_REPLY_MARK(net, skb->mark); 411 int type = icmp_param->data.icmph.type; 412 int code = icmp_param->data.icmph.code; 413 414 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 415 return; 416 417 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 418 local_bh_disable(); 419 420 /* global icmp_msgs_per_sec */ 421 if (!icmpv4_global_allow(net, type, code)) 422 goto out_bh_enable; 423 424 sk = icmp_xmit_lock(net); 425 if (!sk) 426 goto out_bh_enable; 427 inet = inet_sk(sk); 428 429 icmp_param->data.icmph.checksum = 0; 430 431 ipcm_init(&ipc); 432 inet->tos = ip_hdr(skb)->tos; 433 ipc.sockc.mark = mark; 434 daddr = ipc.addr = ip_hdr(skb)->saddr; 435 saddr = fib_compute_spec_dst(skb); 436 437 if (icmp_param->replyopts.opt.opt.optlen) { 438 ipc.opt = &icmp_param->replyopts.opt; 439 if (ipc.opt->opt.srr) 440 daddr = icmp_param->replyopts.opt.opt.faddr; 441 } 442 memset(&fl4, 0, sizeof(fl4)); 443 fl4.daddr = daddr; 444 fl4.saddr = saddr; 445 fl4.flowi4_mark = mark; 446 fl4.flowi4_uid = sock_net_uid(net, NULL); 447 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 448 fl4.flowi4_proto = IPPROTO_ICMP; 449 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 450 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 451 rt = ip_route_output_key(net, &fl4); 452 if (IS_ERR(rt)) 453 goto out_unlock; 454 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 455 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 456 ip_rt_put(rt); 457 out_unlock: 458 icmp_xmit_unlock(sk); 459 out_bh_enable: 460 local_bh_enable(); 461 } 462 463 /* 464 * The device used for looking up which routing table to use for sending an ICMP 465 * error is preferably the source whenever it is set, which should ensure the 466 * icmp error can be sent to the source host, else lookup using the routing 467 * table of the destination device, else use the main routing table (index 0). 468 */ 469 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 470 { 471 struct net_device *route_lookup_dev = NULL; 472 473 if (skb->dev) 474 route_lookup_dev = skb->dev; 475 else if (skb_dst(skb)) 476 route_lookup_dev = skb_dst(skb)->dev; 477 return route_lookup_dev; 478 } 479 480 static struct rtable *icmp_route_lookup(struct net *net, 481 struct flowi4 *fl4, 482 struct sk_buff *skb_in, 483 const struct iphdr *iph, 484 __be32 saddr, u8 tos, u32 mark, 485 int type, int code, 486 struct icmp_bxm *param) 487 { 488 struct net_device *route_lookup_dev; 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_tos = RT_TOS(tos); 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(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 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 515 flowi4_to_flowi(fl4), NULL, 0); 516 if (!IS_ERR(rt)) { 517 if (rt != rt2) 518 return rt; 519 } else if (PTR_ERR(rt) == -EPERM) { 520 rt = NULL; 521 } else 522 return rt; 523 524 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 525 if (err) 526 goto relookup_failed; 527 528 if (inet_addr_type_dev_table(net, route_lookup_dev, 529 fl4_dec.saddr) == RTN_LOCAL) { 530 rt2 = __ip_route_output_key(net, &fl4_dec); 531 if (IS_ERR(rt2)) 532 err = PTR_ERR(rt2); 533 } else { 534 struct flowi4 fl4_2 = {}; 535 unsigned long orefdst; 536 537 fl4_2.daddr = fl4_dec.saddr; 538 rt2 = ip_route_output_key(net, &fl4_2); 539 if (IS_ERR(rt2)) { 540 err = PTR_ERR(rt2); 541 goto relookup_failed; 542 } 543 /* Ugh! */ 544 orefdst = skb_in->_skb_refdst; /* save old refdst */ 545 skb_dst_set(skb_in, NULL); 546 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 547 RT_TOS(tos), rt2->dst.dev); 548 549 dst_release(&rt2->dst); 550 rt2 = skb_rtable(skb_in); 551 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 552 } 553 554 if (err) 555 goto relookup_failed; 556 557 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 558 flowi4_to_flowi(&fl4_dec), NULL, 559 XFRM_LOOKUP_ICMP); 560 if (!IS_ERR(rt2)) { 561 dst_release(&rt->dst); 562 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 563 rt = rt2; 564 } else if (PTR_ERR(rt2) == -EPERM) { 565 if (rt) 566 dst_release(&rt->dst); 567 return rt2; 568 } else { 569 err = PTR_ERR(rt2); 570 goto relookup_failed; 571 } 572 return rt; 573 574 relookup_failed: 575 if (rt) 576 return rt; 577 return ERR_PTR(err); 578 } 579 580 /* 581 * Send an ICMP message in response to a situation 582 * 583 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 584 * MAY send more (we do). 