1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3: Implementation of the ICMP protocol layer.
5 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
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.
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
20 * Alan Cox : Added 216,128 byte paths to the MTU
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
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
38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
40 * Andi Kleen : Check all packet lengths properly
41 * and moved all kfree_skb() up to
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
51 * Tristan Greaves : Added sysctl option to ignore bogus
52 * broadcast responses from broken routers.
56 * - Should use skb_pull() instead of all the manual checking.
57 * This would also greatly simply some upper layer error handlers. --AK
60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
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>
77 #include <net/route.h>
78 #include <net/protocol.h>
84 #include <linux/skbuff.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>
92 #include <net/inet_common.h>
93 #include <net/ip_fib.h>
94 #include <net/l3mdev.h>
95 #include <net/addrconf.h>
98 * Build xmit assembly blocks
107 struct icmphdr icmph;
111 struct ip_options_data replyopts;
114 /* An array of errno for error messages from dest unreach. */
115 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
117 const struct icmp_err icmp_err_convert[] = {
119 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
123 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
127 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
131 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
135 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
139 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
143 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
147 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
151 .errno = ENONET, /* ICMP_HOST_ISOLATED */
155 .errno = ENETUNREACH, /* ICMP_NET_ANO */
159 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
163 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
167 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
171 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
175 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
179 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
183 EXPORT_SYMBOL(icmp_err_convert);
186 * ICMP control array. This specifies what to do with each ICMP.
189 struct icmp_control {
190 enum skb_drop_reason (*handler)(struct sk_buff *skb);
191 short error; /* This ICMP is classed as an error message */
194 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
196 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
198 /* Called with BH disabled */
199 static inline struct sock *icmp_xmit_lock(struct net *net)
203 sk = this_cpu_read(ipv4_icmp_sk);
205 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
206 /* This can happen if the output path signals a
207 * dst_link_failure() for an outgoing ICMP packet.
211 sock_net_set(sk, net);
215 static inline void icmp_xmit_unlock(struct sock *sk)
217 sock_net_set(sk, &init_net);
218 spin_unlock(&sk->sk_lock.slock);
221 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
222 int sysctl_icmp_msgs_burst __read_mostly = 50;
229 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
233 * icmp_global_allow - Are we allowed to send one more ICMP message ?
235 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
236 * Returns false if we reached the limit and can not send another packet.
237 * Note: called with BH disabled
239 bool icmp_global_allow(void)
241 u32 credit, delta, incr = 0, now = (u32)jiffies;
244 /* Check if token bucket is empty and cannot be refilled
245 * without taking the spinlock. The READ_ONCE() are paired
246 * with the following WRITE_ONCE() in this same function.
248 if (!READ_ONCE(icmp_global.credit)) {
249 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
254 spin_lock(&icmp_global.lock);
255 delta = min_t(u32, now - icmp_global.stamp, HZ);
256 if (delta >= HZ / 50) {
257 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
259 WRITE_ONCE(icmp_global.stamp, now);
261 credit = min_t(u32, icmp_global.credit + incr,
262 READ_ONCE(sysctl_icmp_msgs_burst));
264 /* We want to use a credit of one in average, but need to randomize
265 * it for security reasons.
267 credit = max_t(int, credit - get_random_u32_below(3), 0);
270 WRITE_ONCE(icmp_global.credit, credit);
271 spin_unlock(&icmp_global.lock);
274 EXPORT_SYMBOL(icmp_global_allow);
276 static bool icmpv4_mask_allow(struct net *net, int type, int code)
278 if (type > NR_ICMP_TYPES)
281 /* Don't limit PMTU discovery. */
282 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 /* Limit if icmp type is enabled in ratemask. */
286 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
292 static bool icmpv4_global_allow(struct net *net, int type, int code)
294 if (icmpv4_mask_allow(net, type, code))
297 if (icmp_global_allow())
300 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
305 * Send an ICMP frame.
