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>
97 * Build xmit assembly blocks
106 struct icmphdr icmph;
110 struct ip_options_data replyopts;
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'. */
116 const struct icmp_err icmp_err_convert[] = {
118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
150 .errno = ENONET, /* ICMP_HOST_ISOLATED */
154 .errno = ENETUNREACH, /* ICMP_NET_ANO */
158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
182 EXPORT_SYMBOL(icmp_err_convert);
185 * ICMP control array. This specifies what to do with each ICMP.
188 struct icmp_control {
189 enum skb_drop_reason (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
195 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
197 /* Called with BH disabled */
198 static inline struct sock *icmp_xmit_lock(struct net *net)
202 sk = this_cpu_read(ipv4_icmp_sk);
204 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
205 /* This can happen if the output path signals a
206 * dst_link_failure() for an outgoing ICMP packet.
210 sock_net_set(sk, net);
214 static inline void icmp_xmit_unlock(struct sock *sk)
216 sock_net_set(sk, &init_net);
217 spin_unlock(&sk->sk_lock.slock);
220 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
221 int sysctl_icmp_msgs_burst __read_mostly = 50;
228 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
232 * icmp_global_allow - Are we allowed to send one more ICMP message ?
234 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
235 * Returns false if we reached the limit and can not send another packet.
236 * Note: called with BH disabled
238 bool icmp_global_allow(void)
240 u32 credit, delta, incr = 0, now = (u32)jiffies;
243 /* Check if token bucket is empty and cannot be refilled
244 * without taking the spinlock. The READ_ONCE() are paired
245 * with the following WRITE_ONCE() in this same function.
247 if (!READ_ONCE(icmp_global.credit)) {
248 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
253 spin_lock(&icmp_global.lock);
254 delta = min_t(u32, now - icmp_global.stamp, HZ);
255 if (delta >= HZ / 50) {
256 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
258 WRITE_ONCE(icmp_global.stamp, now);
260 credit = min_t(u32, icmp_global.credit + incr,
261 READ_ONCE(sysctl_icmp_msgs_burst));
263 /* We want to use a credit of one in average, but need to randomize
264 * it for security reasons.
266 credit = max_t(int, credit - get_random_u32_below(3), 0);
269 WRITE_ONCE(icmp_global.credit, credit);
270 spin_unlock(&icmp_global.lock);
273 EXPORT_SYMBOL(icmp_global_allow);
275 static bool icmpv4_mask_allow(struct net *net, int type, int code)
277 if (type > NR_ICMP_TYPES)
280 /* Don't limit PMTU discovery. */
281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
284 /* Limit if icmp type is enabled in ratemask. */
285 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
291 static bool icmpv4_global_allow(struct net *net, int type, int code)
293 if (icmpv4_mask_allow(net, type, code))
296 if (icmp_global_allow())
299 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
304 * Send an ICMP frame.
307 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
308 struct flowi4 *fl4, int type, int code)
310 struct dst_entry *dst = &rt->dst;
311 struct inet_peer *peer;
315 if (icmpv4_mask_allow(net, type, code))
318 /* No rate limit on loopback */
319 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
322 vif = l3mdev_master_ifindex(dst->dev);
323 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
324 rc = inet_peer_xrlim_allow(peer,
325 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
330 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
335 * Maintain the counters used in the SNMP statistics for outgoing ICMP
337 void icmp_out_count(struct net *net, unsigned char type)
339 ICMPMSGOUT_INC_STATS(net, type);
340 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
344 * Checksum each fragment, and on the first include the headers and final
347 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
350 struct icmp_bxm *icmp_param = from;
353 csum = skb_copy_and_csum_bits(icmp_param->skb,
354 icmp_param->offset + offset,
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);
363 static void icmp_push_reply(struct sock *sk,
364 struct icmp_bxm *icmp_param,
366 struct ipcm_cookie *ipc, struct rtable **rt)
370 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 icmp_param->data_len+icmp_param->head_len,
372 icmp_param->head_len,
373 ipc, rt, MSG_DONTWAIT) < 0) {
374 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 ip_flush_pending_frames(sk);
376 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 struct icmphdr *icmph = icmp_hdr(skb);
379 struct sk_buff *skb1;
381 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
383 icmp_param->head_len);
384 skb_queue_walk(&sk->sk_write_queue, skb1) {
385 csum = csum_add(csum, skb1->csum);
387 icmph->checksum = csum_fold(csum);
388 skb->ip_summed = CHECKSUM_NONE;
389 ip_push_pending_frames(sk, fl4);
394 * Driving logic for building and sending ICMP messages.
