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 - prandom_u32_max(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())
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));
333 * Maintain the counters used in the SNMP statistics for outgoing ICMP
335 void icmp_out_count(struct net *net, unsigned char type)
337 ICMPMSGOUT_INC_STATS(net, type);
338 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
342 * Checksum each fragment, and on the first include the headers and final
345 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
348 struct icmp_bxm *icmp_param = from;
351 csum = skb_copy_and_csum_bits(icmp_param->skb,
352 icmp_param->offset + offset,
355 skb->csum = csum_block_add(skb->csum, csum, odd);
356 if (icmp_pointers[icmp_param->data.icmph.type].error)
357 nf_ct_attach(skb, icmp_param->skb);
361 static void icmp_push_reply(struct sock *sk,
362 struct icmp_bxm *icmp_param,
364 struct ipcm_cookie *ipc, struct rtable **rt)
368 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
369 icmp_param->data_len+icmp_param->head_len,
370 icmp_param->head_len,
371 ipc, rt, MSG_DONTWAIT) < 0) {
372 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
373 ip_flush_pending_frames(sk);
374 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
375 struct icmphdr *icmph = icmp_hdr(skb);
377 struct sk_buff *skb1;
379 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
381 icmp_param->head_len);
382 skb_queue_walk(&sk->sk_write_queue, skb1) {
383 csum = csum_add(csum, skb1->csum);
385 icmph->checksum = csum_fold(csum);
386 skb->ip_summed = CHECKSUM_NONE;
387 ip_push_pending_frames(sk, fl4);
392 * Driving logic for building and sending ICMP messages.
395 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
397 struct ipcm_cookie ipc;
398 struct rtable *rt = skb_rtable(skb);
399 struct net *net = dev_net(rt->dst.dev);
402 struct inet_sock *inet;
404 u32 mark = IP4_REPLY_MARK(net, skb->mark);
405 int type = icmp_param->data.icmph.type;
406 int code = icmp_param->data.icmph.code;
408 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 /* Needed by both icmp_global_allow and icmp_xmit_lock */
414 /* global icmp_msgs_per_sec */
415 if (!icmpv4_global_allow(net, type, code))
418 sk = icmp_xmit_lock(net);
423 icmp_param->data.icmph.checksum = 0;
426 inet->tos = ip_hdr(skb)->tos;
427 ipc.sockc.mark = mark;
428 daddr = ipc.addr = ip_hdr(skb)->saddr;
429 saddr = fib_compute_spec_dst(skb);
431 if (icmp_param->replyopts.opt.opt.optlen) {
432 ipc.opt = &icmp_param->replyopts.opt;
433 if (ipc.opt->opt.srr)
434 daddr = icmp_param->replyopts.opt.opt.faddr;
436 memset(&fl4, 0, sizeof(fl4));
439 fl4.flowi4_mark = mark;
440 fl4.flowi4_uid = sock_net_uid(net, NULL);
441 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
442 fl4.flowi4_proto = IPPROTO_ICMP;
443 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
444 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
445 rt = ip_route_output_key(net, &fl4);
448 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
449 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
452 icmp_xmit_unlock(sk);
458 * The device used for looking up which routing table to use for sending an ICMP
459 * error is preferably the source whenever it is set, which should ensure the
460 * icmp error can be sent to the source host, else lookup using the routing
461 * table of the destination device, else use the main routing table (index 0).
463 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
465 struct net_device *route_lookup_dev = NULL;
468 route_lookup_dev = skb->dev;
469 else if (skb_dst(skb))
470 route_lookup_dev = skb_dst(skb)->dev;
471 return route_lookup_dev;
474 static struct rtable *icmp_route_lookup(struct net *net,
476 struct sk_buff *skb_in,
477 const struct iphdr *iph,
478 __be32 saddr, u8 tos, u32 mark,
480 struct icmp_bxm *param)
482 struct net_device *route_lookup_dev;
483 struct rtable *rt, *rt2;
484 struct flowi4 fl4_dec;
487 memset(fl4, 0, sizeof(*fl4));
488 fl4->daddr = (param->replyopts.opt.opt.srr ?
