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 100struct 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 116const 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}; 182EXPORT_SYMBOL(icmp_err_convert); 183 184/* 185 * ICMP control array. This specifies what to do with each ICMP. 186 */ 187 188struct icmp_control { 189 bool (*handler)(struct sk_buff *skb); 190 short error; /* This ICMP is classed as an error message */ 191}; 192 193static 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 */ 202static struct sock *icmp_sk(struct net *net) 203{ 204 return this_cpu_read(*net->ipv4.icmp_sk); 205} 206 207/* Called with BH disabled */ 208static 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 223static inline void icmp_xmit_unlock(struct sock *sk) 224{ 225 spin_unlock(&sk->sk_lock.slock); 226} 227 228int sysctl_icmp_msgs_per_sec __read_mostly = 1000; 229int sysctl_icmp_msgs_burst __read_mostly = 50; 230 231static 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 */ 246bool 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 = READ_ONCE(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, 269 READ_ONCE(sysctl_icmp_msgs_burst)); 270 if (credit) { 271 /* We want to use a credit of one in average, but need to randomize 272 * it for security reasons. 273 */ 274 credit = max_t(int, credit - prandom_u32_max(3), 0); 275 rc = true; 276 } 277 WRITE_ONCE(icmp_global.credit, credit); 278 spin_unlock(&icmp_global.lock); 279 return rc; 280} 281EXPORT_SYMBOL(icmp_global_allow); 282 283static bool icmpv4_mask_allow(struct net *net, int type, int code) 284{ 285 if (type > NR_ICMP_TYPES) 286 return true; 287 288 /* Don't limit PMTU discovery. */ 289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 290 return true; 291 292 /* Limit if icmp type is enabled in ratemask. */ 293 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask))) 294 return true; 295 296 return false; 297} 298 299static bool icmpv4_global_allow(struct net *net, int type, int code) 300{ 301 if (icmpv4_mask_allow(net, type, code)) 302 return true; 303 304 if (icmp_global_allow()) 305 return true; 306 307 return false; 308} 309 310/* 311 * Send an ICMP frame. 312 */ 313 314static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt, 315 struct flowi4 *fl4, int type, int code) 316{ 317 struct dst_entry *dst = &rt->dst; 318 struct inet_peer *peer; 319 bool rc = true; 320 int vif; 321 322 if (icmpv4_mask_allow(net, type, code)) 323 goto out; 324 325 /* No rate limit on loopback */ 326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) 327 goto out; 328 329 vif = l3mdev_master_ifindex(dst->dev); 330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1); 331 rc = inet_peer_xrlim_allow(peer, 332 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit)); 333 if (peer) 334 inet_putpeer(peer); 335out: 336 return rc; 337} 338 339/* 340 * Maintain the counters used in the SNMP statistics for outgoing ICMP 341 */ 342void icmp_out_count(struct net *net, unsigned char type) 343{ 344 ICMPMSGOUT_INC_STATS(net, type); 345 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS); 346} 347 348/* 349 * Checksum each fragment, and on the first include the headers and final 350 * checksum. 351 */ 352static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd, 353 struct sk_buff *skb) 354{ 355 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from; 356 __wsum csum; 357 358 csum = skb_copy_and_csum_bits(icmp_param->skb, 359 icmp_param->offset + offset, 360 to, len); 361 362 skb->csum = csum_block_add(skb->csum, csum, odd); 363 if (icmp_pointers[icmp_param->data.icmph.type].error) 364 nf_ct_attach(skb, icmp_param->skb); 365 return 0; 366} 367 368static void icmp_push_reply(struct icmp_bxm *icmp_param, 369 struct flowi4 *fl4, 370 struct ipcm_cookie *ipc, struct rtable **rt) 371{ 372 struct sock *sk; 373 struct sk_buff *skb; 374 375 sk = icmp_sk(dev_net((*rt)->dst.dev)); 376 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param, 377 icmp_param->data_len+icmp_param->head_len, 378 icmp_param->head_len, 379 ipc, rt, MSG_DONTWAIT) < 0) { 380 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS); 381 ip_flush_pending_frames(sk); 382 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { 383 struct icmphdr *icmph = icmp_hdr(skb); 384 __wsum csum; 385 struct sk_buff *skb1; 386 387 csum = csum_partial_copy_nocheck((void *)&icmp_param->data, 388 (char *)icmph, 389 icmp_param->head_len); 390 skb_queue_walk(&sk->sk_write_queue, skb1) { 391 csum = csum_add(csum, skb1->csum); 392 } 393 icmph->checksum = csum_fold(csum); 394 skb->ip_summed = CHECKSUM_NONE; 395 ip_push_pending_frames(sk, fl4); 396 } 397} 398 399/* 400 * Driving logic for building and sending ICMP messages. 401 */ 402 403static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb) 404{ 405 struct ipcm_cookie ipc; 406 struct rtable *rt = skb_rtable(skb); 407 struct net *net = dev_net(rt->dst.dev); 408 struct flowi4 fl4; 409 struct sock *sk; 410 struct inet_sock *inet; 411 __be32 daddr, saddr; 412 u32 mark = IP4_REPLY_MARK(net, skb->mark); 413 int type = icmp_param->data.icmph.type; 414 int code = icmp_param->data.icmph.code; 415 416 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb)) 417 return; 418 419 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 420 local_bh_disable(); 421 422 /* global icmp_msgs_per_sec */ 423 if (!icmpv4_global_allow(net, type, code)) 424 goto out_bh_enable; 425 426 sk = icmp_xmit_lock(net); 427 if (!sk) 428 goto out_bh_enable; 429 inet = inet_sk(sk); 430 431 icmp_param->data.icmph.checksum = 0; 432 433 ipcm_init(&ipc); 434 inet->tos = ip_hdr(skb)->tos; 435 ipc.sockc.mark = mark; 436 daddr = ipc.addr = ip_hdr(skb)->saddr; 437 saddr = fib_compute_spec_dst(skb); 438 439 if (icmp_param->replyopts.opt.opt.optlen) { 440 ipc.opt = &icmp_param->replyopts.opt; 441 if (ipc.opt->opt.srr) 442 daddr = icmp_param->replyopts.opt.opt.faddr; 443 } 444 memset(&fl4, 0, sizeof(fl4)); 445 fl4.daddr = daddr; 446 fl4.saddr = saddr; 447 fl4.flowi4_mark = mark; 448 fl4.flowi4_uid = sock_net_uid(net, NULL); 449 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos); 450 fl4.flowi4_proto = IPPROTO_ICMP; 451 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev); 452 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4)); 453 rt = ip_route_output_key(net, &fl4); 454 if (IS_ERR(rt)) 455 goto out_unlock; 456 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 457 icmp_push_reply(icmp_param, &fl4, &ipc, &rt); 458 ip_rt_put(rt); 459out_unlock: 460 icmp_xmit_unlock(sk); 461out_bh_enable: 462 local_bh_enable(); 463} 464 465/* 466 * The device used for looking up which routing table to use for sending an ICMP 467 * error is preferably the source whenever it is set, which should ensure the 468 * icmp error can be sent to the source host, else lookup using the routing 469 * table of the destination device, else use the main routing table (index 0). 470 */ 471static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb) 472{ 473 struct net_device *route_lookup_dev = NULL; 474 475 if (skb->dev) 476 route_lookup_dev = skb->dev; 477 else if (skb_dst(skb)) 478 route_lookup_dev = skb_dst(skb)->dev; 479 return route_lookup_dev; 480} 481 482static struct rtable *icmp_route_lookup(struct net *net, 483 struct flowi4 *fl4, 484 struct sk_buff *skb_in, 485 const struct iphdr *iph, 486 __be32 saddr, u8 tos, u32 mark, 487 int type, int code, 488 struct icmp_bxm *param) 489{ 490 struct net_device *route_lookup_dev; 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 = RT_TOS(tos); 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 rt = (struct