397 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
399 struct ipcm_cookie ipc;
400 struct rtable *rt = skb_rtable(skb);
401 struct net *net = dev_net(rt->dst.dev);
404 struct inet_sock *inet;
406 u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 int type = icmp_param->data.icmph.type;
408 int code = icmp_param->data.icmph.code;
410 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
413 /* Needed by both icmp_global_allow and icmp_xmit_lock */
416 /* global icmp_msgs_per_sec */
417 if (!icmpv4_global_allow(net, type, code))
420 sk = icmp_xmit_lock(net);
425 icmp_param->data.icmph.checksum = 0;
428 inet->tos = ip_hdr(skb)->tos;
429 ipc.sockc.mark = mark;
430 daddr = ipc.addr = ip_hdr(skb)->saddr;
431 saddr = fib_compute_spec_dst(skb);
433 if (icmp_param->replyopts.opt.opt.optlen) {
434 ipc.opt = &icmp_param->replyopts.opt;
435 if (ipc.opt->opt.srr)
436 daddr = icmp_param->replyopts.opt.opt.faddr;
438 memset(&fl4, 0, sizeof(fl4));
441 fl4.flowi4_mark = mark;
442 fl4.flowi4_uid = sock_net_uid(net, NULL);
443 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 fl4.flowi4_proto = IPPROTO_ICMP;
445 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
447 rt = ip_route_output_key(net, &fl4);
450 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
454 icmp_xmit_unlock(sk);
460 * The device used for looking up which routing table to use for sending an ICMP
461 * error is preferably the source whenever it is set, which should ensure the
462 * icmp error can be sent to the source host, else lookup using the routing
463 * table of the destination device, else use the main routing table (index 0).
465 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
467 struct net_device *route_lookup_dev = NULL;
470 route_lookup_dev = skb->dev;
471 else if (skb_dst(skb))
472 route_lookup_dev = skb_dst(skb)->dev;
473 return route_lookup_dev;
476 static struct rtable *icmp_route_lookup(struct net *net,
478 struct sk_buff *skb_in,
479 const struct iphdr *iph,
480 __be32 saddr, u8 tos, u32 mark,
482 struct icmp_bxm *param)
484 struct net_device *route_lookup_dev;
485 struct rtable *rt, *rt2;
486 struct flowi4 fl4_dec;
489 memset(fl4, 0, sizeof(*fl4));
490 fl4->daddr = (param->replyopts.opt.opt.srr ?
