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 bool (*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];
196 * The ICMP socket(s). This is the most convenient way to flow control
197 * our ICMP output as well as maintain a clean interface throughout
198 * all layers. All Socketless IP sends will soon be gone.
200 * On SMP we have one ICMP socket per-cpu.
202 static struct sock *icmp_sk(struct net *net)
204 return this_cpu_read(*net->ipv4.icmp_sk);
207 /* Called with BH disabled */
208 static inline struct sock *icmp_xmit_lock(struct net *net)
214 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 /* This can happen if the output path signals a
216 * dst_link_failure() for an outgoing ICMP packet.
223 static inline void icmp_xmit_unlock(struct sock *sk)
225 spin_unlock(&sk->sk_lock.slock);
228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229 int sysctl_icmp_msgs_burst __read_mostly = 50;
236 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
240 * icmp_global_allow - Are we allowed to send one more ICMP message ?
242 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
243 * Returns false if we reached the limit and can not send another packet.
244 * Note: called with BH disabled
246 bool icmp_global_allow(void)
248 u32 credit, delta, incr = 0, now = (u32)jiffies;
251 /* Check if token bucket is empty and cannot be refilled
252 * without taking the spinlock. The READ_ONCE() are paired
253 * with the following WRITE_ONCE() in this same function.
255 if (!READ_ONCE(icmp_global.credit)) {
256 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
261 spin_lock(&icmp_global.lock);
262 delta = min_t(u32, now - icmp_global.stamp, HZ);
263 if (delta >= HZ / 50) {
264 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
266 WRITE_ONCE(icmp_global.stamp, now);
268 credit = min_t(u32, icmp_global.credit + incr,
269 READ_ONCE(sysctl_icmp_msgs_burst));
271 /* We want to use a credit of one in average, but need to randomize
272 * it for security reasons.
274 credit = max_t(int, credit - prandom_u32_max(3), 0);
277 WRITE_ONCE(icmp_global.credit, credit);
278 spin_unlock(&icmp_global.lock);
281 EXPORT_SYMBOL(icmp_global_allow);
283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
285 if (type > NR_ICMP_TYPES)
288 /* Don't limit PMTU discovery. */
289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
292 /* Limit if icmp type is enabled in ratemask. */
293 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
299 static bool icmpv4_global_allow(struct net *net, int type, int code)
301 if (icmpv4_mask_allow(net, type, code))
304 if (icmp_global_allow())
311 * Send an ICMP frame.
314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 struct flowi4 *fl4, int type, int code)
317 struct dst_entry *dst = &rt->dst;
318 struct inet_peer *peer;
322 if (icmpv4_mask_allow(net, type, code))
325 /* No rate limit on loopback */
326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
329 vif = l3mdev_master_ifindex(dst->dev);
330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 rc = inet_peer_xrlim_allow(peer,
332 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
340 * Maintain the counters used in the SNMP statistics for outgoing ICMP
342 void icmp_out_count(struct net *net, unsigned char type)
344 ICMPMSGOUT_INC_STATS(net, type);
345 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
349 * Checksum each fragment, and on the first include the headers and final
352 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
355 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
358 csum = skb_copy_and_csum_bits(icmp_param->skb,
359 icmp_param->offset + offset,
362 skb->csum = csum_block_add(skb->csum, csum, odd);
363 if (icmp_pointers[icmp_param->data.icmph.type].error)
364 nf_ct_attach(skb, icmp_param->skb);
368 static void icmp_push_reply(struct icmp_bxm *icmp_param,
370 struct ipcm_cookie *ipc, struct rtable **rt)
375 sk = icmp_sk(dev_net((*rt)->dst.dev));
376 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
377 icmp_param->data_len+icmp_param->head_len,
378 icmp_param->head_len,
379 ipc, rt, MSG_DONTWAIT) < 0) {
380 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
381 ip_flush_pending_frames(sk);
382 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
383 struct icmphdr *icmph = icmp_hdr(skb);
385 struct sk_buff *skb1;
387 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
389 icmp_param->head_len);
390 skb_queue_walk(&sk->sk_write_queue, skb1) {
391 csum = csum_add(csum, skb1->csum);
393 icmph->checksum = csum_fold(csum);
394 skb->ip_summed = CHECKSUM_NONE;
395 ip_push_pending_frames(sk, fl4);
400 * Driving logic for building and sending ICMP messages.