308 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
309 struct flowi4 *fl4, int type, int code)
311 struct dst_entry *dst = &rt->dst;
312 struct inet_peer *peer;
316 if (icmpv4_mask_allow(net, type, code))
319 /* No rate limit on loopback */
320 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
323 vif = l3mdev_master_ifindex(dst->dev);
324 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
325 rc = inet_peer_xrlim_allow(peer,
326 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
331 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
336 * Maintain the counters used in the SNMP statistics for outgoing ICMP
338 void icmp_out_count(struct net *net, unsigned char type)
340 ICMPMSGOUT_INC_STATS(net, type);
341 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
345 * Checksum each fragment, and on the first include the headers and final
348 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
351 struct icmp_bxm *icmp_param = from;
354 csum = skb_copy_and_csum_bits(icmp_param->skb,
355 icmp_param->offset + offset,
358 skb->csum = csum_block_add(skb->csum, csum, odd);
359 if (icmp_pointers[icmp_param->data.icmph.type].error)
360 nf_ct_attach(skb, icmp_param->skb);
364 static void icmp_push_reply(struct sock *sk,
365 struct icmp_bxm *icmp_param,
367 struct ipcm_cookie *ipc, struct rtable **rt)
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);
380 struct sk_buff *skb1;
382 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
384 icmp_param->head_len);
385 skb_queue_walk(&sk->sk_write_queue, skb1) {
386 csum = csum_add(csum, skb1->csum);
388 icmph->checksum = csum_fold(csum);
389 skb->ip_summed = CHECKSUM_NONE;
390 ip_push_pending_frames(sk, fl4);
395 * Driving logic for building and sending ICMP messages.
398 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
400 struct ipcm_cookie ipc;
401 struct rtable *rt = skb_rtable(skb);
402 struct net *net = dev_net(rt->dst.dev);
405 struct inet_sock *inet;
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;
411 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
414 /* Needed by both icmp_global_allow and icmp_xmit_lock */
417 /* global icmp_msgs_per_sec */
418 if (!icmpv4_global_allow(net, type, code))
421 sk = icmp_xmit_lock(net);
426 icmp_param->data.icmph.checksum = 0;
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);
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;
439 memset(&fl4, 0, sizeof(fl4));
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_common(&fl4));
448 rt = ip_route_output_key(net, &fl4);
451 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
452 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
455 icmp_xmit_unlock(sk);
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).
466 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
468 struct net_device *route_lookup_dev = NULL;
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;
477 static struct rtable *icmp_route_lookup(struct net *net,
479 struct sk_buff *skb_in,
480 const struct iphdr *iph,
481 __be32 saddr, u8 tos, u32 mark,
483 struct icmp_bxm *param)
485 struct net_device *route_lookup_dev;
486 struct rtable *rt, *rt2;
487 struct flowi4 fl4_dec;
490 memset(fl4, 0, sizeof(*fl4));
491 fl4->daddr = (param->replyopts.opt.opt.srr ?
492 param->replyopts.opt.opt.faddr : iph->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);
503 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
504 rt = ip_route_output_key_hash(net, fl4, skb_in);
508 /* No need to clone since we're just using its address. */
511 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
512 flowi4_to_flowi(fl4), NULL, 0);
516 } else if (PTR_ERR(rt) == -EPERM) {
521 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
523 goto relookup_failed;
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);
531 struct flowi4 fl4_2 = {};
532 unsigned long orefdst;
534 fl4_2.daddr = fl4_dec.saddr;
535 rt2 = ip_route_output_key(net, &fl4_2);
538 goto relookup_failed;
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);
546 dst_release(&rt2->dst);
547 rt2 = skb_rtable(skb_in);
548 skb_in->_skb_refdst = orefdst; /* restore old refdst */
552 goto relookup_failed;
554 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
555 flowi4_to_flowi(&fl4_dec), NULL,
558 dst_release(&rt->dst);
559 memcpy(fl4, &fl4_dec, sizeof(*fl4));
561 } else if (PTR_ERR(rt2) == -EPERM) {
563 dst_release(&rt->dst);
567 goto relookup_failed;
578 * Send an ICMP message in response to a situation
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.
588 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
589 const struct ip_options *opt)
593 struct icmp_bxm icmp_param;
594 struct rtable *rt = skb_rtable(skb_in);
595 struct ipcm_cookie ipc;
607 net = dev_net(rt->dst.dev);
608 else if (skb_in->dev)
609 net = dev_net(skb_in->dev);
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
618 iph = ip_hdr(skb_in);
620 if ((u8 *)iph < skb_in->head ||
621 (skb_network_header(skb_in) + sizeof(*iph)) >
622 skb_tail_pointer(skb_in))
626 * No replies to physical multicast/broadcast
628 if (skb_in->pkt_type != PACKET_HOST)
632 * Now check at the protocol level
634 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
638 * Only reply to fragment 0. We byte re-order the constant
639 * mask for efficiency.