489 param->replyopts.opt.opt.faddr : iph->saddr);
491 fl4->flowi4_mark = mark;
492 fl4->flowi4_uid = sock_net_uid(net, NULL);
493 fl4->flowi4_tos = RT_TOS(tos);
494 fl4->flowi4_proto = IPPROTO_ICMP;
495 fl4->fl4_icmp_type = type;
496 fl4->fl4_icmp_code = code;
497 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
498 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
500 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
501 rt = ip_route_output_key_hash(net, fl4, skb_in);
505 /* No need to clone since we're just using its address. */
508 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
509 flowi4_to_flowi(fl4), NULL, 0);
513 } else if (PTR_ERR(rt) == -EPERM) {
518 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
520 goto relookup_failed;
522 if (inet_addr_type_dev_table(net, route_lookup_dev,
523 fl4_dec.saddr) == RTN_LOCAL) {
524 rt2 = __ip_route_output_key(net, &fl4_dec);
528 struct flowi4 fl4_2 = {};
529 unsigned long orefdst;
531 fl4_2.daddr = fl4_dec.saddr;
532 rt2 = ip_route_output_key(net, &fl4_2);
535 goto relookup_failed;
538 orefdst = skb_in->_skb_refdst; /* save old refdst */
539 skb_dst_set(skb_in, NULL);
540 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
541 RT_TOS(tos), rt2->dst.dev);
543 dst_release(&rt2->dst);
544 rt2 = skb_rtable(skb_in);
545 skb_in->_skb_refdst = orefdst; /* restore old refdst */
549 goto relookup_failed;
551 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
552 flowi4_to_flowi(&fl4_dec), NULL,
555 dst_release(&rt->dst);
556 memcpy(fl4, &fl4_dec, sizeof(*fl4));
558 } else if (PTR_ERR(rt2) == -EPERM) {
560 dst_release(&rt->dst);
564 goto relookup_failed;
575 * Send an ICMP message in response to a situation
577 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
578 * MAY send more (we do).
579 * MUST NOT change this header information.
580 * MUST NOT reply to a multicast/broadcast IP address.
581 * MUST NOT reply to a multicast/broadcast MAC address.
582 * MUST reply to only the first fragment.
585 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
586 const struct ip_options *opt)
590 struct icmp_bxm icmp_param;
591 struct rtable *rt = skb_rtable(skb_in);
592 struct ipcm_cookie ipc;
604 net = dev_net(rt->dst.dev);
605 else if (skb_in->dev)
606 net = dev_net(skb_in->dev);
611 * Find the original header. It is expected to be valid, of course.
612 * Check this, icmp_send is called from the most obscure devices
615 iph = ip_hdr(skb_in);
617 if ((u8 *)iph < skb_in->head ||
618 (skb_network_header(skb_in) + sizeof(*iph)) >
619 skb_tail_pointer(skb_in))
623 * No replies to physical multicast/broadcast
625 if (skb_in->pkt_type != PACKET_HOST)
629 * Now check at the protocol level
631 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
635 * Only reply to fragment 0. We byte re-order the constant
636 * mask for efficiency.
638 if (iph->frag_off & htons(IP_OFFSET))
642 * If we send an ICMP error to an ICMP error a mess would result..
644 if (icmp_pointers[type].error) {
646 * We are an error, check if we are replying to an
649 if (iph->protocol == IPPROTO_ICMP) {
650 u8 _inner_type, *itp;
652 itp = skb_header_pointer(skb_in,
653 skb_network_header(skb_in) +
655 offsetof(struct icmphdr,
664 * Assume any unknown ICMP type is an error. This
665 * isn't specified by the RFC, but think about it..
667 if (*itp > NR_ICMP_TYPES ||
668 icmp_pointers[*itp].error)
673 /* Needed by both icmp_global_allow and icmp_xmit_lock */
676 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
677 * incoming dev is loopback. If outgoing dev change to not be
678 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
680 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
681 !icmpv4_global_allow(net, type, code))
684 sk = icmp_xmit_lock(net);
689 * Construct source address and options.
693 if (!(rt->rt_flags & RTCF_LOCAL)) {
694 struct net_device *dev = NULL;
697 if (rt_is_input_route(rt) &&
698 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
699 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
702 saddr = inet_select_addr(dev, iph->saddr,
709 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
710 IPTOS_PREC_INTERNETCONTROL) :
712 mark = IP4_REPLY_MARK(net, skb_in->mark);
714 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
719 * Prepare data for ICMP header.