403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
405 struct ipcm_cookie ipc;
406 struct rtable *rt = skb_rtable(skb);
407 struct net *net = dev_net(rt->dst.dev);
410 struct inet_sock *inet;
412 u32 mark = IP4_REPLY_MARK(net, skb->mark);
413 int type = icmp_param->data.icmph.type;
414 int code = icmp_param->data.icmph.code;
416 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
419 /* Needed by both icmp_global_allow and icmp_xmit_lock */
422 /* global icmp_msgs_per_sec */
423 if (!icmpv4_global_allow(net, type, code))
426 sk = icmp_xmit_lock(net);
431 icmp_param->data.icmph.checksum = 0;
434 inet->tos = ip_hdr(skb)->tos;
435 ipc.sockc.mark = mark;
436 daddr = ipc.addr = ip_hdr(skb)->saddr;
437 saddr = fib_compute_spec_dst(skb);
439 if (icmp_param->replyopts.opt.opt.optlen) {
440 ipc.opt = &icmp_param->replyopts.opt;
441 if (ipc.opt->opt.srr)
442 daddr = icmp_param->replyopts.opt.opt.faddr;
444 memset(&fl4, 0, sizeof(fl4));
447 fl4.flowi4_mark = mark;
448 fl4.flowi4_uid = sock_net_uid(net, NULL);
449 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
450 fl4.flowi4_proto = IPPROTO_ICMP;
451 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
452 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
453 rt = ip_route_output_key(net, &fl4);
456 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
457 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
460 icmp_xmit_unlock(sk);
466 * The device used for looking up which routing table to use for sending an ICMP
467 * error is preferably the source whenever it is set, which should ensure the
468 * icmp error can be sent to the source host, else lookup using the routing
469 * table of the destination device, else use the main routing table (index 0).
471 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
473 struct net_device *route_lookup_dev = NULL;
476 route_lookup_dev = skb->dev;
477 else if (skb_dst(skb))
478 route_lookup_dev = skb_dst(skb)->dev;
479 return route_lookup_dev;
482 static struct rtable *icmp_route_lookup(struct net *net,
484 struct sk_buff *skb_in,
485 const struct iphdr *iph,
486 __be32 saddr, u8 tos, u32 mark,
488 struct icmp_bxm *param)
490 struct net_device *route_lookup_dev;
491 struct rtable *rt, *rt2;
492 struct flowi4 fl4_dec;
495 memset(fl4, 0, sizeof(*fl4));
496 fl4->daddr = (param->replyopts.opt.opt.srr ?