641 if (iph->frag_off & htons(IP_OFFSET))
645 * If we send an ICMP error to an ICMP error a mess would result..
647 if (icmp_pointers[type].error) {
649 * We are an error, check if we are replying to an
652 if (iph->protocol == IPPROTO_ICMP) {
653 u8 _inner_type, *itp;
655 itp = skb_header_pointer(skb_in,
656 skb_network_header(skb_in) +
658 offsetof(struct icmphdr,
667 * Assume any unknown ICMP type is an error. This
668 * isn't specified by the RFC, but think about it..
670 if (*itp > NR_ICMP_TYPES ||
671 icmp_pointers[*itp].error)
676 /* Needed by both icmp_global_allow and icmp_xmit_lock */
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)
683 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
684 !icmpv4_global_allow(net, type, code))
687 sk = icmp_xmit_lock(net);
692 * Construct source address and options.
696 if (!(rt->rt_flags & RTCF_LOCAL)) {
697 struct net_device *dev = NULL;
700 if (rt_is_input_route(rt) &&
701 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
702 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
705 saddr = inet_select_addr(dev, iph->saddr,
712 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
713 IPTOS_PREC_INTERNETCONTROL) :
715 mark = IP4_REPLY_MARK(net, skb_in->mark);
717 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
722 * Prepare data for ICMP header.
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;
733 ipc.addr = iph->saddr;
734 ipc.opt = &icmp_param.replyopts.opt;
735 ipc.sockc.mark = mark;
737 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
738 type, code, &icmp_param);
742 /* peer icmp_ratelimit */
743 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
746 /* RFC says return as much as we can without exceeding 576 bytes. */
748 room = dst_mtu(&rt->dst);
751 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
752 room -= sizeof(struct icmphdr);
753 /* Guard against tiny mtu. We need to include at least one
754 * IP network header for this message to make any sense.
756 if (room <= (int)sizeof(struct iphdr))
759 icmp_param.data_len = skb_in->len - icmp_param.offset;
760 if (icmp_param.data_len > room)
761 icmp_param.data_len = room;
762 icmp_param.head_len = sizeof(struct icmphdr);
764 /* if we don't have a source address at this point, fall back to the
765 * dummy address instead of sending out a packet with a source address
769 fl4.saddr = htonl(INADDR_DUMMY);
771 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
775 icmp_xmit_unlock(sk);
780 EXPORT_SYMBOL(__icmp_send);
782 #if IS_ENABLED(CONFIG_NF_NAT)
783 #include <net/netfilter/nf_conntrack.h>
784 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
786 struct sk_buff *cloned_skb = NULL;
787 struct ip_options opts = { 0 };
788 enum ip_conntrack_info ctinfo;
792 ct = nf_ct_get(skb_in, &ctinfo);
793 if (!ct || !(ct->status & IPS_SRC_NAT)) {
794 __icmp_send(skb_in, type, code, info, &opts);
798 if (skb_shared(skb_in))
799 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
801 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
802 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
803 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
804 skb_network_offset(skb_in) + sizeof(struct iphdr))))
807 orig_ip = ip_hdr(skb_in)->saddr;
808 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
809 __icmp_send(skb_in, type, code, info, &opts);
810 ip_hdr(skb_in)->saddr = orig_ip;
812 consume_skb(cloned_skb);
814 EXPORT_SYMBOL(icmp_ndo_send);
817 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
819 const struct iphdr *iph = (const struct iphdr *)skb->data;
820 const struct net_protocol *ipprot;
821 int protocol = iph->protocol;
823 /* Checkin full IP header plus 8 bytes of protocol to
824 * avoid additional coding at protocol handlers.
826 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
827 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
831 raw_icmp_error(skb, protocol, info);
833 ipprot = rcu_dereference(inet_protos[protocol]);
834 if (ipprot && ipprot->err_handler)
835 ipprot->err_handler(skb, info);
838 static bool icmp_tag_validation(int proto)
843 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
849 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
850 * ICMP_PARAMETERPROB.
853 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
855 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
856 const struct iphdr *iph;
857 struct icmphdr *icmph;
861 net = dev_net(skb_dst(skb)->dev);
864 * Incomplete header ?
865 * Only checks for the IP header, there should be an
866 * additional check for longer headers in upper levels.