722 icmp_param.data.icmph.type = type;
723 icmp_param.data.icmph.code = code;
724 icmp_param.data.icmph.un.gateway = info;
725 icmp_param.data.icmph.checksum = 0;
726 icmp_param.skb = skb_in;
727 icmp_param.offset = skb_network_offset(skb_in);
728 inet_sk(sk)->tos = tos;
730 ipc.addr = iph->saddr;
731 ipc.opt = &icmp_param.replyopts.opt;
732 ipc.sockc.mark = mark;
734 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
735 type, code, &icmp_param);
739 /* peer icmp_ratelimit */
740 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
743 /* RFC says return as much as we can without exceeding 576 bytes. */
745 room = dst_mtu(&rt->dst);
748 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
749 room -= sizeof(struct icmphdr);
750 /* Guard against tiny mtu. We need to include at least one
751 * IP network header for this message to make any sense.
753 if (room <= (int)sizeof(struct iphdr))
756 icmp_param.data_len = skb_in->len - icmp_param.offset;
757 if (icmp_param.data_len > room)
758 icmp_param.data_len = room;
759 icmp_param.head_len = sizeof(struct icmphdr);
761 /* if we don't have a source address at this point, fall back to the
762 * dummy address instead of sending out a packet with a source address
766 fl4.saddr = htonl(INADDR_DUMMY);
768 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
772 icmp_xmit_unlock(sk);
777 EXPORT_SYMBOL(__icmp_send);
779 #if IS_ENABLED(CONFIG_NF_NAT)
780 #include <net/netfilter/nf_conntrack.h>
781 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
783 struct sk_buff *cloned_skb = NULL;
784 struct ip_options opts = { 0 };
785 enum ip_conntrack_info ctinfo;
789 ct = nf_ct_get(skb_in, &ctinfo);
790 if (!ct || !(ct->status & IPS_SRC_NAT)) {
791 __icmp_send(skb_in, type, code, info, &opts);
795 if (skb_shared(skb_in))
796 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
798 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
799 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
800 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
801 skb_network_offset(skb_in) + sizeof(struct iphdr))))
804 orig_ip = ip_hdr(skb_in)->saddr;
805 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
806 __icmp_send(skb_in, type, code, info, &opts);
807 ip_hdr(skb_in)->saddr = orig_ip;
809 consume_skb(cloned_skb);
811 EXPORT_SYMBOL(icmp_ndo_send);
814 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
816 const struct iphdr *iph = (const struct iphdr *)skb->data;
817 const struct net_protocol *ipprot;
818 int protocol = iph->protocol;
820 /* Checkin full IP header plus 8 bytes of protocol to
821 * avoid additional coding at protocol handlers.
823 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
824 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
828 raw_icmp_error(skb, protocol, info);
830 ipprot = rcu_dereference(inet_protos[protocol]);
831 if (ipprot && ipprot->err_handler)
832 ipprot->err_handler(skb, info);
835 static bool icmp_tag_validation(int proto)
840 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
846 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
847 * ICMP_PARAMETERPROB.
850 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
852 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
853 const struct iphdr *iph;
854 struct icmphdr *icmph;
858 net = dev_net(skb_dst(skb)->dev);
861 * Incomplete header ?
862 * Only checks for the IP header, there should be an
863 * additional check for longer headers in upper levels.
866 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
869 icmph = icmp_hdr(skb);
870 iph = (const struct iphdr *)skb->data;
872 if (iph->ihl < 5) { /* Mangled header, drop. */
873 reason = SKB_DROP_REASON_IP_INHDR;
877 switch (icmph->type) {
878 case ICMP_DEST_UNREACH:
879 switch (icmph->code & 15) {
880 case ICMP_NET_UNREACH:
881 case ICMP_HOST_UNREACH:
882 case ICMP_PROT_UNREACH:
883 case ICMP_PORT_UNREACH:
885 case ICMP_FRAG_NEEDED:
886 /* for documentation of the ip_no_pmtu_disc
888 * Documentation/networking/ip-sysctl.rst
890 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
892 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
898 if (!icmp_tag_validation(iph->protocol))
902 info = ntohs(icmph->un.frag.mtu);
906 net_dbg_ratelimited("%pI4: Source Route Failed\n",
912 if (icmph->code > NR_ICMP_UNREACH)
915 case ICMP_PARAMETERPROB:
916 info = ntohl(icmph->un.gateway) >> 24;
918 case ICMP_TIME_EXCEEDED:
919 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
920 if (icmph->code == ICMP_EXC_FRAGTIME)
926 * Throw it at our lower layers
928 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
930 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
932 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
937 * Check the other end isn't violating RFC 1122. Some routers send
938 * bogus responses to broadcast frames. If you see this message
939 * first check your netmask matches at both ends, if it does then
940 * get the other vendor to fix their kit.