497 param->replyopts.opt.opt.faddr : iph->saddr);
499 fl4->flowi4_mark = mark;
500 fl4->flowi4_uid = sock_net_uid(net, NULL);
501 fl4->flowi4_tos = RT_TOS(tos);
502 fl4->flowi4_proto = IPPROTO_ICMP;
503 fl4->fl4_icmp_type = type;
504 fl4->fl4_icmp_code = code;
505 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
506 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
508 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
509 rt = ip_route_output_key_hash(net, fl4, skb_in);
513 /* No need to clone since we're just using its address. */
516 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
517 flowi4_to_flowi(fl4), NULL, 0);
521 } else if (PTR_ERR(rt) == -EPERM) {
526 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
528 goto relookup_failed;
530 if (inet_addr_type_dev_table(net, route_lookup_dev,
531 fl4_dec.saddr) == RTN_LOCAL) {
532 rt2 = __ip_route_output_key(net, &fl4_dec);
536 struct flowi4 fl4_2 = {};
537 unsigned long orefdst;
539 fl4_2.daddr = fl4_dec.saddr;
540 rt2 = ip_route_output_key(net, &fl4_2);
543 goto relookup_failed;
546 orefdst = skb_in->_skb_refdst; /* save old refdst */
547 skb_dst_set(skb_in, NULL);
548 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
549 RT_TOS(tos), rt2->dst.dev);
551 dst_release(&rt2->dst);
552 rt2 = skb_rtable(skb_in);
553 skb_in->_skb_refdst = orefdst; /* restore old refdst */
557 goto relookup_failed;
559 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
560 flowi4_to_flowi(&fl4_dec), NULL,
563 dst_release(&rt->dst);
564 memcpy(fl4, &fl4_dec, sizeof(*fl4));
566 } else if (PTR_ERR(rt2) == -EPERM) {
568 dst_release(&rt->dst);
572 goto relookup_failed;
583 * Send an ICMP message in response to a situation
585 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
586 * MAY send more (we do).
587 * MUST NOT change this header information.
588 * MUST NOT reply to a multicast/broadcast IP address.
589 * MUST NOT reply to a multicast/broadcast MAC address.
590 * MUST reply to only the first fragment.
593 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
594 const struct ip_options *opt)
598 struct icmp_bxm icmp_param;
599 struct rtable *rt = skb_rtable(skb_in);
600 struct ipcm_cookie ipc;
612 net = dev_net(rt->dst.dev);
613 else if (skb_in->dev)
614 net = dev_net(skb_in->dev);
619 * Find the original header. It is expected to be valid, of course.
620 * Check this, icmp_send is called from the most obscure devices
623 iph = ip_hdr(skb_in);
625 if ((u8 *)iph < skb_in->head ||
626 (skb_network_header(skb_in) + sizeof(*iph)) >
627 skb_tail_pointer(skb_in))
631 * No replies to physical multicast/broadcast
633 if (skb_in->pkt_type != PACKET_HOST)
637 * Now check at the protocol level
639 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
643 * Only reply to fragment 0. We byte re-order the constant
644 * mask for efficiency.
646 if (iph->frag_off & htons(IP_OFFSET))
650 * If we send an ICMP error to an ICMP error a mess would result..
652 if (icmp_pointers[type].error) {
654 * We are an error, check if we are replying to an
657 if (iph->protocol == IPPROTO_ICMP) {
658 u8 _inner_type, *itp;
660 itp = skb_header_pointer(skb_in,
661 skb_network_header(skb_in) +
663 offsetof(struct icmphdr,
672 * Assume any unknown ICMP type is an error. This
673 * isn't specified by the RFC, but think about it..
675 if (*itp > NR_ICMP_TYPES ||
676 icmp_pointers[*itp].error)
681 /* Needed by both icmp_global_allow and icmp_xmit_lock */
684 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
685 * incoming dev is loopback. If outgoing dev change to not be
686 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
688 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
689 !icmpv4_global_allow(net, type, code))
692 sk = icmp_xmit_lock(net);
697 * Construct source address and options.
701 if (!(rt->rt_flags & RTCF_LOCAL)) {
702 struct net_device *dev = NULL;
705 if (rt_is_input_route(rt) &&
706 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
707 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
710 saddr = inet_select_addr(dev, iph->saddr,
717 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
718 IPTOS_PREC_INTERNETCONTROL) :
720 mark = IP4_REPLY_MARK(net, skb_in->mark);
722 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
727 * Prepare data for ICMP header.