869 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
872 icmph = icmp_hdr(skb);
873 iph = (const struct iphdr *)skb->data;
875 if (iph->ihl < 5) { /* Mangled header, drop. */
876 reason = SKB_DROP_REASON_IP_INHDR;
880 switch (icmph->type) {
881 case ICMP_DEST_UNREACH:
882 switch (icmph->code & 15) {
883 case ICMP_NET_UNREACH:
884 case ICMP_HOST_UNREACH:
885 case ICMP_PROT_UNREACH:
886 case ICMP_PORT_UNREACH:
888 case ICMP_FRAG_NEEDED:
889 /* for documentation of the ip_no_pmtu_disc
891 * Documentation/networking/ip-sysctl.rst
893 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
895 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
901 if (!icmp_tag_validation(iph->protocol))
905 info = ntohs(icmph->un.frag.mtu);
909 net_dbg_ratelimited("%pI4: Source Route Failed\n",
915 if (icmph->code > NR_ICMP_UNREACH)
918 case ICMP_PARAMETERPROB:
919 info = ntohl(icmph->un.gateway) >> 24;
921 case ICMP_TIME_EXCEEDED:
922 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
923 if (icmph->code == ICMP_EXC_FRAGTIME)
929 * Throw it at our lower layers
931 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
933 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
935 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
940 * Check the other end isn't violating RFC 1122. Some routers send
941 * bogus responses to broadcast frames. If you see this message
942 * first check your netmask matches at both ends, if it does then
943 * get the other vendor to fix their kit.
946 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
947 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
948 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
950 icmph->type, icmph->code,
951 &iph->daddr, skb->dev->name);
955 icmp_socket_deliver(skb, info);
960 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
961 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
966 * Handle ICMP_REDIRECT.
969 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
971 if (skb->len < sizeof(struct iphdr)) {
972 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
973 return SKB_DROP_REASON_PKT_TOO_SMALL;
976 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
977 /* there aught to be a stat */
978 return SKB_DROP_REASON_NOMEM;
981 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
982 return SKB_NOT_DROPPED_YET;
986 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
988 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
990 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
991 * included in the reply.
992 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
993 * echo requests, MUST have default=NOT.
994 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
995 * Extended Echo Functionality, MUST be disabled by default
996 * See also WRT handling of options once they are done and working.
999 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1001 struct icmp_bxm icmp_param;
1004 net = dev_net(skb_dst(skb)->dev);
1005 /* should there be an ICMP stat for ignored echos? */
1006 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1007 return SKB_NOT_DROPPED_YET;
1009 icmp_param.data.icmph = *icmp_hdr(skb);
1010 icmp_param.skb = skb;
1011 icmp_param.offset = 0;
1012 icmp_param.data_len = skb->len;
1013 icmp_param.head_len = sizeof(struct icmphdr);
1015 if (icmp_param.data.icmph.type == ICMP_ECHO)
1016 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1017 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1018 return SKB_NOT_DROPPED_YET;
1020 icmp_reply(&icmp_param, skb);
1021 return SKB_NOT_DROPPED_YET;
1024 /* Helper for icmp_echo and icmpv6_echo_reply.
1025 * Searches for net_device that matches PROBE interface identifier
1026 * and builds PROBE reply message in icmphdr.
1028 * Returns false if PROBE responses are disabled via sysctl
1031 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1033 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1034 struct icmp_ext_echo_iio *iio, _iio;
1035 struct net *net = dev_net(skb->dev);
1036 struct inet6_dev *in6_dev;
1037 struct in_device *in_dev;
1038 struct net_device *dev;
1039 char buff[IFNAMSIZ];
1043 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1046 /* We currently only support probing interfaces on the proxy node
1047 * Check to ensure L-bit is set
1049 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1051 /* Clear status bits in reply message */
1052 icmphdr->un.echo.sequence &= htons(0xFF00);
1053 if (icmphdr->type == ICMP_EXT_ECHO)
1054 icmphdr->type = ICMP_EXT_ECHOREPLY;
1056 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1057 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1058 /* Size of iio is class_type dependent.