943 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
944 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
945 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
947 icmph->type, icmph->code,
948 &iph->daddr, skb->dev->name);
952 icmp_socket_deliver(skb, info);
957 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
958 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
963 * Handle ICMP_REDIRECT.
966 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
968 if (skb->len < sizeof(struct iphdr)) {
969 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
970 return SKB_DROP_REASON_PKT_TOO_SMALL;
973 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
974 /* there aught to be a stat */
975 return SKB_DROP_REASON_NOMEM;
978 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
979 return SKB_NOT_DROPPED_YET;
983 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
985 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
987 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
988 * included in the reply.
989 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
990 * echo requests, MUST have default=NOT.
991 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
992 * Extended Echo Functionality, MUST be disabled by default
993 * See also WRT handling of options once they are done and working.
996 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
998 struct icmp_bxm icmp_param;
1001 net = dev_net(skb_dst(skb)->dev);
1002 /* should there be an ICMP stat for ignored echos? */
1003 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1004 return SKB_NOT_DROPPED_YET;
1006 icmp_param.data.icmph = *icmp_hdr(skb);
1007 icmp_param.skb = skb;
1008 icmp_param.offset = 0;
1009 icmp_param.data_len = skb->len;
1010 icmp_param.head_len = sizeof(struct icmphdr);
1012 if (icmp_param.data.icmph.type == ICMP_ECHO)
1013 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1014 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1015 return SKB_NOT_DROPPED_YET;
1017 icmp_reply(&icmp_param, skb);
1018 return SKB_NOT_DROPPED_YET;
1021 /* Helper for icmp_echo and icmpv6_echo_reply.
1022 * Searches for net_device that matches PROBE interface identifier
1023 * and builds PROBE reply message in icmphdr.
1025 * Returns false if PROBE responses are disabled via sysctl
1028 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1030 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1031 struct icmp_ext_echo_iio *iio, _iio;
1032 struct net *net = dev_net(skb->dev);
1033 struct inet6_dev *in6_dev;
1034 struct in_device *in_dev;
1035 struct net_device *dev;
1036 char buff[IFNAMSIZ];
1040 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1043 /* We currently only support probing interfaces on the proxy node
1044 * Check to ensure L-bit is set
1046 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1048 /* Clear status bits in reply message */
1049 icmphdr->un.echo.sequence &= htons(0xFF00);
1050 if (icmphdr->type == ICMP_EXT_ECHO)
1051 icmphdr->type = ICMP_EXT_ECHOREPLY;
1053 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1054 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1055 /* Size of iio is class_type dependent.