rtable *) xfrm_lookup(net, &rt->dst, 517 flowi4_to_flowi(fl4), NULL, 0); 518 if (!IS_ERR(rt)) { 519 if (rt != rt2) 520 return rt; 521 } else if (PTR_ERR(rt) == -EPERM) { 522 rt = NULL; 523 } else 524 return rt; 525 526 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET); 527 if (err) 528 goto relookup_failed; 529 530 if (inet_addr_type_dev_table(net, route_lookup_dev, 531 fl4_dec.saddr) == RTN_LOCAL) { 532 rt2 = __ip_route_output_key(net, &fl4_dec); 533 if (IS_ERR(rt2)) 534 err = PTR_ERR(rt2); 535 } else { 536 struct flowi4 fl4_2 = {}; 537 unsigned long orefdst; 538 539 fl4_2.daddr = fl4_dec.saddr; 540 rt2 = ip_route_output_key(net, &fl4_2); 541 if (IS_ERR(rt2)) { 542 err = PTR_ERR(rt2); 543 goto relookup_failed; 544 } 545 /* Ugh! */ 546 orefdst = skb_in->_skb_refdst; /* save old refdst */ 547 skb_dst_set(skb_in, NULL); 548 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr, 549 RT_TOS(tos), rt2->dst.dev); 550 551 dst_release(&rt2->dst); 552 rt2 = skb_rtable(skb_in); 553 skb_in->_skb_refdst = orefdst; /* restore old refdst */ 554 } 555 556 if (err) 557 goto relookup_failed; 558 559 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst, 560 flowi4_to_flowi(&fl4_dec), NULL, 561 XFRM_LOOKUP_ICMP); 562 if (!IS_ERR(rt2)) { 563 dst_release(&rt->dst); 564 memcpy(fl4, &fl4_dec, sizeof(*fl4)); 565 rt = rt2; 566 } else if (PTR_ERR(rt2) == -EPERM) { 567 if (rt) 568 dst_release(&rt->dst); 569 return rt2; 570 } else { 571 err = PTR_ERR(rt2); 572 goto relookup_failed; 573 } 574 return rt; 575 576relookup_failed: 577 if (rt) 578 return rt; 579 return ERR_PTR(err); 580} 581 582/* 583 * Send an ICMP message in response to a situation 584 * 585 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header. 586 * MAY send more (we do). 587 * MUST NOT change this header information. 588 * MUST NOT reply to a multicast/broadcast IP address. 589 * MUST NOT reply to a multicast/broadcast MAC address. 590 * MUST reply to only the first fragment. 591 */ 592 593void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info, 594 const struct ip_options *opt) 595{ 596 struct iphdr *iph; 597 int room; 598 struct icmp_bxm icmp_param; 599 struct rtable *rt = skb_rtable(skb_in); 600 struct ipcm_cookie ipc; 601 struct flowi4 fl4; 602 __be32 saddr; 603 u8 tos; 604 u32 mark; 605 struct net *net; 606 struct sock *sk; 607 608 if (!rt) 609 goto out; 610 611 if (rt->dst.dev) 612 net = dev_net(rt->dst.dev); 613 else if (skb_in->dev) 614 net = dev_net(skb_in->dev); 615 else 616 goto out; 617 618 /* 619 * Find the original header. It is expected to be valid, of course. 620 * Check this, icmp_send is called from the most obscure devices 621 * sometimes. 622 */ 623 iph = ip_hdr(skb_in); 624 625 if ((u8 *)iph < skb_in->head || 626 (skb_network_header(skb_in) + sizeof(*iph)) > 627 skb_tail_pointer(skb_in)) 628 goto out; 629 630 /* 631 * No replies to physical multicast/broadcast 632 */ 633 if (skb_in->pkt_type != PACKET_HOST) 634 goto out; 635 636 /* 637 * Now check at the protocol level 638 */ 639 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 640 goto out; 641 642 /* 643 * Only reply to fragment 0. We byte re-order the constant 644 * mask for efficiency. 645 */ 646 if (iph->frag_off & htons(IP_OFFSET)) 647 goto out; 648 649 /* 650 * If we send an ICMP error to an ICMP error a mess would result.. 