491 param->replyopts.opt.opt.faddr : iph->saddr);
493 fl4->flowi4_mark = mark;
494 fl4->flowi4_uid = sock_net_uid(net, NULL);
495 fl4->flowi4_tos = RT_TOS(tos);
496 fl4->flowi4_proto = IPPROTO_ICMP;
497 fl4->fl4_icmp_type = type;
498 fl4->fl4_icmp_code = code;
499 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
500 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
502 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
503 rt = ip_route_output_key_hash(net, fl4, skb_in);
507 /* No need to clone since we're just using its address. */
510 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
511 flowi4_to_flowi(fl4), NULL, 0);
515 } else if (PTR_ERR(rt) == -EPERM) {
520 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
522 goto relookup_failed;
524 if (inet_addr_type_dev_table(net, route_lookup_dev,
525 fl4_dec.saddr) == RTN_LOCAL) {
526 rt2 = __ip_route_output_key(net, &fl4_dec);
530 struct flowi4 fl4_2 = {};
531 unsigned long orefdst;
533 fl4_2.daddr = fl4_dec.saddr;
534 rt2 = ip_route_output_key(net, &fl4_2);
537 goto relookup_failed;
540 orefdst = skb_in->_skb_refdst; /* save old refdst */
541 skb_dst_set(skb_in, NULL);
542 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
543 RT_TOS(tos), rt2->dst.dev);
545 dst_release(&rt2->dst);
546 rt2 = skb_rtable(skb_in);
547 skb_in->_skb_refdst = orefdst; /* restore old refdst */
551 goto relookup_failed;
553 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
554 flowi4_to_flowi(&fl4_dec), NULL,
557 dst_release(&rt->dst);
558 memcpy(fl4, &fl4_dec, sizeof(*fl4));
560 } else if (PTR_ERR(rt2) == -EPERM) {
562 dst_release(&rt->dst);
566 goto relookup_failed;
577 * Send an ICMP message in response to a situation
579 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
580 * MAY send more (we do).
581 * MUST NOT change this header information.
582 * MUST NOT reply to a multicast/broadcast IP address.
583 * MUST NOT reply to a multicast/broadcast MAC address.
584 * MUST reply to only the first fragment.
587 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
588 const struct ip_options *opt)
592 struct icmp_bxm icmp_param;
593 struct rtable *rt = skb_rtable(skb_in);
594 struct ipcm_cookie ipc;
606 net = dev_net(rt->dst.dev);
607 else if (skb_in->dev)
608 net = dev_net(skb_in->dev);
613 * Find the original header. It is expected to be valid, of course.
614 * Check this, icmp_send is called from the most obscure devices
617 iph = ip_hdr(skb_in);
619 if ((u8 *)iph < skb_in->head ||
620 (skb_network_header(skb_in) + sizeof(*iph)) >
621 skb_tail_pointer(skb_in))
625 * No replies to physical multicast/broadcast
627 if (skb_in->pkt_type != PACKET_HOST)
631 * Now check at the protocol level
633 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
637 * Only reply to fragment 0. We byte re-order the constant
638 * mask for efficiency.
640 if (iph->frag_off & htons(IP_OFFSET))
644 * If we send an ICMP error to an ICMP error a mess would result..
646 if (icmp_pointers[type].error) {
648 * We are an error, check if we are replying to an
651 if (iph->protocol == IPPROTO_ICMP) {
652 u8 _inner_type, *itp;
654 itp = skb_header_pointer(skb_in,
655 skb_network_header(skb_in) +
657 offsetof(struct icmphdr,
666 * Assume any unknown ICMP type is an error. This
667 * isn't specified by the RFC, but think about it..
669 if (*itp > NR_ICMP_TYPES ||
670 icmp_pointers[*itp].error)
675 /* Needed by both icmp_global_allow and icmp_xmit_lock */
678 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
679 * incoming dev is loopback. If outgoing dev change to not be
680 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
682 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
683 !icmpv4_global_allow(net, type, code))
686 sk = icmp_xmit_lock(net);
691 * Construct source address and options.
695 if (!(rt->rt_flags & RTCF_LOCAL)) {
696 struct net_device *dev = NULL;
699 if (rt_is_input_route(rt) &&
700 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
701 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
704 saddr = inet_select_addr(dev, iph->saddr,
711 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
712 IPTOS_PREC_INTERNETCONTROL) :
714 mark = IP4_REPLY_MARK(net, skb_in->mark);
716 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
721 * Prepare data for ICMP header.