585 * MUST NOT change this header information. 586 * MUST NOT reply to a multicast/broadcast IP address. 587 * MUST NOT reply to a multicast/broadcast MAC address. 588 * MUST reply to only the first fragment. 589 */ 590 591 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 592 const struct ip_options *opt) 593 { 594 struct iphdr *iph; 595 int room; 596 struct icmp_bxm icmp_param; 597 struct rtable *rt = skb_rtable(skb_in); 598 struct ipcm_cookie ipc; 599 struct flowi4 fl4; 600 __be32 saddr; 601 u8 tos; 602 u32 mark; 603 struct net *net; 604 struct sock *sk; 605 606 if (!rt) 607 goto out; 608 609 if (rt->dst.dev) 610 net = dev_net(rt->dst.dev); 611 else if (skb_in->dev) 612 net = dev_net(skb_in->dev); 613 else 614 goto out; 615 616 /* 617 * Find the original header. It is expected to be valid, of course. 618 * Check this, icmp_send is called from the most obscure devices 619 * sometimes. 620 */ 621 iph = ip_hdr(skb_in); 622 623 if ((u8 *)iph < skb_in->head || 624 (skb_network_header(skb_in) + sizeof(*iph)) > 625 skb_tail_pointer(skb_in)) 626 goto out; 627 628 /* 629 * No replies to physical multicast/broadcast 630 */ 631 if (skb_in->pkt_type != PACKET_HOST) 632 goto out; 633 634 /* 635 * Now check at the protocol level 636 */ 637 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 638 goto out; 639 640 /* 641 * Only reply to fragment 0. We byte re-order the constant 642 * mask for efficiency. 643 */ 644 if (iph->frag_off & htons(IP_OFFSET)) 645 goto out; 646 647 /* 648 * If we send an ICMP error to an ICMP error a mess would result.. 649 */ 650 if (icmp_pointers[type].error) { 651 /* 652 * We are an error, check if we are replying to an 653 * ICMP error 654 */ 655 if (iph->protocol == IPPROTO_ICMP) { 656 u8 _inner_type, *itp; 657 658 itp = skb_header_pointer(skb_in, 659 skb_network_header(skb_in) + 660 (iph->ihl << 2) + 661 offsetof(struct icmphdr, 662 type) - 663 skb_in->data, 664 sizeof(_inner_type), 665 &_inner_type); 666 if (!itp) 667 goto out; 668 669 /* 670 * Assume any unknown ICMP type is an error. This 671 * isn't specified by the RFC, but think about it.. 672 */ 673 if (*itp > NR_ICMP_TYPES || 674 icmp_pointers[*itp].error) 675 goto out; 676 } 677 } 678 679 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 680 local_bh_disable(); 681 682 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 683 * incoming dev is loopback. If outgoing dev change to not be 684 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 685 */ 686 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 687 !icmpv4_global_allow(net, type, code)) 688 goto out_bh_enable; 689 690 sk = icmp_xmit_lock(net); 691 if (!sk) 692 goto out_bh_enable; 693 694 /* 695 * Construct source address and options. 696 */ 697 698 saddr = iph->daddr; 699 if (!(rt->rt_flags & RTCF_LOCAL)) { 700 struct net_device *dev = NULL; 701 702 rcu_read_lock(); 703 if (rt_is_input_route(rt) && 704 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 705 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 706 707 if (dev) 708 saddr = inet_select_addr(dev, iph->saddr, 709 RT_SCOPE_LINK); 710 else 711 saddr = 0; 712 rcu_read_unlock(); 713 } 714 715 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 716 IPTOS_PREC_INTERNETCONTROL) : 717 iph->tos; 718 mark = IP4_REPLY_MARK(net, skb_in->mark); 719 720 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 721 goto out_unlock; 722 723 724 /* 725 * Prepare data for ICMP header. 