1056 * Only check header here and assign length based on ctype in the switch statement
1058 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1059 if (!ext_hdr || !iio)
1060 goto send_mal_query;
1061 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1062 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1063 goto send_mal_query;
1064 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1065 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1066 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1068 goto send_mal_query;
1072 switch (iio->extobj_hdr.class_type) {
1073 case ICMP_EXT_ECHO_CTYPE_NAME:
1074 if (ident_len >= IFNAMSIZ)
1075 goto send_mal_query;
1076 memset(buff, 0, sizeof(buff));
1077 memcpy(buff, &iio->ident.name, ident_len);
1078 dev = dev_get_by_name(net, buff);
1080 case ICMP_EXT_ECHO_CTYPE_INDEX:
1081 if (ident_len != sizeof(iio->ident.ifindex))
1082 goto send_mal_query;
1083 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1085 case ICMP_EXT_ECHO_CTYPE_ADDR:
1086 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1087 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1088 iio->ident.addr.ctype3_hdr.addrlen)
1089 goto send_mal_query;
1090 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1092 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1093 goto send_mal_query;
1094 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1096 #if IS_ENABLED(CONFIG_IPV6)
1098 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1099 goto send_mal_query;
1100 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1105 goto send_mal_query;
1109 goto send_mal_query;
1112 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1115 /* Fill bits in reply message */
1116 if (dev->flags & IFF_UP)
1117 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1119 in_dev = __in_dev_get_rcu(dev);
1120 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1121 status |= ICMP_EXT_ECHOREPLY_IPV4;
1123 in6_dev = __in6_dev_get(dev);
1124 if (in6_dev && !list_empty(&in6_dev->addr_list))
1125 status |= ICMP_EXT_ECHOREPLY_IPV6;
1128 icmphdr->un.echo.sequence |= htons(status);
1131 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1134 EXPORT_SYMBOL_GPL(icmp_build_probe);
1137 * Handle ICMP Timestamp requests.
1138 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1139 * SHOULD be in the kernel for minimum random latency.
1140 * MUST be accurate to a few minutes.
1141 * MUST be updated at least at 15Hz.
1143 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1145 struct icmp_bxm icmp_param;
1153 * Fill in the current time as ms since midnight UT:
1155 icmp_param.data.times[1] = inet_current_timestamp();
1156 icmp_param.data.times[2] = icmp_param.data.times[1];
1158 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1160 icmp_param.data.icmph = *icmp_hdr(skb);
1161 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1162 icmp_param.data.icmph.code = 0;
1163 icmp_param.skb = skb;
1164 icmp_param.offset = 0;
1165 icmp_param.data_len = 0;
1166 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1167 icmp_reply(&icmp_param, skb);
1168 return SKB_NOT_DROPPED_YET;
1171 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1172 return SKB_DROP_REASON_PKT_TOO_SMALL;
1175 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1177 /* pretend it was a success */
1178 return SKB_NOT_DROPPED_YET;
1182 * Deal with incoming ICMP packets.
1184 int icmp_rcv(struct sk_buff *skb)
1186 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1187 struct rtable *rt = skb_rtable(skb);
1188 struct net *net = dev_net(rt->dst.dev);
1189 struct icmphdr *icmph;
1191 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1192 struct sec_path *sp = skb_sec_path(skb);
1195 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1197 reason = SKB_DROP_REASON_XFRM_POLICY;
1201 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1204 nh = skb_network_offset(skb);
1205 skb_set_network_header(skb, sizeof(*icmph));
1207 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1209 reason = SKB_DROP_REASON_XFRM_POLICY;
1213 skb_set_network_header(skb, nh);
1216 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1218 if (skb_checksum_simple_validate(skb))
1221 if (!pskb_pull(skb, sizeof(*icmph)))
1224 icmph = icmp_hdr(skb);
1226 ICMPMSGIN_INC_STATS(net, icmph->type);
1228 /* Check for ICMP Extended Echo (PROBE) messages */
1229 if (icmph->type == ICMP_EXT_ECHO) {
1230 /* We can't use icmp_pointers[].handler() because it is an array of
1231 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1233 reason = icmp_echo(skb);
1237 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1238 reason = ping_rcv(skb);
1243 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1245 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1248 if (icmph->type > NR_ICMP_TYPES) {
1249 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1254 * Parse the ICMP message
1257 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1259 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1260 * silently ignored (we let user decide with a sysctl).
1261 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1262 * discarded if to broadcast/multicast.