651 */ 652 if (icmp_pointers[type].error) { 653 /* 654 * We are an error, check if we are replying to an 655 * ICMP error 656 */ 657 if (iph->protocol == IPPROTO_ICMP) { 658 u8 _inner_type, *itp; 659 660 itp = skb_header_pointer(skb_in, 661 skb_network_header(skb_in) + 662 (iph->ihl << 2) + 663 offsetof(struct icmphdr, 664 type) - 665 skb_in->data, 666 sizeof(_inner_type), 667 &_inner_type); 668 if (!itp) 669 goto out; 670 671 /* 672 * Assume any unknown ICMP type is an error. This 673 * isn't specified by the RFC, but think about it.. 674 */ 675 if (*itp > NR_ICMP_TYPES || 676 icmp_pointers[*itp].error) 677 goto out; 678 } 679 } 680 681 /* Needed by both icmp_global_allow and icmp_xmit_lock */ 682 local_bh_disable(); 683 684 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless 685 * incoming dev is loopback. If outgoing dev change to not be 686 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow) 687 */ 688 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) && 689 !icmpv4_global_allow(net, type, code)) 690 goto out_bh_enable; 691 692 sk = icmp_xmit_lock(net); 693 if (!sk) 694 goto out_bh_enable; 695 696 /* 697 * Construct source address and options. 698 */ 699 700 saddr = iph->daddr; 701 if (!(rt->rt_flags & RTCF_LOCAL)) { 702 struct net_device *dev = NULL; 703 704 rcu_read_lock(); 705 if (rt_is_input_route(rt) && 706 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr) 707 dev = dev_get_by_index_rcu(net, inet_iif(skb_in)); 708 709 if (dev) 710 saddr = inet_select_addr(dev, iph->saddr, 711 RT_SCOPE_LINK); 712 else 713 saddr = 0; 714 rcu_read_unlock(); 715 } 716 717 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) | 718 IPTOS_PREC_INTERNETCONTROL) : 719 iph->tos; 720 mark = IP4_REPLY_MARK(net, skb_in->mark); 721 722 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt)) 723 goto out_unlock; 724 725 726 /* 727 * Prepare data for ICMP header. 728 */ 729 730 icmp_param.data.icmph.type = type; 731 icmp_param.data.icmph.code = code; 732 icmp_param.data.icmph.un.gateway = info; 733 icmp_param.data.icmph.checksum = 0; 734 icmp_param.skb = skb_in; 735 icmp_param.offset = skb_network_offset(skb_in); 736 inet_sk(sk)->tos = tos; 737 ipcm_init(&ipc); 738 ipc.addr = iph->saddr; 739 ipc.opt = &icmp_param.replyopts.opt; 740 ipc.sockc.mark = mark; 741 742 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark, 743 type, code, &icmp_param); 744 if (IS_ERR(rt)) 745 goto out_unlock; 746 747 /* peer icmp_ratelimit */ 748 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code)) 749 goto ende; 750 751 /* RFC says return as much as we can without exceeding 576 bytes. */ 752 753 room = dst_mtu(&rt->dst); 754 if (room > 576) 755 room = 576; 756 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen; 757 room -= sizeof(struct icmphdr); 758 /* Guard against tiny mtu. We need to include at least one 759 * IP network header for this message to make any sense. 760 */ 761 if (room <= (int)sizeof(struct iphdr)) 762 goto ende; 763 764 icmp_param.data_len = skb_in->len - icmp_param.offset; 765 if (icmp_param.data_len > room) 766 icmp_param.data_len = room; 767 icmp_param.head_len = sizeof(struct icmphdr); 768 769 /* if we don't have a source address at this point, fall back to the 770 * dummy address instead of sending out a packet with a source address 771 * of 0.0.0.0 772 */ 773 if (!fl4.saddr) 774 fl4.saddr = htonl(INADDR_DUMMY); 775 776 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt); 777ende: 778 ip_rt_put(rt); 779out_unlock: 780 icmp_xmit_unlock(sk); 781out_bh_enable: 782 local_bh_enable(); 783out:; 784} 785EXPORT_SYMBOL(__icmp_send); 786 787#if IS_ENABLED(CONFIG_NF_NAT) 788#include <net/netfilter/nf_conntrack.h> 789void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info) 790{ 791 struct sk_buff *cloned_skb = NULL; 792 struct ip_options opts = { 0 }; 793 enum ip_conntrack_info ctinfo; 794 struct nf_conn *ct; 795 __be32 orig_ip; 796 797 ct = nf_ct_get(skb_in, &ctinfo); 798 if (!ct || !