724 icmp_param.data.icmph.type = type;
725 icmp_param.data.icmph.code = code;
726 icmp_param.data.icmph.un.gateway = info;
727 icmp_param.data.icmph.checksum = 0;
728 icmp_param.skb = skb_in;
729 icmp_param.offset = skb_network_offset(skb_in);
730 inet_sk(sk)->tos = tos;
732 ipc.addr = iph->saddr;
733 ipc.opt = &icmp_param.replyopts.opt;
734 ipc.sockc.mark = mark;
736 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
737 type, code, &icmp_param);
741 /* peer icmp_ratelimit */
742 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
745 /* RFC says return as much as we can without exceeding 576 bytes. */
747 room = dst_mtu(&rt->dst);
750 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
751 room -= sizeof(struct icmphdr);
753 icmp_param.data_len = skb_in->len - icmp_param.offset;
754 if (icmp_param.data_len > room)
755 icmp_param.data_len = room;
756 icmp_param.head_len = sizeof(struct icmphdr);
758 /* if we don't have a source address at this point, fall back to the
759 * dummy address instead of sending out a packet with a source address
763 fl4.saddr = htonl(INADDR_DUMMY);
765 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
769 icmp_xmit_unlock(sk);
774 EXPORT_SYMBOL(__icmp_send);
776 #if IS_ENABLED(CONFIG_NF_NAT)
777 #include <net/netfilter/nf_conntrack.h>
778 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
780 struct sk_buff *cloned_skb = NULL;
781 struct ip_options opts = { 0 };
782 enum ip_conntrack_info ctinfo;
786 ct = nf_ct_get(skb_in, &ctinfo);
787 if (!ct || !(ct->status & IPS_SRC_NAT)) {
788 __icmp_send(skb_in, type, code, info, &opts);
792 if (skb_shared(skb_in))
793 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
795 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
796 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
797 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
798 skb_network_offset(skb_in) + sizeof(struct iphdr))))
801 orig_ip = ip_hdr(skb_in)->saddr;
802 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
803 __icmp_send(skb_in, type, code, info, &opts);
804 ip_hdr(skb_in)->saddr = orig_ip;
806 consume_skb(cloned_skb);
808 EXPORT_SYMBOL(icmp_ndo_send);
811 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
813 const struct iphdr *iph = (const struct iphdr *)skb->data;
814 const struct net_protocol *ipprot;
815 int protocol = iph->protocol;
817 /* Checkin full IP header plus 8 bytes of protocol to
818 * avoid additional coding at protocol handlers.
820 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
821 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
825 raw_icmp_error(skb, protocol, info);
827 ipprot = rcu_dereference(inet_protos[protocol]);
828 if (ipprot && ipprot->err_handler)
829 ipprot->err_handler(skb, info);
832 static bool icmp_tag_validation(int proto)
837 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
843 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
844 * ICMP_PARAMETERPROB.
847 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
849 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
850 const struct iphdr *iph;
851 struct icmphdr *icmph;
855 net = dev_net(skb_dst(skb)->dev);
858 * Incomplete header ?
859 * Only checks for the IP header, there should be an
860 * additional check for longer headers in upper levels.
863 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
866 icmph = icmp_hdr(skb);
867 iph = (const struct iphdr *)skb->data;
869 if (iph->ihl < 5) { /* Mangled header, drop. */
870 reason = SKB_DROP_REASON_IP_INHDR;
874 switch (icmph->type) {
875 case ICMP_DEST_UNREACH:
876 switch (icmph->code & 15) {
877 case ICMP_NET_UNREACH:
878 case ICMP_HOST_UNREACH:
879 case ICMP_PROT_UNREACH:
880 case ICMP_PORT_UNREACH:
882 case ICMP_FRAG_NEEDED:
883 /* for documentation of the ip_no_pmtu_disc
885 * Documentation/networking/ip-sysctl.rst
887 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
889 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
895 if (!icmp_tag_validation(iph->protocol))
899 info = ntohs(icmph->un.frag.mtu);
903 net_dbg_ratelimited("%pI4: Source Route Failed\n",
909 if (icmph->code > NR_ICMP_UNREACH)
912 case ICMP_PARAMETERPROB:
913 info = ntohl(icmph->un.gateway) >> 24;
915 case ICMP_TIME_EXCEEDED:
916 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
917 if (icmph->code == ICMP_EXC_FRAGTIME)
923 * Throw it at our lower layers
925 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
927 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
929 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
934 * Check the other end isn't violating RFC 1122. Some routers send
935 * bogus responses to broadcast frames. If you see this message
936 * first check your netmask matches at both ends, if it does then
937 * get the other vendor to fix their kit.