730 icmp_param.data.icmph.type = type;
731 icmp_param.data.icmph.code = code;
732 icmp_param.data.icmph.un.gateway = info;
733 icmp_param.data.icmph.checksum = 0;
734 icmp_param.skb = skb_in;
735 icmp_param.offset = skb_network_offset(skb_in);
736 inet_sk(sk)->tos = tos;
738 ipc.addr = iph->saddr;
739 ipc.opt = &icmp_param.replyopts.opt;
740 ipc.sockc.mark = mark;
742 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
743 type, code, &icmp_param);
747 /* peer icmp_ratelimit */
748 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
751 /* RFC says return as much as we can without exceeding 576 bytes. */
753 room = dst_mtu(&rt->dst);
756 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
757 room -= sizeof(struct icmphdr);
758 /* Guard against tiny mtu. We need to include at least one
759 * IP network header for this message to make any sense.
761 if (room <= (int)sizeof(struct iphdr))
764 icmp_param.data_len = skb_in->len - icmp_param.offset;
765 if (icmp_param.data_len > room)
766 icmp_param.data_len = room;
767 icmp_param.head_len = sizeof(struct icmphdr);
769 /* if we don't have a source address at this point, fall back to the
770 * dummy address instead of sending out a packet with a source address
774 fl4.saddr = htonl(INADDR_DUMMY);
776 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
780 icmp_xmit_unlock(sk);
785 EXPORT_SYMBOL(__icmp_send);
787 #if IS_ENABLED(CONFIG_NF_NAT)
788 #include <net/netfilter/nf_conntrack.h>
789 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
791 struct sk_buff *cloned_skb = NULL;
792 struct ip_options opts = { 0 };
793 enum ip_conntrack_info ctinfo;
797 ct = nf_ct_get(skb_in, &ctinfo);
798 if (!ct || !(ct->status & IPS_SRC_NAT)) {
799 __icmp_send(skb_in, type, code, info, &opts);
803 if (skb_shared(skb_in))
804 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
806 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
807 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
808 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
809 skb_network_offset(skb_in) + sizeof(struct iphdr))))
812 orig_ip = ip_hdr(skb_in)->saddr;
813 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
814 __icmp_send(skb_in, type, code, info, &opts);
815 ip_hdr(skb_in)->saddr = orig_ip;
817 consume_skb(cloned_skb);
819 EXPORT_SYMBOL(icmp_ndo_send);
822 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
824 const struct iphdr *iph = (const struct iphdr *)skb->data;
825 const struct net_protocol *ipprot;
826 int protocol = iph->protocol;
828 /* Checkin full IP header plus 8 bytes of protocol to
829 * avoid additional coding at protocol handlers.
831 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
832 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
836 raw_icmp_error(skb, protocol, info);
838 ipprot = rcu_dereference(inet_protos[protocol]);
839 if (ipprot && ipprot->err_handler)
840 ipprot->err_handler(skb, info);
843 static bool icmp_tag_validation(int proto)
848 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
854 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
855 * ICMP_PARAMETERPROB.
858 static bool icmp_unreach(struct sk_buff *skb)
860 const struct iphdr *iph;
861 struct icmphdr *icmph;
865 net = dev_net(skb_dst(skb)->dev);
868 * Incomplete header ?
869 * Only checks for the IP header, there should be an
870 * additional check for longer headers in upper levels.
873 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
876 icmph = icmp_hdr(skb);
877 iph = (const struct iphdr *)skb->data;
879 if (iph->ihl < 5) /* Mangled header, drop. */
882 switch (icmph->type) {
883 case ICMP_DEST_UNREACH:
884 switch (icmph->code & 15) {
885 case ICMP_NET_UNREACH:
886 case ICMP_HOST_UNREACH:
887 case ICMP_PROT_UNREACH:
888 case ICMP_PORT_UNREACH:
890 case ICMP_FRAG_NEEDED:
891 /* for documentation of the ip_no_pmtu_disc
893 * Documentation/networking/ip-sysctl.rst
895 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
897 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
903 if (!icmp_tag_validation(iph->protocol))
907 info = ntohs(icmph->un.frag.mtu);
911 net_dbg_ratelimited("%pI4: Source Route Failed\n",
917 if (icmph->code > NR_ICMP_UNREACH)
920 case ICMP_PARAMETERPROB:
921 info = ntohl(icmph->un.gateway) >> 24;
923 case ICMP_TIME_EXCEEDED:
924 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
925 if (icmph->code == ICMP_EXC_FRAGTIME)
931 * Throw it at our lower layers
933 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
935 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
937 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
942 * Check the other end isn't violating RFC 1122. Some routers send
943 * bogus responses to broadcast frames. If you see this message
944 * first check your netmask matches at both ends, if it does then
945 * get the other vendor to fix their kit.
