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 credit--; 271 rc = true; 272 } 273 WRITE_ONCE(icmp_global.credit, credit); 274 spin_unlock(&icmp_global.lock); 275 return rc; 276 } 277 EXPORT_SYMBOL(icmp_global_allow); 278 279 static bool icmpv4_mask_allow(struct net *net, int type, int code) 280 { 281 if (type > NR_ICMP_TYPES) 282 return true; 283 284 /* Don't limit PMTU discovery. */ 285 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 286 return true; 287 288 /* Limit if icmp type is enabled in ratemask. */ 289 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask)) 290 return true; 291 292 return false; 293 } 294 295 static bool icmpv4_global_allow(struct net *net, int type, int code) 296 { 297 if (icmpv4_mask_allow(net, type, code)) 298 return true; 299 300 if (icmp_global_allow()) 301 return true; 302 303 return false; 304 } 305 306 /* 307 * Send an ICMP frame. 308 */ 309 310 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 311 struct flowi4 *fl4, int type, int code) 312 { 313 struct dst_entry *dst = &rt->dst; 314 struct inet_peer *peer; 315 bool rc = true; 316 int vif; 317 318 if (icmpv4_mask_allow(net, type, code)) 319 goto out; 320 321 /* No rate limit on loopback */ 322 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 323 goto out; 324 325 vif = l3mdev_master_ifindex(dst->dev); 326 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1); 327 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit); 328 if (peer) 329 inet_putpeer(peer); 330 out: 331 return rc; 332 } 333 334 /* 335 * Maintain the counters used in the SNMP statistics for outgoing ICMP 336 */ 337 void icmp_out_count(struct net *net, unsigned char type) 338 { 339 ICMPMSGOUT_INC_STATS(net, type); 340 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 341 } 342 343 /* 344 * Checksum each fragment, and on the first include the headers and final 345 * checksum. 346 */ 347 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 348 struct sk_buff *skb) 349 { 350 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 351 __wsum csum; 352 353 csum = skb_copy_and_csum_bits(icmp_param->skb, 354 icmp_param->offset + offset, 355 to, len); 356 357 skb->csum = csum_block_add(skb->csum, csum, odd); 358 if (icmp_pointers[icmp_param->data.icmph.type].error) 359 nf_ct_attach(skb, icmp_param->skb); 360 return 0; 361 } 362 363 static void icmp_push_reply(struct icmp_bxm *icmp_param, 364 struct flowi4 *fl4, 365 struct ipcm_cookie *ipc, struct rtable **rt) 366 { 367 struct sock *sk; 368 struct sk_buff *skb; 369 370 sk = icmp_sk(dev_net((*rt)->dst.dev)); 371 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 372 icmp_param->data_len+icmp_param->head_len, 373 icmp_param->head_len, 374 ipc, rt, MSG_DONTWAIT) < 0) { 375 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 376 ip_flush_pending_frames(sk); 377 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 378 struct icmphdr *icmph = icmp_hdr(skb); 379 __wsum csum; 380 struct sk_buff *skb1; 381 382 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 383 (char *)icmph, 384 icmp_param->head_len); 385 skb_queue_walk(&sk->sk_write_queue, skb1) { 386 csum = csum_add(csum, skb1->csum); 387 } 388 icmph->checksum = csum_fold(csum); 389 skb->ip_summed = CHECKSUM_NONE; 390 ip_push_pending_frames(sk, fl4); 391 } 392 } 393 394 /* 395 * Driving logic for building and sending ICMP messages. 