940 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
941 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
942 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
944 icmph->type, icmph->code,
945 &iph->daddr, skb->dev->name);
949 icmp_socket_deliver(skb, info);
954 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
955 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
960 * Handle ICMP_REDIRECT.
963 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
965 if (skb->len < sizeof(struct iphdr)) {
966 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
967 return SKB_DROP_REASON_PKT_TOO_SMALL;
970 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
971 /* there aught to be a stat */
972 return SKB_DROP_REASON_NOMEM;
975 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
976 return SKB_NOT_DROPPED_YET;
980 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
982 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
984 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
985 * included in the reply.
986 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
987 * echo requests, MUST have default=NOT.
988 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
989 * Extended Echo Functionality, MUST be disabled by default
990 * See also WRT handling of options once they are done and working.
993 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
995 struct icmp_bxm icmp_param;
998 net = dev_net(skb_dst(skb)->dev);
999 /* should there be an ICMP stat for ignored echos? */
1000 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1001 return SKB_NOT_DROPPED_YET;
1003 icmp_param.data.icmph = *icmp_hdr(skb);
1004 icmp_param.skb = skb;
1005 icmp_param.offset = 0;
1006 icmp_param.data_len = skb->len;
1007 icmp_param.head_len = sizeof(struct icmphdr);
1009 if (icmp_param.data.icmph.type == ICMP_ECHO)
1010 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1011 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1012 return SKB_NOT_DROPPED_YET;
1014 icmp_reply(&icmp_param, skb);
1015 return SKB_NOT_DROPPED_YET;
1018 /* Helper for icmp_echo and icmpv6_echo_reply.
1019 * Searches for net_device that matches PROBE interface identifier
1020 * and builds PROBE reply message in icmphdr.
1022 * Returns false if PROBE responses are disabled via sysctl
1025 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1027 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1028 struct icmp_ext_echo_iio *iio, _iio;
1029 struct net *net = dev_net(skb->dev);
1030 struct net_device *dev;
1031 char buff[IFNAMSIZ];
1035 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1038 /* We currently only support probing interfaces on the proxy node
1039 * Check to ensure L-bit is set
1041 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1043 /* Clear status bits in reply message */
1044 icmphdr->un.echo.sequence &= htons(0xFF00);
1045 if (icmphdr->type == ICMP_EXT_ECHO)
1046 icmphdr->type = ICMP_EXT_ECHOREPLY;
1048 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1049 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1050 /* Size of iio is class_type dependent.
1051 * Only check header here and assign length based on ctype in the switch statement
1053 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1054 if (!ext_hdr || !iio)
1055 goto send_mal_query;
1056 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1057 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1058 goto send_mal_query;
1059 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1060 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1061 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1063 goto send_mal_query;
1067 switch (iio->extobj_hdr.class_type) {
1068 case ICMP_EXT_ECHO_CTYPE_NAME:
1069 if (ident_len >= IFNAMSIZ)
1070 goto send_mal_query;
1071 memset(buff, 0, sizeof(buff));
1072 memcpy(buff, &iio->ident.name, ident_len);
1073 dev = dev_get_by_name(net, buff);
1075 case ICMP_EXT_ECHO_CTYPE_INDEX:
1076 if (ident_len != sizeof(iio->ident.ifindex))
1077 goto send_mal_query;
1078 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1080 case ICMP_EXT_ECHO_CTYPE_ADDR:
1081 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1082 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1083 iio->ident.addr.ctype3_hdr.addrlen)
1084 goto send_mal_query;
1085 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1087 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1088 goto send_mal_query;
1089 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1091 #if IS_ENABLED(CONFIG_IPV6)
1093 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1094 goto send_mal_query;
1095 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1100 goto send_mal_query;
1104 goto send_mal_query;
1107 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1110 /* Fill bits in reply message */
1111 if (dev->flags & IFF_UP)
1112 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1113 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1114 status |= ICMP_EXT_ECHOREPLY_IPV4;
1115 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1116 status |= ICMP_EXT_ECHOREPLY_IPV6;
1118 icmphdr->un.echo.sequence |= htons(status);
1121 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1124 EXPORT_SYMBOL_GPL(icmp_build_probe);
1127 * Handle ICMP Timestamp requests.