726 */ 727 728 icmp_param.data.icmph.type = type; 729 icmp_param.data.icmph.code = code; 730 icmp_param.data.icmph.un.gateway = info; 731 icmp_param.data.icmph.checksum = 0; 732 icmp_param.skb = skb_in; 733 icmp_param.offset = skb_network_offset(skb_in); 734 inet_sk(sk)->tos = tos; 735 ipcm_init(&ipc); 736 ipc.addr = iph->saddr; 737 ipc.opt = &icmp_param.replyopts.opt; 738 ipc.sockc.mark = mark; 739 740 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark, 741 type, code, &icmp_param); 742 if (IS_ERR(rt)) 743 goto out_unlock; 744 745 /* peer icmp_ratelimit */ 746 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 747 goto ende; 748 749 /* RFC says return as much as we can without exceeding 576 bytes. */ 750 751 room = dst_mtu(&rt->dst); 752 if (room > 576) 753 room = 576; 754 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 755 room -= sizeof(struct icmphdr); 756 757 icmp_param.data_len = skb_in->len - icmp_param.offset; 758 if (icmp_param.data_len > room) 759 icmp_param.data_len = room; 760 icmp_param.head_len = sizeof(struct icmphdr); 761 762 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 763 ende: 764 ip_rt_put(rt); 765 out_unlock: 766 icmp_xmit_unlock(sk); 767 out_bh_enable: 768 local_bh_enable(); 769 out:; 770 } 771 EXPORT_SYMBOL(__icmp_send); 772 773 #if IS_ENABLED(CONFIG_NF_NAT) 774 #include <net/netfilter/nf_conntrack.h> 775 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 776 { 777 struct sk_buff *cloned_skb = NULL; 778 enum ip_conntrack_info ctinfo; 779 struct nf_conn *ct; 780 __be32 orig_ip; 781 782 ct = nf_ct_get(skb_in, &ctinfo); 783 if (!ct || !(ct->status & IPS_SRC_NAT)) { 784 icmp_send(skb_in, type, code, info); 785 return; 786 } 787 788 if (skb_shared(skb_in)) 789 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 790 791 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 792 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 793 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 794 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 795 goto out; 796 797 orig_ip = ip_hdr(skb_in)->saddr; 798 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 799 icmp_send(skb_in, type, code, info); 800 ip_hdr(skb_in)->saddr = orig_ip; 801 out: 802 consume_skb(cloned_skb); 803 } 804 EXPORT_SYMBOL(icmp_ndo_send); 805 #endif 806 807 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 808 { 809 const struct iphdr *iph = (const struct iphdr *)skb->data; 810 const struct net_protocol *ipprot; 811 int protocol = iph->protocol; 812 813 /* Checkin full IP header plus 8 bytes of protocol to 814 * avoid additional coding at protocol handlers. 815 */ 816 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 817 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 818 return; 819 } 820 821 raw_icmp_error(skb, protocol, info); 822 823 ipprot = rcu_dereference(inet_protos[protocol]); 824 if (ipprot && ipprot->err_handler) 825 ipprot->err_handler(skb, info); 826 } 827 828 static bool icmp_tag_validation(int proto) 829 { 830 bool ok; 831 832 rcu_read_lock(); 833 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 834 rcu_read_unlock(); 835 return ok; 836 } 837 838 /* 839 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 840 * ICMP_PARAMETERPROB. 841 */ 842 843 static bool icmp_unreach(struct sk_buff *skb) 844 { 845 const struct iphdr *iph; 846 struct icmphdr *icmph; 847 struct net *net; 848 u32 info = 0; 849 850 net = dev_net(skb_dst(skb)->dev); 851 852 /* 853 * Incomplete header ? 854 * Only checks for the IP header, there should be an 855 * additional check for longer headers in upper levels. 856 */ 857 858 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 859 goto out_err; 860 861 icmph = icmp_hdr(skb); 862 iph = (const struct iphdr *)skb->data; 863 864 if (iph->ihl < 5) /* Mangled header, drop. */ 865 goto out_err; 866 867 switch (icmph->type) { 868 case ICMP_DEST_UNREACH: 869 switch (icmph->code & 15) { 870 case ICMP_NET_UNREACH: 871 case ICMP_HOST_UNREACH: 872 case ICMP_PROT_UNREACH: 873 case ICMP_PORT_UNREACH: 874 break; 875 case ICMP_FRAG_NEEDED: 876 /* for documentation of the ip_no_pmtu_disc 877 * values please see 878 * Documentation/networking/ip-sysctl.rst 879 */ 880 switch (net->ipv4.sysctl_ip_no_pmtu_disc) { 881 default: 882 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 883 &iph->daddr); 884 break; 885 case 2: 886 goto out; 887 case 3: 888 if (!icmp_tag_validation(iph->protocol)) 889 goto out; 890 fallthrough; 891 case 0: 892 info = ntohs(icmph->un.frag.mtu); 893 } 894 break; 895 case ICMP_SR_FAILED: 896 net_dbg_ratelimited("%pI4: Source Route Failed\n", 897 &iph->daddr); 898 break; 899 default: 900 break; 901 } 902 if (icmph->code > NR_ICMP_UNREACH) 903 goto out; 904 break; 905 case ICMP_PARAMETERPROB: 906 info = ntohl(icmph->un.gateway) >> 24; 907 break; 908 case ICMP_TIME_EXCEEDED: 909 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 910 if (icmph->code == ICMP_EXC_FRAGTIME) 911 goto out; 912 break; 913 } 914 915 /* 916 * Throw it at our lower layers 917 * 918 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 919 * header. 