396 */ 397 398 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 399 { 400 struct ipcm_cookie ipc; 401 struct rtable *rt = skb_rtable(skb); 402 struct net *net = dev_net(rt->dst.dev); 403 struct flowi4 fl4; 404 struct sock *sk; 405 struct inet_sock *inet; 406 __be32 daddr, saddr; 407 u32 mark = IP4_REPLY_MARK(net, skb->mark); 408 int type = icmp_param->data.icmph.type; 409 int code = icmp_param->data.icmph.code; 410 411 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 412 return; 413 414 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 415 local_bh_disable(); 416 417 /* global icmp_msgs_per_sec */ 418 if (!icmpv4_global_allow(net, type, code)) 419 goto out_bh_enable; 420 421 sk = icmp_xmit_lock(net); 422 if (!sk) 423 goto out_bh_enable; 424 inet = inet_sk(sk); 425 426 icmp_param->data.icmph.checksum = 0; 427 428 ipcm_init(&ipc); 429 inet->tos = ip_hdr(skb)->tos; 430 ipc.sockc.mark = mark; 431 daddr = ipc.addr = ip_hdr(skb)->saddr; 432 saddr = fib_compute_spec_dst(skb); 433 434 if (icmp_param->replyopts.opt.opt.optlen) { 435 ipc.opt = &icmp_param->replyopts.opt; 436 if (ipc.opt->opt.srr) 437 daddr = icmp_param->replyopts.opt.opt.faddr; 438 } 439 memset(&fl4, 0, sizeof(fl4)); 440 fl4.daddr = daddr; 441 fl4.saddr = saddr; 442 fl4.flowi4_mark = mark; 443 fl4.flowi4_uid = sock_net_uid(net, NULL); 444 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 445 fl4.flowi4_proto = IPPROTO_ICMP; 446 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 447 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); 448 rt = ip_route_output_key(net, &fl4); 449 if (IS_ERR(rt)) 450 goto out_unlock; 451 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 452 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 453 ip_rt_put(rt); 454 out_unlock: 455 icmp_xmit_unlock(sk); 456 out_bh_enable: 457 local_bh_enable(); 458 } 459 460 /* 461 * The device used for looking up which routing table to use for sending an ICMP 462 * error is preferably the source whenever it is set, which should ensure the 463 * icmp error can be sent to the source host, else lookup using the routing 464 * table of the destination device, else use the main routing table (index 0). 465 */ 466 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 467 { 468 struct net_device *route_lookup_dev = NULL; 469 470 if (skb->dev) 471 route_lookup_dev = skb->dev; 472 else if (skb_dst(skb)) 473 route_lookup_dev = skb_dst(skb)->dev; 474 return route_lookup_dev; 475 } 476 477 static struct rtable *icmp_route_lookup(struct net *net, 478 struct flowi4 *fl4, 479 struct sk_buff *skb_in, 480 const struct iphdr *iph, 481 __be32 saddr, u8 tos, u32 mark, 482 int type, int code, 483 struct icmp_bxm *param) 484 { 485 struct net_device *route_lookup_dev; 486 struct rtable *rt, *rt2; 487 struct flowi4 fl4_dec; 488 int err; 489 490 memset(fl4, 0, sizeof(*fl4)); 491 fl4->daddr = (param->replyopts.opt.opt.srr ? 492 param->replyopts.opt.opt.faddr : iph->saddr); 493 fl4->saddr = saddr; 494 fl4->flowi4_mark = mark; 495 fl4->flowi4_uid = sock_net_uid(net, NULL); 496 fl4->flowi4_tos = RT_TOS(tos); 497 fl4->flowi4_proto = IPPROTO_ICMP; 498 fl4->fl4_icmp_type = type; 499 fl4->fl4_icmp_code = code; 500 route_lookup_dev = icmp_get_route_lookup_dev(skb_in); 501 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev); 502 503 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4)); 504 rt = ip_route_output_key_hash(net, fl4, skb_in); 505 if (IS_ERR(rt)) 506 return rt; 507 508 /* No need to clone since we're just using its address. */ 509 rt2 = rt; 510 511 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 512 flowi4_to_flowi(fl4), NULL, 0); 513 if (!IS_ERR(rt)) { 514 if (rt != rt2) 515 return rt; 516 } else if (PTR_ERR(rt) == -EPERM) { 517 rt = NULL; 518 } else 519 return rt; 520 521 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 522 if (err) 523 goto relookup_failed; 524 525 if (inet_addr_type_dev_table(net, route_lookup_dev, 