1264 if ((icmph->type == ICMP_ECHO ||
1265 icmph->type == ICMP_TIMESTAMP) &&
1266 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1267 reason = SKB_DROP_REASON_INVALID_PROTO;
1270 if (icmph->type != ICMP_ECHO &&
1271 icmph->type != ICMP_TIMESTAMP &&
1272 icmph->type != ICMP_ADDRESS &&
1273 icmph->type != ICMP_ADDRESSREPLY) {
1274 reason = SKB_DROP_REASON_INVALID_PROTO;
1279 reason = icmp_pointers[icmph->type].handler(skb);
1283 return NET_RX_SUCCESS;
1287 kfree_skb_reason(skb, reason);
1290 reason = SKB_DROP_REASON_ICMP_CSUM;
1291 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1293 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1297 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1299 struct icmp_extobj_hdr *objh, _objh;
1300 struct icmp_ext_hdr *exth, _exth;
1303 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1306 if (exth->version != 2)
1309 if (exth->checksum &&
1310 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1313 off += sizeof(_exth);
1314 while (off < skb->len) {
1315 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1319 olen = ntohs(objh->length);
1320 if (olen < sizeof(_objh))
1331 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1332 struct sock_ee_data_rfc4884 *out,
1337 /* original datagram headers: end of icmph to payload (skb->data) */
1338 hlen = -skb_transport_offset(skb) - thlen;
1340 /* per rfc 4884: minimal datagram length of 128 bytes */
1341 if (off < 128 || off < hlen)
1344 /* kernel has stripped headers: return payload offset in bytes */
1346 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1351 if (!ip_icmp_error_rfc4884_validate(skb, off))
1352 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1354 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1356 int icmp_err(struct sk_buff *skb, u32 info)
1358 struct iphdr *iph = (struct iphdr *)skb->data;
1359 int offset = iph->ihl<<2;
1360 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1361 int type = icmp_hdr(skb)->type;
1362 int code = icmp_hdr(skb)->code;
1363 struct net *net = dev_net(skb->dev);
1366 * Use ping_err to handle all icmp errors except those
1367 * triggered by ICMP_ECHOREPLY which sent from kernel.
1369 if (icmph->type != ICMP_ECHOREPLY) {
1370 ping_err(skb, offset, info);
1374 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1375 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1376 else if (type == ICMP_REDIRECT)
1377 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1383 * This table is the definition of how we handle ICMP.
1385 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1386 [ICMP_ECHOREPLY] = {
1387 .handler = ping_rcv,
1390 .handler = icmp_discard,
1394 .handler = icmp_discard,
1397 [ICMP_DEST_UNREACH] = {
1398 .handler = icmp_unreach,
1401 [ICMP_SOURCE_QUENCH] = {
1402 .handler = icmp_unreach,
1406 .handler = icmp_redirect,
1410 .handler = icmp_discard,
1414 .handler = icmp_discard,
1418 .handler = icmp_echo,
1421 .handler = icmp_discard,
1425 .handler = icmp_discard,
1428 [ICMP_TIME_EXCEEDED] = {
1429 .handler = icmp_unreach,
1432 [ICMP_PARAMETERPROB] = {
1433 .handler = icmp_unreach,
1436 [ICMP_TIMESTAMP] = {
1437 .handler = icmp_timestamp,
1439 [ICMP_TIMESTAMPREPLY] = {
1440 .handler = icmp_discard,
1442 [ICMP_INFO_REQUEST] = {
1443 .handler = icmp_discard,
1445 [ICMP_INFO_REPLY] = {
1446 .handler = icmp_discard,
1449 .handler = icmp_discard,
1451 [ICMP_ADDRESSREPLY] = {
1452 .handler = icmp_discard,
1456 static int __net_init icmp_sk_init(struct net *net)
1458 /* Control parameters for ECHO replies. */
1459 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1460 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1461 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1463 /* Control parameter - ignore bogus broadcast responses? */
1464 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1467 * Configurable global rate limit.
1469 * ratelimit defines tokens/packet consumed for dst->rate_token
1470 * bucket ratemask defines which icmp types are ratelimited by
1471 * setting it's bit position.
1474 * dest unreachable (3), source quench (4),
1475 * time exceeded (11), parameter problem (12)
1478 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1479 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1480 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1485 static struct pernet_operations __net_initdata icmp_sk_ops = {
1486 .init = icmp_sk_init,
1489 int __init icmp_init(void)
1493 for_each_possible_cpu(i) {
1496 err = inet_ctl_sock_create(&sk, PF_INET,
1497 SOCK_RAW, IPPROTO_ICMP, &init_net);
1501 per_cpu(ipv4_icmp_sk, i) = sk;
1503 /* Enough space for 2 64K ICMP packets, including
1504 * sk_buff/skb_shared_info struct overhead.
1506 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1509 * Speedup sock_wfree()
1511 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1512 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1514 return register_pernet_subsys(&icmp_sk_ops);