920 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 921 * transport layer. 922 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 923 * transport layer. 924 */ 925 926 /* 927 * Check the other end isn't violating RFC 1122. Some routers send 928 * bogus responses to broadcast frames. If you see this message 929 * first check your netmask matches at both ends, if it does then 930 * get the other vendor to fix their kit. 931 */ 932 933 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 934 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 935 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 936 &ip_hdr(skb)->saddr, 937 icmph->type, icmph->code, 938 &iph->daddr, skb->dev->name); 939 goto out; 940 } 941 942 icmp_socket_deliver(skb, info); 943 944 out: 945 return true; 946 out_err: 947 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 948 return false; 949 } 950 951 952 /* 953 * Handle ICMP_REDIRECT. 954 */ 955 956 static bool icmp_redirect(struct sk_buff *skb) 957 { 958 if (skb->len < sizeof(struct iphdr)) { 959 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 960 return false; 961 } 962 963 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 964 /* there aught to be a stat */ 965 return false; 966 } 967 968 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 969 return true; 970 } 971 972 /* 973 * Handle ICMP_ECHO ("ping") requests. 974 * 975 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 976 * requests. 977 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 978 * included in the reply. 979 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 980 * echo requests, MUST have default=NOT. 981 * See also WRT handling of options once they are done and working. 982 */ 983 984 static bool icmp_echo(struct sk_buff *skb) 985 { 986 struct net *net; 987 988 net = dev_net(skb_dst(skb)->dev); 989 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 990 struct icmp_bxm icmp_param; 991 992 icmp_param.data.icmph = *icmp_hdr(skb); 993 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 994 icmp_param.skb = skb; 995 icmp_param.offset = 0; 996 icmp_param.data_len = skb->len; 997 icmp_param.head_len = sizeof(struct icmphdr); 998 icmp_reply(&icmp_param, skb); 999 } 1000 /* should there be an ICMP stat for ignored echos? */ 1001 return true; 1002 } 1003 1004 /* 1005 * Handle ICMP Timestamp requests. 1006 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1007 * SHOULD be in the kernel for minimum random latency. 1008 * MUST be accurate to a few minutes. 1009 * MUST be updated at least at 15Hz. 1010 */ 1011 static bool icmp_timestamp(struct sk_buff *skb) 1012 { 1013 struct icmp_bxm icmp_param; 1014 /* 1015 * Too short. 1016 */ 1017 if (skb->len < 4) 1018 goto out_err; 1019 1020 /* 1021 * Fill in the current time as ms since midnight UT: 1022 */ 1023 icmp_param.data.times[1] = inet_current_timestamp(); 1024 icmp_param.data.times[2] = icmp_param.data.times[1]; 1025 1026 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1027 1028 icmp_param.data.icmph = *icmp_hdr(skb); 1029 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1030 icmp_param.data.icmph.code = 0; 1031 icmp_param.skb = skb; 1032 icmp_param.offset = 0; 1033 icmp_param.data_len = 0; 1034 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1035 icmp_reply(&icmp_param, skb); 1036 return true; 1037 1038 out_err: 1039 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 1040 return false; 1041 } 1042 1043 static bool icmp_discard(struct sk_buff *skb) 1044 { 1045 /* pretend it was a success */ 1046 return true; 1047 } 1048 1049 /* 1050 * Deal with incoming ICMP packets. 