526 fl4_dec.saddr) == RTN_LOCAL) { 527 rt2 = __ip_route_output_key(net, &fl4_dec); 528 if (IS_ERR(rt2)) 529 err = PTR_ERR(rt2); 530 } else { 531 struct flowi4 fl4_2 = {}; 532 unsigned long orefdst; 533 534 fl4_2.daddr = fl4_dec.saddr; 535 rt2 = ip_route_output_key(net, &fl4_2); 536 if (IS_ERR(rt2)) { 537 err = PTR_ERR(rt2); 538 goto relookup_failed; 539 } 540 /* Ugh! */ 541 orefdst = skb_in->_skb_refdst; /* save old refdst */ 542 skb_dst_set(skb_in, NULL); 543 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 544 RT_TOS(tos), rt2->dst.dev); 545 546 dst_release(&rt2->dst); 547 rt2 = skb_rtable(skb_in); 548 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 549 } 550 551 if (err) 552 goto relookup_failed; 553 554 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 555 flowi4_to_flowi(&fl4_dec), NULL, 556 XFRM_LOOKUP_ICMP); 557 if (!IS_ERR(rt2)) { 558 dst_release(&rt->dst); 559 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 560 rt = rt2; 561 } else if (PTR_ERR(rt2) == -EPERM) { 562 if (rt) 563 dst_release(&rt->dst); 564 return rt2; 565 } else { 566 err = PTR_ERR(rt2); 567 goto relookup_failed; 568 } 569 return rt; 570 571 relookup_failed: 572 if (rt) 573 return rt; 574 return ERR_PTR(err); 575 } 576 577 /* 578 * Send an ICMP message in response to a situation 579 * 580 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 581 * MAY send more (we do). 582 * MUST NOT change this header information. 583 * MUST NOT reply to a multicast/broadcast IP address. 584 * MUST NOT reply to a multicast/broadcast MAC address. 585 * MUST reply to only the first fragment. 586 */ 587 588 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 589 const struct ip_options *opt) 590 { 591 struct iphdr *iph; 592 int room; 593 struct icmp_bxm icmp_param; 594 struct rtable *rt = skb_rtable(skb_in); 595 struct ipcm_cookie ipc; 596 struct flowi4 fl4; 597 __be32 saddr; 598 u8 tos; 599 u32 mark; 600 struct net *net; 601 struct sock *sk; 602 603 if (!rt) 604 goto out; 605 606 if (rt->dst.dev) 607 net = dev_net(rt->dst.dev); 608 else if (skb_in->dev) 609 net = dev_net(skb_in->dev); 610 else 611 goto out; 612 613 /* 614 * Find the original header. It is expected to be valid, of course. 615 * Check this, icmp_send is called from the most obscure devices 616 * sometimes. 617 */ 618 iph = ip_hdr(skb_in); 619 620 if ((u8 *)iph < skb_in->head || 621 (skb_network_header(skb_in) + sizeof(*iph)) > 622 skb_tail_pointer(skb_in)) 623 goto out; 624 625 /* 626 * No replies to physical multicast/broadcast 627 */ 628 if (skb_in->pkt_type != PACKET_HOST) 629 goto out; 630 631 /* 632 * Now check at the protocol level 633 */ 634 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 635 goto out; 636 637 /* 638 * Only reply to fragment 0. We byte re-order the constant 639 * mask for efficiency. 640 */ 641 if (iph->frag_off & htons(IP_OFFSET)) 642 goto out; 643 644 /* 645 * If we send an ICMP error to an ICMP error a mess would result.. 646 */ 647 if (icmp_pointers[type].error) { 648 /* 649 * We are an error, check if we are replying to an 650 * ICMP error 651 */ 652 if (iph->protocol == IPPROTO_ICMP) { 653 u8 _inner_type, *itp; 654 655 itp = skb_header_pointer(skb_in, 656 skb_network_header(skb_in) + 657 (iph->ihl << 2) + 658 offsetof(struct icmphdr, 659 type) - 660 skb_in->data, 661 sizeof(_inner_type), 662 &_inner_type); 663 if (!itp) 664 goto out; 665 666 /* 667 * Assume any unknown ICMP type is an error. This 668 * isn't specified by the RFC, but think about it.. 669 */ 670 if (*itp > NR_ICMP_TYPES || 671 icmp_pointers[*itp].error) 672 goto out; 673 } 674 } 675 676 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 677 local_bh_disable(); 678 679 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 680 * incoming dev is loopback. If outgoing dev change to not be 681 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 682 */ 683 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 684 !icmpv4_global_allow(net, type, code)) 685 goto out_bh_enable; 686 687 sk = icmp_xmit_lock(net); 688 if (!sk) 689 goto out_bh_enable; 690 691 /* 692 * Construct source address and options. 