1059 * Only check header here and assign length based on ctype in the switch statement
1061 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1062 if (!ext_hdr || !iio)
1063 goto send_mal_query;
1064 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1065 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1066 goto send_mal_query;
1067 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1068 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1069 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1071 goto send_mal_query;
1075 switch (iio->extobj_hdr.class_type) {
1076 case ICMP_EXT_ECHO_CTYPE_NAME:
1077 if (ident_len >= IFNAMSIZ)
1078 goto send_mal_query;
1079 memset(buff, 0, sizeof(buff));
1080 memcpy(buff, &iio->ident.name, ident_len);
1081 dev = dev_get_by_name(net, buff);
1083 case ICMP_EXT_ECHO_CTYPE_INDEX:
1084 if (ident_len != sizeof(iio->ident.ifindex))
1085 goto send_mal_query;
1086 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1088 case ICMP_EXT_ECHO_CTYPE_ADDR:
1089 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1090 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1091 iio->ident.addr.ctype3_hdr.addrlen)
1092 goto send_mal_query;
1093 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1095 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1096 goto send_mal_query;
1097 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1099 #if IS_ENABLED(CONFIG_IPV6)
1101 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1102 goto send_mal_query;
1103 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1108 goto send_mal_query;
1112 goto send_mal_query;
1115 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1118 /* Fill bits in reply message */
1119 if (dev->flags & IFF_UP)
1120 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1122 in_dev = __in_dev_get_rcu(dev);
1123 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1124 status |= ICMP_EXT_ECHOREPLY_IPV4;
1126 in6_dev = __in6_dev_get(dev);
1127 if (in6_dev && !list_empty(&in6_dev->addr_list))
1128 status |= ICMP_EXT_ECHOREPLY_IPV6;
1131 icmphdr->un.echo.sequence |= htons(status);
1134 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1137 EXPORT_SYMBOL_GPL(icmp_build_probe);
1140 * Handle ICMP Timestamp requests.
1141 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1142 * SHOULD be in the kernel for minimum random latency.
1143 * MUST be accurate to a few minutes.
1144 * MUST be updated at least at 15Hz.
1146 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1148 struct icmp_bxm icmp_param;
1156 * Fill in the current time as ms since midnight UT:
1158 icmp_param.data.times[1] = inet_current_timestamp();
1159 icmp_param.data.times[2] = icmp_param.data.times[1];
1161 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1163 icmp_param.data.icmph = *icmp_hdr(skb);
1164 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1165 icmp_param.data.icmph.code = 0;
1166 icmp_param.skb = skb;
1167 icmp_param.offset = 0;
1168 icmp_param.data_len = 0;
1169 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1170 icmp_reply(&icmp_param, skb);
1171 return SKB_NOT_DROPPED_YET;
1174 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1175 return SKB_DROP_REASON_PKT_TOO_SMALL;
1178 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1180 /* pretend it was a success */
1181 return SKB_NOT_DROPPED_YET;
1185 * Deal with incoming ICMP packets.
1187 int icmp_rcv(struct sk_buff *skb)
1189 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1190 struct rtable *rt = skb_rtable(skb);
1191 struct net *net = dev_net(rt->dst.dev);
1192 struct icmphdr *icmph;
1194 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1195 struct sec_path *sp = skb_sec_path(skb);
1198 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1200 reason = SKB_DROP_REASON_XFRM_POLICY;
1204 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1207 nh = skb_network_offset(skb);
1208 skb_set_network_header(skb, sizeof(*icmph));
1210 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1212 reason = SKB_DROP_REASON_XFRM_POLICY;
1216 skb_set_network_header(skb, nh);
1219 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1221 if (skb_checksum_simple_validate(skb))
1224 if (!pskb_pull(skb, sizeof(*icmph)))
1227 icmph = icmp_hdr(skb);
1229 ICMPMSGIN_INC_STATS(net, icmph->type);
1231 /* Check for ICMP Extended Echo (PROBE) messages */
1232 if (icmph->type == ICMP_EXT_ECHO) {
1233 /* We can't use icmp_pointers[].handler() because it is an array of
1234 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1236 reason = icmp_echo(skb);
1240 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1241 reason = ping_rcv(skb);
1246 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1248 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1251 if (icmph->type > NR_ICMP_TYPES) {
1252 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1257 * Parse the ICMP message
1260 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1262 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1263 * silently ignored (we let user decide with a sysctl).
1264 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1265 * discarded if to broadcast/multicast.