1051 */ 1052 int icmp_rcv(struct sk_buff *skb) 1053 { 1054 struct icmphdr *icmph; 1055 struct rtable *rt = skb_rtable(skb); 1056 struct net *net = dev_net(rt->dst.dev); 1057 bool success; 1058 1059 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1060 struct sec_path *sp = skb_sec_path(skb); 1061 int nh; 1062 1063 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1064 XFRM_STATE_ICMP)) 1065 goto drop; 1066 1067 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1068 goto drop; 1069 1070 nh = skb_network_offset(skb); 1071 skb_set_network_header(skb, sizeof(*icmph)); 1072 1073 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 1074 goto drop; 1075 1076 skb_set_network_header(skb, nh); 1077 } 1078 1079 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1080 1081 if (skb_checksum_simple_validate(skb)) 1082 goto csum_error; 1083 1084 if (!pskb_pull(skb, sizeof(*icmph))) 1085 goto error; 1086 1087 icmph = icmp_hdr(skb); 1088 1089 ICMPMSGIN_INC_STATS(net, icmph->type); 1090 /* 1091 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1092 * 1093 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1094 * discarded. 1095 */ 1096 if (icmph->type > NR_ICMP_TYPES) 1097 goto error; 1098 1099 1100 /* 1101 * Parse the ICMP message 1102 */ 1103 1104 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1105 /* 1106 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1107 * silently ignored (we let user decide with a sysctl). 1108 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1109 * discarded if to broadcast/multicast. 1110 */ 1111 if ((icmph->type == ICMP_ECHO || 1112 icmph->type == ICMP_TIMESTAMP) && 1113 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1114 goto error; 1115 } 1116 if (icmph->type != ICMP_ECHO && 1117 icmph->type != ICMP_TIMESTAMP && 1118 icmph->type != ICMP_ADDRESS && 1119 icmph->type != ICMP_ADDRESSREPLY) { 1120 goto error; 1121 } 1122 } 1123 1124 success = icmp_pointers[icmph->type].handler(skb); 1125 1126 if (success) { 1127 consume_skb(skb); 1128 return NET_RX_SUCCESS; 1129 } 1130 1131 drop: 1132 kfree_skb(skb); 1133 return NET_RX_DROP; 1134 csum_error: 1135 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1136 error: 1137 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1138 goto drop; 1139 } 1140 1141 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1142 { 1143 struct icmp_extobj_hdr *objh, _objh; 1144 struct icmp_ext_hdr *exth, _exth; 1145 u16 olen; 1146 1147 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1148 if (!exth) 1149 return false; 1150 if (exth->version != 2) 1151 return true; 1152 1153 if (exth->checksum && 1154 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1155 return false; 1156 1157 off += sizeof(_exth); 1158 while (off < skb->len) { 1159 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1160 if (!objh) 1161 return false; 1162 1163 olen = ntohs(objh->length); 1164 if (olen < sizeof(_objh)) 1165 return false; 1166 1167 off += olen; 1168 if (off > skb->len) 1169 return false; 1170 } 1171 1172 return true; 1173 } 1174 1175 void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1176 struct sock_ee_data_rfc4884 *out, 1177 int thlen, int off) 1178 { 1179 int hlen; 1180 1181 /* original datagram headers: end of icmph to payload (skb->data) */ 1182 hlen = -skb_transport_offset(skb) - thlen; 1183 1184 /* per rfc 4884: minimal datagram length of 128 bytes */ 1185 if (off < 128 || off < hlen) 1186 return; 1187 1188 /* kernel has stripped headers: return payload offset in bytes */ 1189 off -= hlen; 1190 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1191 return; 1192 1193 out->len = off; 1194 1195 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1196 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1197 } 1198 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1199 1200 int icmp_err(struct sk_buff *skb, u32 info) 1201 { 1202 struct iphdr *iph = (struct iphdr *)skb->data; 1203 int offset = iph->ihl<<2; 1204 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1205 int type = icmp_hdr(skb)->type; 1206 int code = icmp_hdr(skb)->code; 1207 struct net *net = dev_net(skb->dev); 1208 1209 /* 1210 * Use ping_err to handle all icmp errors except those 1211 * triggered by ICMP_ECHOREPLY which sent from kernel. 