693 */ 694 695 saddr = iph->daddr; 696 if (!(rt->rt_flags & RTCF_LOCAL)) { 697 struct net_device *dev = NULL; 698 699 rcu_read_lock(); 700 if (rt_is_input_route(rt) && 701 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 702 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 703 704 if (dev) 705 saddr = inet_select_addr(dev, iph->saddr, 706 RT_SCOPE_LINK); 707 else 708 saddr = 0; 709 rcu_read_unlock(); 710 } 711 712 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 713 IPTOS_PREC_INTERNETCONTROL) : 714 iph->tos; 715 mark = IP4_REPLY_MARK(net, skb_in->mark); 716 717 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 718 goto out_unlock; 719 720 721 /* 722 * Prepare data for ICMP header. 723 */ 724 725 icmp_param.data.icmph.type = type; 726 icmp_param.data.icmph.code = code; 727 icmp_param.data.icmph.un.gateway = info; 728 icmp_param.data.icmph.checksum = 0; 729 icmp_param.skb = skb_in; 730 icmp_param.offset = skb_network_offset(skb_in); 731 inet_sk(sk)->tos = tos; 732 ipcm_init(&ipc); 733 ipc.addr = iph->saddr; 734 ipc.opt = &icmp_param.replyopts.opt; 735 ipc.sockc.mark = mark; 736 737 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark, 738 type, code, &icmp_param); 739 if (IS_ERR(rt)) 740 goto out_unlock; 741 742 /* peer icmp_ratelimit */ 743 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 744 goto ende; 745 746 /* RFC says return as much as we can without exceeding 576 bytes. */ 747 748 room = dst_mtu(&rt->dst); 749 if (room > 576) 750 room = 576; 751 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 752 room -= sizeof(struct icmphdr); 753 754 icmp_param.data_len = skb_in->len - icmp_param.offset; 755 if (icmp_param.data_len > room) 756 icmp_param.data_len = room; 757 icmp_param.head_len = sizeof(struct icmphdr); 758 759 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 760 ende: 761 ip_rt_put(rt); 762 out_unlock: 763 icmp_xmit_unlock(sk); 764 out_bh_enable: 765 local_bh_enable(); 766 out:; 767 } 768 EXPORT_SYMBOL(__icmp_send); 769 770 #if IS_ENABLED(CONFIG_NF_NAT) 771 #include <net/netfilter/nf_conntrack.h> 772 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 773 { 774 struct sk_buff *cloned_skb = NULL; 775 enum ip_conntrack_info ctinfo; 776 struct nf_conn *ct; 777 __be32 orig_ip; 778 779 ct = nf_ct_get(skb_in, &ctinfo); 780 if (!ct || !(ct->status & IPS_SRC_NAT)) { 781 icmp_send(skb_in, type, code, info); 782 return; 783 } 784 785 if (skb_shared(skb_in)) 786 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 787 788 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 789 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 790 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 791 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 792 goto out; 793 794 orig_ip = ip_hdr(skb_in)->saddr; 795 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 796 icmp_send(skb_in, type, code, info); 797 ip_hdr(skb_in)->saddr = orig_ip; 798 out: 799 consume_skb(cloned_skb); 800 } 801 EXPORT_SYMBOL(icmp_ndo_send); 802 #endif 803 804 static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 805 { 806 const struct iphdr *iph = (const struct iphdr *)skb->data; 807 const struct net_protocol *ipprot; 808 int protocol = iph->protocol; 809 810 /* Checkin full IP header plus 8 bytes of protocol to 811 * avoid additional coding at protocol handlers. 