1267 if ((icmph->type == ICMP_ECHO ||
1268 icmph->type == ICMP_TIMESTAMP) &&
1269 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1270 reason = SKB_DROP_REASON_INVALID_PROTO;
1273 if (icmph->type != ICMP_ECHO &&
1274 icmph->type != ICMP_TIMESTAMP &&
1275 icmph->type != ICMP_ADDRESS &&
1276 icmph->type != ICMP_ADDRESSREPLY) {
1277 reason = SKB_DROP_REASON_INVALID_PROTO;
1282 reason = icmp_pointers[icmph->type].handler(skb);
1286 return NET_RX_SUCCESS;
1290 kfree_skb_reason(skb, reason);
1293 reason = SKB_DROP_REASON_ICMP_CSUM;
1294 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1296 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1300 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1302 struct icmp_extobj_hdr *objh, _objh;
1303 struct icmp_ext_hdr *exth, _exth;
1306 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1309 if (exth->version != 2)
1312 if (exth->checksum &&
1313 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1316 off += sizeof(_exth);
1317 while (off < skb->len) {
1318 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1322 olen = ntohs(objh->length);
1323 if (olen < sizeof(_objh))
1334 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1335 struct sock_ee_data_rfc4884 *out,
1340 /* original datagram headers: end of icmph to payload (skb->data) */
1341 hlen = -skb_transport_offset(skb) - thlen;
1343 /* per rfc 4884: minimal datagram length of 128 bytes */
1344 if (off < 128 || off < hlen)
1347 /* kernel has stripped headers: return payload offset in bytes */
1349 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1354 if (!ip_icmp_error_rfc4884_validate(skb, off))
1355 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1357 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1359 int icmp_err(struct sk_buff *skb, u32 info)
1361 struct iphdr *iph = (struct iphdr *)skb->data;
1362 int offset = iph->ihl<<2;
1363 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1364 int type = icmp_hdr(skb)->type;
1365 int code = icmp_hdr(skb)->code;
1366 struct net *net = dev_net(skb->dev);
1369 * Use ping_err to handle all icmp errors except those
1370 * triggered by ICMP_ECHOREPLY which sent from kernel.
1372 if (icmph->type != ICMP_ECHOREPLY) {
1373 ping_err(skb, offset, info);
1377 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1378 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1379 else if (type == ICMP_REDIRECT)
1380 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1386 * This table is the definition of how we handle ICMP.
1388 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1389 [ICMP_ECHOREPLY] = {
1390 .handler = ping_rcv,
1393 .handler = icmp_discard,
1397 .handler = icmp_discard,
1400 [ICMP_DEST_UNREACH] = {
1401 .handler = icmp_unreach,
1404 [ICMP_SOURCE_QUENCH] = {
1405 .handler = icmp_unreach,
1409 .handler = icmp_redirect,
1413 .handler = icmp_discard,
1417 .handler = icmp_discard,
1421 .handler = icmp_echo,
1424 .handler = icmp_discard,
1428 .handler = icmp_discard,
1431 [ICMP_TIME_EXCEEDED] = {
1432 .handler = icmp_unreach,
1435 [ICMP_PARAMETERPROB] = {
1436 .handler = icmp_unreach,
1439 [ICMP_TIMESTAMP] = {
1440 .handler = icmp_timestamp,
1442 [ICMP_TIMESTAMPREPLY] = {
1443 .handler = icmp_discard,
1445 [ICMP_INFO_REQUEST] = {
1446 .handler = icmp_discard,
1448 [ICMP_INFO_REPLY] = {
1449 .handler = icmp_discard,
1452 .handler = icmp_discard,
1454 [ICMP_ADDRESSREPLY] = {
1455 .handler = icmp_discard,
1459 static int __net_init icmp_sk_init(struct net *net)
1461 /* Control parameters for ECHO replies. */
1462 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1463 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1464 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1466 /* Control parameter - ignore bogus broadcast responses? */
1467 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1470 * Configurable global rate limit.
1472 * ratelimit defines tokens/packet consumed for dst->rate_token
1473 * bucket ratemask defines which icmp types are ratelimited by
1474 * setting it's bit position.
1477 * dest unreachable (3), source quench (4),
1478 * time exceeded (11), parameter problem (12)
1481 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1482 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1483 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1488 static struct pernet_operations __net_initdata icmp_sk_ops = {
1489 .init = icmp_sk_init,
1492 int __init icmp_init(void)
1496 for_each_possible_cpu(i) {
1499 err = inet_ctl_sock_create(&sk, PF_INET,
1500 SOCK_RAW, IPPROTO_ICMP, &init_net);
1504 per_cpu(ipv4_icmp_sk, i) = sk;
1506 /* Enough space for 2 64K ICMP packets, including
1507 * sk_buff/skb_shared_info struct overhead.
1509 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1512 * Speedup sock_wfree()
1514 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1515 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1517 return register_pernet_subsys(&icmp_sk_ops);