(ct->status & IPS_SRC_NAT)) { 799 __icmp_send(skb_in, type, code, info, &opts); 800 return; 801 } 802 803 if (skb_shared(skb_in)) 804 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC); 805 806 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head || 807 (skb_network_header(skb_in) + sizeof(struct iphdr)) > 808 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in, 809 skb_network_offset(skb_in) + sizeof(struct iphdr)))) 810 goto out; 811 812 orig_ip = ip_hdr(skb_in)->saddr; 813 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip; 814 __icmp_send(skb_in, type, code, info, &opts); 815 ip_hdr(skb_in)->saddr = orig_ip; 816out: 817 consume_skb(cloned_skb); 818} 819EXPORT_SYMBOL(icmp_ndo_send); 820#endif 821 822static void icmp_socket_deliver(struct sk_buff *skb, u32 info) 823{ 824 const struct iphdr *iph = (const struct iphdr *)skb->data; 825 const struct net_protocol *ipprot; 826 int protocol = iph->protocol; 827 828 /* Checkin full IP header plus 8 bytes of protocol to 829 * avoid additional coding at protocol handlers. 830 */ 831 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) { 832 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 833 return; 834 } 835 836 raw_icmp_error(skb, protocol, info); 837 838 ipprot = rcu_dereference(inet_protos[protocol]); 839 if (ipprot && ipprot->err_handler) 840 ipprot->err_handler(skb, info); 841} 842 843static bool icmp_tag_validation(int proto) 844{ 845 bool ok; 846 847 rcu_read_lock(); 848 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation; 849 rcu_read_unlock(); 850 return ok; 851} 852 853/* 854 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and 855 * ICMP_PARAMETERPROB. 856 */ 857 858static bool icmp_unreach(struct sk_buff *skb) 859{ 860 const struct iphdr *iph; 861 struct icmphdr *icmph; 862 struct net *net; 863 u32 info = 0; 864 865 net = dev_net(skb_dst(skb)->dev); 866 867 /* 868 * Incomplete header ? 869 * Only checks for the IP header, there should be an 870 * additional check for longer headers in upper levels. 871 */ 872 873 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 874 goto out_err; 875 876 icmph = icmp_hdr(skb); 877 iph = (const struct iphdr *)skb->data; 878 879 if (iph->ihl < 5) /* Mangled header, drop. */ 880 goto out_err; 881 882 switch (icmph->type) { 883 case ICMP_DEST_UNREACH: 884 switch (icmph->code & 15) { 885 case ICMP_NET_UNREACH: 886 case ICMP_HOST_UNREACH: 887 case ICMP_PROT_UNREACH: 888 case ICMP_PORT_UNREACH: 889 break; 890 case ICMP_FRAG_NEEDED: 891 /* for documentation of the ip_no_pmtu_disc 892 * values please see 893 * Documentation/networking/ip-sysctl.rst 894 */ 895 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) { 896 default: 897 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n", 898 &iph->daddr); 899 break; 900 case 2: 901 goto out; 902 case 3: 903 if (!icmp_tag_validation(iph->protocol)) 904 goto out; 905 fallthrough; 906 case 0: 907 info = ntohs(icmph->un.frag.mtu); 908 } 909 break; 910 case ICMP_SR_FAILED: 911 net_dbg_ratelimited("%pI4: Source Route Failed\n", 912 &iph->daddr); 913 break; 914 default: 915 break; 916 } 917 if (icmph->code > NR_ICMP_UNREACH) 918 goto out; 919 break; 920 case ICMP_PARAMETERPROB: 921 info = ntohl(icmph->un.gateway) >> 24; 922 break; 923 case ICMP_TIME_EXCEEDED: 924 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS); 925 if (icmph->code == ICMP_EXC_FRAGTIME) 926 goto out; 927 break; 928 } 929 930 /* 931 * Throw it at our lower layers 932 * 933 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed 934 * header. 935 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the 936 * transport layer. 937 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to 938 * transport layer. 939 */ 940 941 /* 942 * Check the other end isn't violating RFC 1122. Some routers send 943 * bogus responses to broadcast frames. If you see this message 944 * first check your netmask matches at both ends, if it does then 945 * get the other vendor to fix their kit. 946 */ 947 948 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses && 949 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) { 950 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n", 951 &ip_hdr(skb)->saddr, 952 icmph->type, icmph->code, 953 &iph->daddr, skb->dev->name); 954 goto out; 955 } 956 957 icmp_socket_deliver(skb, info); 958 959out: 960 return true; 961out_err: 962 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 963 return false; 964} 965 966 967/* 968 * Handle ICMP_REDIRECT. 