812 */ 813 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 814 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 815 return; 816 } 817 818 raw_icmp_error(skb, protocol, info); 819 820 ipprot = rcu_dereference(inet_protos[protocol]); 821 if (ipprot && ipprot->err_handler) 822 ipprot->err_handler(skb, info); 823 } 824 825 static bool icmp_tag_validation(int proto) 826 { 827 bool ok; 828 829 rcu_read_lock(); 830 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 831 rcu_read_unlock(); 832 return ok; 833 } 834 835 /* 836 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 837 * ICMP_PARAMETERPROB. 838 */ 839 840 static bool icmp_unreach(struct sk_buff *skb) 841 { 842 const struct iphdr *iph; 843 struct icmphdr *icmph; 844 struct net *net; 845 u32 info = 0; 846 847 net = dev_net(skb_dst(skb)->dev); 848 849 /* 850 * Incomplete header ? 851 * Only checks for the IP header, there should be an 852 * additional check for longer headers in upper levels. 853 */ 854 855 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 856 goto out_err; 857 858 icmph = icmp_hdr(skb); 859 iph = (const struct iphdr *)skb->data; 860 861 if (iph->ihl < 5) /* Mangled header, drop. */ 862 goto out_err; 863 864 switch (icmph->type) { 865 case ICMP_DEST_UNREACH: 866 switch (icmph->code & 15) { 867 case ICMP_NET_UNREACH: 868 case ICMP_HOST_UNREACH: 869 case ICMP_PROT_UNREACH: 870 case ICMP_PORT_UNREACH: 871 break; 872 case ICMP_FRAG_NEEDED: 873 /* for documentation of the ip_no_pmtu_disc 874 * values please see 875 * Documentation/networking/ip-sysctl.rst 876 */ 877 switch (net->ipv4.sysctl_ip_no_pmtu_disc) { 878 default: 879 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 880 &iph->daddr); 881 break; 882 case 2: 883 goto out; 884 case 3: 885 if (!icmp_tag_validation(iph->protocol)) 886 goto out; 887 fallthrough; 888 case 0: 889 info = ntohs(icmph->un.frag.mtu); 890 } 891 break; 892 case ICMP_SR_FAILED: 893 net_dbg_ratelimited("%pI4: Source Route Failed\n", 894 &iph->daddr); 895 break; 896 default: 897 break; 898 } 899 if (icmph->code > NR_ICMP_UNREACH) 900 goto out; 901 break; 902 case ICMP_PARAMETERPROB: 903 info = ntohl(icmph->un.gateway) >> 24; 904 break; 905 case ICMP_TIME_EXCEEDED: 906 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 907 if (icmph->code == ICMP_EXC_FRAGTIME) 908 goto out; 909 break; 910 } 911 912 /* 913 * Throw it at our lower layers 914 * 915 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 916 * header. 917 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 918 * transport layer. 919 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 920 * transport layer. 921 */ 922 923 /* 924 * Check the other end isn't violating RFC 1122. Some routers send 925 * bogus responses to broadcast frames. If you see this message 926 * first check your netmask matches at both ends, if it does then 927 * get the other vendor to fix their kit. 928 */ 929 930 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 931 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 932 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 933 &ip_hdr(skb)->saddr, 934 icmph->type, icmph->code, 935 &iph->daddr, skb->dev->name); 936 goto out; 937 } 938 939 icmp_socket_deliver(skb, info); 940 941 out: 942 return true; 943 out_err: 944 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 945 return false; 946 } 947 948 949 /* 950 * Handle ICMP_REDIRECT. 