1128 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1129 * SHOULD be in the kernel for minimum random latency.
1130 * MUST be accurate to a few minutes.
1131 * MUST be updated at least at 15Hz.
1133 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1135 struct icmp_bxm icmp_param;
1143 * Fill in the current time as ms since midnight UT:
1145 icmp_param.data.times[1] = inet_current_timestamp();
1146 icmp_param.data.times[2] = icmp_param.data.times[1];
1148 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1150 icmp_param.data.icmph = *icmp_hdr(skb);
1151 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1152 icmp_param.data.icmph.code = 0;
1153 icmp_param.skb = skb;
1154 icmp_param.offset = 0;
1155 icmp_param.data_len = 0;
1156 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1157 icmp_reply(&icmp_param, skb);
1158 return SKB_NOT_DROPPED_YET;
1161 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1162 return SKB_DROP_REASON_PKT_TOO_SMALL;
1165 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1167 /* pretend it was a success */
1168 return SKB_NOT_DROPPED_YET;
1172 * Deal with incoming ICMP packets.
1174 int icmp_rcv(struct sk_buff *skb)
1176 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1177 struct rtable *rt = skb_rtable(skb);
1178 struct net *net = dev_net(rt->dst.dev);
1179 struct icmphdr *icmph;
1181 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1182 struct sec_path *sp = skb_sec_path(skb);
1185 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1187 reason = SKB_DROP_REASON_XFRM_POLICY;
1191 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1194 nh = skb_network_offset(skb);
1195 skb_set_network_header(skb, sizeof(*icmph));
1197 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1199 reason = SKB_DROP_REASON_XFRM_POLICY;
1203 skb_set_network_header(skb, nh);
1206 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1208 if (skb_checksum_simple_validate(skb))
1211 if (!pskb_pull(skb, sizeof(*icmph)))
1214 icmph = icmp_hdr(skb);
1216 ICMPMSGIN_INC_STATS(net, icmph->type);
1218 /* Check for ICMP Extended Echo (PROBE) messages */
1219 if (icmph->type == ICMP_EXT_ECHO) {
1220 /* We can't use icmp_pointers[].handler() because it is an array of
1221 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1223 reason = icmp_echo(skb);
1227 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1228 reason = ping_rcv(skb);
1233 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1235 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1238 if (icmph->type > NR_ICMP_TYPES) {
1239 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1244 * Parse the ICMP message
1247 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1249 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1250 * silently ignored (we let user decide with a sysctl).
1251 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1252 * discarded if to broadcast/multicast.