969 */ 970 971static bool icmp_redirect(struct sk_buff *skb) 972{ 973 if (skb->len < sizeof(struct iphdr)) { 974 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS); 975 return false; 976 } 977 978 if (!pskb_may_pull(skb, sizeof(struct iphdr))) { 979 /* there aught to be a stat */ 980 return false; 981 } 982 983 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway)); 984 return true; 985} 986 987/* 988 * Handle ICMP_ECHO ("ping") requests. 989 * 990 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo 991 * requests. 992 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be 993 * included in the reply. 994 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring 995 * echo requests, MUST have default=NOT. 996 * See also WRT handling of options once they are done and working. 997 */ 998 999static bool icmp_echo(struct sk_buff *skb) 1000{ 1001 struct net *net; 1002 1003 net = dev_net(skb_dst(skb)->dev); 1004 if (!net->ipv4.sysctl_icmp_echo_ignore_all) { 1005 struct icmp_bxm icmp_param; 1006 1007 icmp_param.data.icmph = *icmp_hdr(skb); 1008 icmp_param.data.icmph.type = ICMP_ECHOREPLY; 1009 icmp_param.skb = skb; 1010 icmp_param.offset = 0; 1011 icmp_param.data_len = skb->len; 1012 icmp_param.head_len = sizeof(struct icmphdr); 1013 icmp_reply(&icmp_param, skb); 1014 } 1015 /* should there be an ICMP stat for ignored echos? */ 1016 return true; 1017} 1018 1019/* 1020 * Handle ICMP Timestamp requests. 1021 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests. 1022 * SHOULD be in the kernel for minimum random latency. 1023 * MUST be accurate to a few minutes. 1024 * MUST be updated at least at 15Hz. 1025 */ 1026static bool icmp_timestamp(struct sk_buff *skb) 1027{ 1028 struct icmp_bxm icmp_param; 1029 /* 1030 * Too short. 1031 */ 1032 if (skb->len < 4) 1033 goto out_err; 1034 1035 /* 1036 * Fill in the current time as ms since midnight UT: 1037 */ 1038 icmp_param.data.times[1] = inet_current_timestamp(); 1039 icmp_param.data.times[2] = icmp_param.data.times[1]; 1040 1041 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4)); 1042 1043 icmp_param.data.icmph = *icmp_hdr(skb); 1044 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY; 1045 icmp_param.data.icmph.code = 0; 1046 icmp_param.skb = skb; 1047 icmp_param.offset = 0; 1048 icmp_param.data_len = 0; 1049 icmp_param.head_len = sizeof(struct icmphdr) + 12; 1050 icmp_reply(&icmp_param, skb); 1051 return true; 1052 1053out_err: 1054 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS); 1055 return false; 1056} 1057 1058static bool icmp_discard(struct sk_buff *skb) 1059{ 1060 /* pretend it was a success */ 1061 return true; 1062} 1063 1064/* 1065 * Deal with incoming ICMP packets. 1066 */ 1067int icmp_rcv(struct sk_buff *skb) 1068{ 1069 struct icmphdr *icmph; 1070 struct rtable *rt = skb_rtable(skb); 1071 struct net *net = dev_net(rt->dst.dev); 1072 bool success; 1073 1074 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { 1075 struct sec_path *sp = skb_sec_path(skb); 1076 int nh; 1077 1078 if (!