951 */ 952 953 static bool icmp_redirect(struct sk_buff *skb) 954 { 955 if (skb->len < sizeof(struct iphdr)) { 956 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 957 return false; 958 } 959 960 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 961 /* there aught to be a stat */ 962 return false; 963 } 964 965 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 966 return true; 967 } 968 969 /* 970 * Handle ICMP_ECHO ("ping") requests. 971 * 972 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 973 * requests. 974 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 975 * included in the reply. 976 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 977 * echo requests, MUST have default=NOT. 978 * See also WRT handling of options once they are done and working. 979 */ 980 981 static bool icmp_echo(struct sk_buff *skb) 982 { 983 struct net *net; 984 985 net = dev_net(skb_dst(skb)->dev); 986 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 987 struct icmp_bxm icmp_param; 988 989 icmp_param.data.icmph = *icmp_hdr(skb); 990 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 991 icmp_param.skb = skb; 992 icmp_param.offset = 0; 993 icmp_param.data_len = skb->len; 994 icmp_param.head_len = sizeof(struct icmphdr); 995 icmp_reply(&icmp_param, skb); 996 } 997 /* should there be an ICMP stat for ignored echos? */ 998 return true; 999 } 1000 1001 /* 1002 * Handle ICMP Timestamp requests. 1003 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1004 * SHOULD be in the kernel for minimum random latency. 1005 * MUST be accurate to a few minutes. 1006 * MUST be updated at least at 15Hz. 1007 */ 1008 static bool icmp_timestamp(struct sk_buff *skb) 1009 { 1010 struct icmp_bxm icmp_param; 1011 /* 1012 * Too short. 1013 */ 1014 if (skb->len < 4) 1015 goto out_err; 1016 1017 /* 1018 * Fill in the current time as ms since midnight UT: 1019 */ 1020 icmp_param.data.times[1] = inet_current_timestamp(); 1021 icmp_param.data.times[2] = icmp_param.data.times[1]; 1022 1023 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1024 1025 icmp_param.data.icmph = *icmp_hdr(skb); 1026 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1027 icmp_param.data.icmph.code = 0; 1028 icmp_param.skb = skb; 1029 icmp_param.offset = 0; 1030 icmp_param.data_len = 0; 1031 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1032 icmp_reply(&icmp_param, skb); 1033 return true; 1034 1035 out_err: 1036 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 1037 return false; 1038 } 1039 1040 static bool icmp_discard(struct sk_buff *skb) 1041 { 1042 /* pretend it was a success */ 1043 return true; 1044 } 1045 1046 /* 1047 * Deal with incoming ICMP packets. 1048 */ 1049 int icmp_rcv(struct sk_buff *skb) 1050 { 1051 struct icmphdr *icmph; 1052 struct rtable *rt = skb_rtable(skb); 1053 struct net *net = dev_net(rt->dst.dev); 1054 bool success; 1055 1056 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1057 struct sec_path *sp = skb_sec_path(skb); 1058 int nh; 1059 1060 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1061 XFRM_STATE_ICMP)) 1062 goto drop; 1063 1064 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1065 goto drop; 1066 1067 nh = skb_network_offset(skb); 1068 skb_set_network_header(skb, sizeof(*icmph)); 1069 1070 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 1071 goto drop; 1072 1073 skb_set_network_header(skb, nh); 1074 } 1075 1076 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1077 1078 if (skb_checksum_simple_validate(skb)) 1079 goto csum_error; 1080 1081 if (!pskb_pull(skb, sizeof(*icmph))) 1082 goto error; 1083 1084 icmph = icmp_hdr(skb); 1085 1086 ICMPMSGIN_INC_STATS(net, icmph->type); 1087 /* 1088 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1089 * 1090 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1091 * discarded. 