1254 if ((icmph->type == ICMP_ECHO ||
1255 icmph->type == ICMP_TIMESTAMP) &&
1256 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1257 reason = SKB_DROP_REASON_INVALID_PROTO;
1260 if (icmph->type != ICMP_ECHO &&
1261 icmph->type != ICMP_TIMESTAMP &&
1262 icmph->type != ICMP_ADDRESS &&
1263 icmph->type != ICMP_ADDRESSREPLY) {
1264 reason = SKB_DROP_REASON_INVALID_PROTO;
1269 reason = icmp_pointers[icmph->type].handler(skb);
1273 return NET_RX_SUCCESS;
1277 kfree_skb_reason(skb, reason);
1280 reason = SKB_DROP_REASON_ICMP_CSUM;
1281 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1283 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1287 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1289 struct icmp_extobj_hdr *objh, _objh;
1290 struct icmp_ext_hdr *exth, _exth;
1293 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1296 if (exth->version != 2)
1299 if (exth->checksum &&
1300 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1303 off += sizeof(_exth);
1304 while (off < skb->len) {
1305 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1309 olen = ntohs(objh->length);
1310 if (olen < sizeof(_objh))
1321 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1322 struct sock_ee_data_rfc4884 *out,
1327 /* original datagram headers: end of icmph to payload (skb->data) */
1328 hlen = -skb_transport_offset(skb) - thlen;
1330 /* per rfc 4884: minimal datagram length of 128 bytes */
1331 if (off < 128 || off < hlen)
1334 /* kernel has stripped headers: return payload offset in bytes */
1336 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1341 if (!ip_icmp_error_rfc4884_validate(skb, off))
1342 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1344 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1346 int icmp_err(struct sk_buff *skb, u32 info)
1348 struct iphdr *iph = (struct iphdr *)skb->data;
1349 int offset = iph->ihl<<2;
1350 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1351 int type = icmp_hdr(skb)->type;
1352 int code = icmp_hdr(skb)->code;
1353 struct net *net = dev_net(skb->dev);
1356 * Use ping_err to handle all icmp errors except those
1357 * triggered by ICMP_ECHOREPLY which sent from kernel.
1359 if (icmph->type != ICMP_ECHOREPLY) {
1360 ping_err(skb, offset, info);
1364 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1365 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1366 else if (type == ICMP_REDIRECT)
1367 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1373 * This table is the definition of how we handle ICMP.
1375 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1376 [ICMP_ECHOREPLY] = {
1377 .handler = ping_rcv,
1380 .handler = icmp_discard,
1384 .handler = icmp_discard,
1387 [ICMP_DEST_UNREACH] = {
1388 .handler = icmp_unreach,
1391 [ICMP_SOURCE_QUENCH] = {
1392 .handler = icmp_unreach,
1396 .handler = icmp_redirect,
1400 .handler = icmp_discard,
1404 .handler = icmp_discard,
1408 .handler = icmp_echo,
1411 .handler = icmp_discard,
1415 .handler = icmp_discard,
1418 [ICMP_TIME_EXCEEDED] = {
1419 .handler = icmp_unreach,
1422 [ICMP_PARAMETERPROB] = {
1423 .handler = icmp_unreach,
1426 [ICMP_TIMESTAMP] = {
1427 .handler = icmp_timestamp,
1429 [ICMP_TIMESTAMPREPLY] = {
1430 .handler = icmp_discard,
1432 [ICMP_INFO_REQUEST] = {
1433 .handler = icmp_discard,
1435 [ICMP_INFO_REPLY] = {
1436 .handler = icmp_discard,
1439 .handler = icmp_discard,
1441 [ICMP_ADDRESSREPLY] = {
1442 .handler = icmp_discard,
1446 static int __net_init icmp_sk_init(struct net *net)
1448 /* Control parameters for ECHO replies. */
1449 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1450 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1451 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1453 /* Control parameter - ignore bogus broadcast responses? */
1454 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1457 * Configurable global rate limit.
1459 * ratelimit defines tokens/packet consumed for dst->rate_token
1460 * bucket ratemask defines which icmp types are ratelimited by
1461 * setting it's bit position.
1464 * dest unreachable (3), source quench (4),
1465 * time exceeded (11), parameter problem (12)
1468 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1469 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1470 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1475 static struct pernet_operations __net_initdata icmp_sk_ops = {
1476 .init = icmp_sk_init,
1479 int __init icmp_init(void)
1483 for_each_possible_cpu(i) {
1486 err = inet_ctl_sock_create(&sk, PF_INET,
1487 SOCK_RAW, IPPROTO_ICMP, &init_net);
1491 per_cpu(ipv4_icmp_sk, i) = sk;
1493 /* Enough space for 2 64K ICMP packets, including
1494 * sk_buff/skb_shared_info struct overhead.
1496 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1499 * Speedup sock_wfree()
1501 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1502 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1504 return register_pernet_subsys(&icmp_sk_ops);