(sp && sp->xvec[sp->len - 1]->props.flags & 1079 XFRM_STATE_ICMP)) 1080 goto drop; 1081 1082 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr))) 1083 goto drop; 1084 1085 nh = skb_network_offset(skb); 1086 skb_set_network_header(skb, sizeof(*icmph)); 1087 1088 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb)) 1089 goto drop; 1090 1091 skb_set_network_header(skb, nh); 1092 } 1093 1094 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS); 1095 1096 if (skb_checksum_simple_validate(skb)) 1097 goto csum_error; 1098 1099 if (!pskb_pull(skb, sizeof(*icmph))) 1100 goto error; 1101 1102 icmph = icmp_hdr(skb); 1103 1104 ICMPMSGIN_INC_STATS(net, icmph->type); 1105 /* 1106 * 18 is the highest 'known' ICMP type. Anything else is a mystery 1107 * 1108 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently 1109 * discarded. 1110 */ 1111 if (icmph->type > NR_ICMP_TYPES) 1112 goto error; 1113 1114 1115 /* 1116 * Parse the ICMP message 1117 */ 1118 1119 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) { 1120 /* 1121 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be 1122 * silently ignored (we let user decide with a sysctl). 1123 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently 1124 * discarded if to broadcast/multicast. 1125 */ 1126 if ((icmph->type == ICMP_ECHO || 1127 icmph->type == ICMP_TIMESTAMP) && 1128 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) { 1129 goto error; 1130 } 1131 if (icmph->type != ICMP_ECHO && 1132 icmph->type != ICMP_TIMESTAMP && 1133 icmph->type != ICMP_ADDRESS && 1134 icmph->type != ICMP_ADDRESSREPLY) { 1135 goto error; 1136 } 1137 } 1138 1139 success = icmp_pointers[icmph->type].handler(skb); 1140 1141 if (success) { 1142 consume_skb(skb); 1143 return NET_RX_SUCCESS; 1144 } 1145 1146drop: 1147 kfree_skb(skb); 1148 return NET_RX_DROP; 1149csum_error: 1150 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS); 1151error: 1152 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); 1153 goto drop; 1154} 1155 1156static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off) 1157{ 1158 struct icmp_extobj_hdr *objh, _objh; 1159 struct icmp_ext_hdr *exth, _exth; 1160 u16 olen; 1161 1162 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth); 1163 if (!exth) 1164 return false; 1165 if (exth->version != 2) 1166 return true; 1167 1168 if (exth->checksum && 1169 csum_fold(skb_checksum(skb, off, skb->len - off, 0))) 1170 return false; 1171 1172 off += sizeof(_exth); 1173 while (off < skb->len) { 1174 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh); 1175 if (!objh) 1176 return false; 1177 1178 olen = ntohs(objh->length); 1179 if (olen < sizeof(_objh)) 1180 return false; 1181 1182 off += olen; 1183 if (off > skb->len) 1184 return false; 1185 } 1186 1187 return true; 1188} 1189 1190void ip_icmp_error_rfc4884(const struct sk_buff *skb, 1191 struct sock_ee_data_rfc4884 *out, 1192 int thlen, int off) 1193{ 1194 int hlen; 1195 1196 /* original datagram headers: end of icmph to payload (skb->data) */ 1197 hlen = -skb_transport_offset(skb) - thlen; 1198 1199 /* per rfc 4884: minimal datagram length of 128 bytes */ 1200 if (off < 128 || off < hlen) 1201 return; 1202 1203 /* kernel has stripped headers: return payload offset in bytes */ 1204 off -= hlen; 1205 if (off + sizeof(struct icmp_ext_hdr) > skb->len) 1206 return; 1207 1208 out->len = off; 1209 1210 if (!ip_icmp_error_rfc4884_validate(skb, off)) 1211 out->flags |= SO_EE_RFC4884_FLAG_INVALID; 1212} 1213EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884); 1214 1215int icmp_err(struct sk_buff *skb, u32 info) 1216{ 1217 struct iphdr *iph = (struct iphdr *)skb->data; 1218 int offset = iph->ihl<<2; 1219 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset); 1220 int type = icmp_hdr(skb)->type; 1221 int code = icmp_hdr(skb)->code; 1222 struct net *net = dev_net(skb->dev); 1223 1224 /* 1225 * Use ping_err to handle all icmp errors except those 1226 * triggered by ICMP_ECHOREPLY which sent from kernel. 