1092 */ 1093 if (icmph->type > NR_ICMP_TYPES) 1094 goto error; 1095 1096 1097 /* 1098 * Parse the ICMP message 1099 */ 1100 1101 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1102 /* 1103 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1104 * silently ignored (we let user decide with a sysctl). 1105 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1106 * discarded if to broadcast/multicast. 1107 */ 1108 if ((icmph->type == ICMP_ECHO || 1109 icmph->type == ICMP_TIMESTAMP) && 1110 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1111 goto error; 1112 } 1113 if (icmph->type != ICMP_ECHO && 1114 icmph->type != ICMP_TIMESTAMP && 1115 icmph->type != ICMP_ADDRESS && 1116 icmph->type != ICMP_ADDRESSREPLY) { 1117 goto error; 1118 } 1119 } 1120 1121 success = icmp_pointers[icmph->type].handler(skb); 1122 1123 if (success) { 1124 consume_skb(skb); 1125 return NET_RX_SUCCESS; 1126 } 1127 1128 drop: 1129 kfree_skb(skb); 1130 return NET_RX_DROP; 1131 csum_error: 1132 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1133 error: 1134 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1135 goto drop; 1136 } 1137 1138 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1139 { 1140 struct icmp_extobj_hdr *objh, _objh; 1141 struct icmp_ext_hdr *exth, _exth; 1142 u16 olen; 1143 1144 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1145 if (!exth) 1146 return false; 1147 if (exth->version != 2) 1148 return true; 1149 1150 if (exth->checksum && 1151 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1152 return false; 1153 1154 off += sizeof(_exth); 1155 while (off < skb->len) { 1156 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1157 if (!objh) 1158 return false; 1159 1160 olen = ntohs(objh->length); 1161 if (olen < sizeof(_objh)) 1162 return false; 1163 1164 off += olen; 1165 if (off > skb->len) 1166 return false; 1167 } 1168 1169 return true; 1170 } 1171 1172 void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1173 struct sock_ee_data_rfc4884 *out, 1174 int thlen, int off) 1175 { 1176 int hlen; 1177 1178 /* original datagram headers: end of icmph to payload (skb->data) */ 1179 hlen = -skb_transport_offset(skb) - thlen; 1180 1181 /* per rfc 4884: minimal datagram length of 128 bytes */ 1182 if (off < 128 || off < hlen) 1183 return; 1184 1185 /* kernel has stripped headers: return payload offset in bytes */ 1186 off -= hlen; 1187 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1188 return; 1189 1190 out->len = off; 1191 1192 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1193 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1194 } 1195 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1196 1197 int icmp_err(struct sk_buff *skb, u32 info) 1198 { 1199 struct iphdr *iph = (struct iphdr *)skb->data; 1200 int offset = iph->ihl<<2; 1201 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1202 int type = icmp_hdr(skb)->type; 1203 int code = icmp_hdr(skb)->code; 1204 struct net *net = dev_net(skb->dev); 1205 1206 /* 1207 * Use ping_err to handle all icmp errors except those 1208 * triggered by ICMP_ECHOREPLY which sent from kernel. 