1212 */ 1213 if (icmph->type != ICMP_ECHOREPLY) { 1214 ping_err(skb, offset, info); 1215 return 0; 1216 } 1217 1218 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1219 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1220 else if (type == ICMP_REDIRECT) 1221 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1222 1223 return 0; 1224 } 1225 1226 /* 1227 * This table is the definition of how we handle ICMP. 1228 */ 1229 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1230 [ICMP_ECHOREPLY] = { 1231 .handler = ping_rcv, 1232 }, 1233 [1] = { 1234 .handler = icmp_discard, 1235 .error = 1, 1236 }, 1237 [2] = { 1238 .handler = icmp_discard, 1239 .error = 1, 1240 }, 1241 [ICMP_DEST_UNREACH] = { 1242 .handler = icmp_unreach, 1243 .error = 1, 1244 }, 1245 [ICMP_SOURCE_QUENCH] = { 1246 .handler = icmp_unreach, 1247 .error = 1, 1248 }, 1249 [ICMP_REDIRECT] = { 1250 .handler = icmp_redirect, 1251 .error = 1, 1252 }, 1253 [6] = { 1254 .handler = icmp_discard, 1255 .error = 1, 1256 }, 1257 [7] = { 1258 .handler = icmp_discard, 1259 .error = 1, 1260 }, 1261 [ICMP_ECHO] = { 1262 .handler = icmp_echo, 1263 }, 1264 [9] = { 1265 .handler = icmp_discard, 1266 .error = 1, 1267 }, 1268 [10] = { 1269 .handler = icmp_discard, 1270 .error = 1, 1271 }, 1272 [ICMP_TIME_EXCEEDED] = { 1273 .handler = icmp_unreach, 1274 .error = 1, 1275 }, 1276 [ICMP_PARAMETERPROB] = { 1277 .handler = icmp_unreach, 1278 .error = 1, 1279 }, 1280 [ICMP_TIMESTAMP] = { 1281 .handler = icmp_timestamp, 1282 }, 1283 [ICMP_TIMESTAMPREPLY] = { 1284 .handler = icmp_discard, 1285 }, 1286 [ICMP_INFO_REQUEST] = { 1287 .handler = icmp_discard, 1288 }, 1289 [ICMP_INFO_REPLY] = { 1290 .handler = icmp_discard, 1291 }, 1292 [ICMP_ADDRESS] = { 1293 .handler = icmp_discard, 1294 }, 1295 [ICMP_ADDRESSREPLY] = { 1296 .handler = icmp_discard, 1297 }, 1298 }; 1299 1300 static void __net_exit icmp_sk_exit(struct net *net) 1301 { 1302 int i; 1303 1304 for_each_possible_cpu(i) 1305 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i)); 1306 free_percpu(net->ipv4.icmp_sk); 1307 net->ipv4.icmp_sk = NULL; 1308 } 1309 1310 static int __net_init icmp_sk_init(struct net *net) 1311 { 1312 int i, err; 1313 1314 net->ipv4.icmp_sk = alloc_percpu(struct sock *); 1315 if (!net->ipv4.icmp_sk) 1316 return -ENOMEM; 1317 1318 for_each_possible_cpu(i) { 1319 struct sock *sk; 1320 1321 err = inet_ctl_sock_create(&sk, PF_INET, 1322 SOCK_RAW, IPPROTO_ICMP, net); 1323 if (err < 0) 1324 goto fail; 1325 1326 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk; 1327 1328 /* Enough space for 2 64K ICMP packets, including 1329 * sk_buff/skb_shared_info struct overhead. 1330 */ 1331 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1332 1333 /* 1334 * Speedup sock_wfree() 1335 */ 1336 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1337 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1338 } 1339 1340 /* Control parameters for ECHO replies. */ 1341 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1342 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1343 1344 /* Control parameter - ignore bogus broadcast responses? */ 1345 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1346 1347 /* 1348 * Configurable global rate limit. 1349 * 1350 * ratelimit defines tokens/packet consumed for dst->rate_token 1351 * bucket ratemask defines which icmp types are ratelimited by 1352 * setting it's bit position. 1353 * 1354 * default: 1355 * dest unreachable (3), source quench (4), 1356 * time exceeded (11), parameter problem (12) 1357 */ 1358 1359 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1360 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1361 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1362 1363 return 0; 1364 1365 fail: 1366 icmp_sk_exit(net); 1367 return err; 1368 } 1369 1370 static struct pernet_operations __net_initdata icmp_sk_ops = { 1371 .init = icmp_sk_init, 1372 .exit = icmp_sk_exit, 1373 }; 1374 1375 int __init icmp_init(void) 1376 { 1377 return register_pernet_subsys(&icmp_sk_ops); 1378 } 1379