1227 */ 1228 if (icmph->type != ICMP_ECHOREPLY) { 1229 ping_err(skb, offset, info); 1230 return 0; 1231 } 1232 1233 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) 1234 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP); 1235 else if (type == ICMP_REDIRECT) 1236 ipv4_redirect(skb, net, 0, IPPROTO_ICMP); 1237 1238 return 0; 1239} 1240 1241/* 1242 * This table is the definition of how we handle ICMP. 1243 */ 1244static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = { 1245 [ICMP_ECHOREPLY] = { 1246 .handler = ping_rcv, 1247 }, 1248 [1] = { 1249 .handler = icmp_discard, 1250 .error = 1, 1251 }, 1252 [2] = { 1253 .handler = icmp_discard, 1254 .error = 1, 1255 }, 1256 [ICMP_DEST_UNREACH] = { 1257 .handler = icmp_unreach, 1258 .error = 1, 1259 }, 1260 [ICMP_SOURCE_QUENCH] = { 1261 .handler = icmp_unreach, 1262 .error = 1, 1263 }, 1264 [ICMP_REDIRECT] = { 1265 .handler = icmp_redirect, 1266 .error = 1, 1267 }, 1268 [6] = { 1269 .handler = icmp_discard, 1270 .error = 1, 1271 }, 1272 [7] = { 1273 .handler = icmp_discard, 1274 .error = 1, 1275 }, 1276 [ICMP_ECHO] = { 1277 .handler = icmp_echo, 1278 }, 1279 [9] = { 1280 .handler = icmp_discard, 1281 .error = 1, 1282 }, 1283 [10] = { 1284 .handler = icmp_discard, 1285 .error = 1, 1286 }, 1287 [ICMP_TIME_EXCEEDED] = { 1288 .handler = icmp_unreach, 1289 .error = 1, 1290 }, 1291 [ICMP_PARAMETERPROB] = { 1292 .handler = icmp_unreach, 1293 .error = 1, 1294 }, 1295 [ICMP_TIMESTAMP] = { 1296 .handler = icmp_timestamp, 1297 }, 1298 [ICMP_TIMESTAMPREPLY] = { 1299 .handler = icmp_discard, 1300 }, 1301 [ICMP_INFO_REQUEST] = { 1302 .handler = icmp_discard, 1303 }, 1304 [ICMP_INFO_REPLY] = { 1305 .handler = icmp_discard, 1306 }, 1307 [ICMP_ADDRESS] = { 1308 .handler = icmp_discard, 1309 }, 1310 [ICMP_ADDRESSREPLY] = { 1311 .handler = icmp_discard, 1312 }, 1313}; 1314 1315static void __net_exit icmp_sk_exit(struct net *net) 1316{ 1317 int i; 1318 1319 for_each_possible_cpu(i) 1320 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i)); 1321 free_percpu(net->ipv4.icmp_sk); 1322 net->ipv4.icmp_sk = NULL; 1323} 1324 1325static int __net_init icmp_sk_init(struct net *net) 1326{ 1327 int i, err; 1328 1329 net->ipv4.icmp_sk = alloc_percpu(struct sock *); 1330 if (!net->ipv4.icmp_sk) 1331 return -ENOMEM; 1332 1333 for_each_possible_cpu(i) { 1334 struct sock *sk; 1335 1336 err = inet_ctl_sock_create(&sk, PF_INET, 1337 SOCK_RAW, IPPROTO_ICMP, net); 1338 if (err < 0) 1339 goto fail; 1340 1341 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk; 1342 1343 /* Enough space for 2 64K ICMP packets, including 1344 * sk_buff/skb_shared_info struct overhead. 1345 */ 1346 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024); 1347 1348 /* 1349 * Speedup sock_wfree() 1350 */ 1351 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); 1352 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT; 1353 } 1354 1355 /* Control parameters for ECHO replies. */ 1356 net->ipv4.sysctl_icmp_echo_ignore_all = 0; 1357 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1; 1358 1359 /* Control parameter - ignore bogus broadcast responses? */ 1360 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1; 1361 1362 /* 1363 * Configurable global rate limit. 1364 * 1365 * ratelimit defines tokens/packet consumed for dst->rate_token 1366 * bucket ratemask defines which icmp types are ratelimited by 1367 * setting it's bit position. 1368 * 1369 * default: 1370 * dest unreachable (3), source quench (4), 1371 * time exceeded (11), parameter problem (12) 1372 */ 1373 1374 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ; 1375 net->ipv4.sysctl_icmp_ratemask = 0x1818; 1376 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0; 1377 1378 return 0; 1379 1380fail: 1381 icmp_sk_exit(net); 1382 return err; 1383} 1384 1385static struct pernet_operations __net_initdata icmp_sk_ops = { 1386 .init = icmp_sk_init, 1387 .exit = icmp_sk_exit, 1388}; 1389 1390int __init icmp_init(void) 1391{ 1392 return register_pernet_subsys(&icmp_sk_ops); 1393} 1394