1209 */ 1210 if (icmph->type != ICMP_ECHOREPLY) { 1211 ping_err(skb, offset, info); 1212 return 0; 1213 } 1214 1215 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1216 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1217 else if (type == ICMP_REDIRECT) 1218 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1219 1220 return 0; 1221 } 1222 1223 /* 1224 * This table is the definition of how we handle ICMP. 1225 */ 1226 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1227 [ICMP_ECHOREPLY] = { 1228 .handler = ping_rcv, 1229 }, 1230 [1] = { 1231 .handler = icmp_discard, 1232 .error = 1, 1233 }, 1234 [2] = { 1235 .handler = icmp_discard, 1236 .error = 1, 1237 }, 1238 [ICMP_DEST_UNREACH] = { 1239 .handler = icmp_unreach, 1240 .error = 1, 1241 }, 1242 [ICMP_SOURCE_QUENCH] = { 1243 .handler = icmp_unreach, 1244 .error = 1, 1245 }, 1246 [ICMP_REDIRECT] = { 1247 .handler = icmp_redirect, 1248 .error = 1, 1249 }, 1250 [6] = { 1251 .handler = icmp_discard, 1252 .error = 1, 1253 }, 1254 [7] = { 1255 .handler = icmp_discard, 1256 .error = 1, 1257 }, 1258 [ICMP_ECHO] = { 1259 .handler = icmp_echo, 1260 }, 1261 [9] = { 1262 .handler = icmp_discard, 1263 .error = 1, 1264 }, 1265 [10] = { 1266 .handler = icmp_discard, 1267 .error = 1, 1268 }, 1269 [ICMP_TIME_EXCEEDED] = { 1270 .handler = icmp_unreach, 1271 .error = 1, 1272 }, 1273 [ICMP_PARAMETERPROB] = { 1274 .handler = icmp_unreach, 1275 .error = 1, 1276 }, 1277 [ICMP_TIMESTAMP] = { 1278 .handler = icmp_timestamp, 1279 }, 1280 [ICMP_TIMESTAMPREPLY] = { 1281 .handler = icmp_discard, 1282 }, 1283 [ICMP_INFO_REQUEST] = { 1284 .handler = icmp_discard, 1285 }, 1286 [ICMP_INFO_REPLY] = { 1287 .handler = icmp_discard, 1288 }, 1289 [ICMP_ADDRESS] = { 1290 .handler = icmp_discard, 1291 }, 1292 [ICMP_ADDRESSREPLY] = { 1293 .handler = icmp_discard, 1294 }, 1295 }; 1296 1297 static void __net_exit icmp_sk_exit(struct net *net) 1298 { 1299 int i; 1300 1301 for_each_possible_cpu(i) 1302 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i)); 1303 free_percpu(net->ipv4.icmp_sk); 1304 net->ipv4.icmp_sk = NULL; 1305 } 1306 1307 static int __net_init icmp_sk_init(struct net *net) 1308 { 1309 int i, err; 1310 1311 net->ipv4.icmp_sk = alloc_percpu(struct sock *); 1312 if (!net->ipv4.icmp_sk) 1313 return -ENOMEM; 1314 1315 for_each_possible_cpu(i) { 1316 struct sock *sk; 1317 1318 err = inet_ctl_sock_create(&sk, PF_INET, 1319 SOCK_RAW, IPPROTO_ICMP, net); 1320 if (err < 0) 1321 goto fail; 1322 1323 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk; 1324 1325 /* Enough space for 2 64K ICMP packets, including 1326 * sk_buff/skb_shared_info struct overhead. 1327 */ 1328 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1329 1330 /* 1331 * Speedup sock_wfree() 1332 */ 1333 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1334 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1335 } 1336 1337 /* Control parameters for ECHO replies. */ 1338 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1339 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1340 1341 /* Control parameter - ignore bogus broadcast responses? */ 1342 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1343 1344 /* 1345 * Configurable global rate limit. 1346 * 1347 * ratelimit defines tokens/packet consumed for dst->rate_token 1348 * bucket ratemask defines which icmp types are ratelimited by 1349 * setting it's bit position. 1350 * 1351 * default: 1352 * dest unreachable (3), source quench (4), 1353 * time exceeded (11), parameter problem (12) 1354 */ 1355 1356 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1357 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1358 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1359 1360 return 0; 1361 1362 fail: 1363 icmp_sk_exit(net); 1364 return err; 1365 } 1366 1367 static struct pernet_operations __net_initdata icmp_sk_ops = { 1368 .init = icmp_sk_init, 1369 .exit = icmp_sk_exit, 1370 }; 1371 1372 int __init icmp_init(void) 1373 { 1374 return register_pernet_subsys(&icmp_sk_ops); 1375 } 1376