948 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
949 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
950 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
952 icmph->type, icmph->code,
953 &iph->daddr, skb->dev->name);
957 icmp_socket_deliver(skb, info);
962 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
968 * Handle ICMP_REDIRECT.
971 static bool icmp_redirect(struct sk_buff *skb)
973 if (skb->len < sizeof(struct iphdr)) {
974 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
978 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
979 /* there aught to be a stat */
983 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
988 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
990 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
992 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
993 * included in the reply.
994 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
995 * echo requests, MUST have default=NOT.
996 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
997 * Extended Echo Functionality, MUST be disabled by default
998 * See also WRT handling of options once they are done and working.
1001 static bool icmp_echo(struct sk_buff *skb)
1003 struct icmp_bxm icmp_param;
1006 net = dev_net(skb_dst(skb)->dev);
1007 /* should there be an ICMP stat for ignored echos? */
1008 if (net->ipv4.sysctl_icmp_echo_ignore_all)
1011 icmp_param.data.icmph = *icmp_hdr(skb);
1012 icmp_param.skb = skb;
1013 icmp_param.offset = 0;
1014 icmp_param.data_len = skb->len;
1015 icmp_param.head_len = sizeof(struct icmphdr);
1017 if (icmp_param.data.icmph.type == ICMP_ECHO)
1018 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1019 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1022 icmp_reply(&icmp_param, skb);
1026 /* Helper for icmp_echo and icmpv6_echo_reply.
1027 * Searches for net_device that matches PROBE interface identifier
1028 * and builds PROBE reply message in icmphdr.
1030 * Returns false if PROBE responses are disabled via sysctl
1033 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1035 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1036 struct icmp_ext_echo_iio *iio, _iio;
1037 struct net *net = dev_net(skb->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;
1121 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1122 status |= ICMP_EXT_ECHOREPLY_IPV4;
1123 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1124 status |= ICMP_EXT_ECHOREPLY_IPV6;
1126 icmphdr->un.echo.sequence |= htons(status);
1129 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1132 EXPORT_SYMBOL_GPL(icmp_build_probe);
1135 * Handle ICMP Timestamp requests.
1136 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1137 * SHOULD be in the kernel for minimum random latency.
1138 * MUST be accurate to a few minutes.
1139 * MUST be updated at least at 15Hz.
1141 static bool icmp_timestamp(struct sk_buff *skb)
1143 struct icmp_bxm icmp_param;
1151 * Fill in the current time as ms since midnight UT:
1153 icmp_param.data.times[1] = inet_current_timestamp();
1154 icmp_param.data.times[2] = icmp_param.data.times[1];
1156 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1158 icmp_param.data.icmph = *icmp_hdr(skb);
1159 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1160 icmp_param.data.icmph.code = 0;
1161 icmp_param.skb = skb;
1162 icmp_param.offset = 0;
1163 icmp_param.data_len = 0;
1164 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1165 icmp_reply(&icmp_param, skb);
1169 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1173 static bool icmp_discard(struct sk_buff *skb)
1175 /* pretend it was a success */
1180 * Deal with incoming ICMP packets.
1182 int icmp_rcv(struct sk_buff *skb)
1184 struct icmphdr *icmph;
1185 struct rtable *rt = skb_rtable(skb);
1186 struct net *net = dev_net(rt->dst.dev);
1189 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1190 struct sec_path *sp = skb_sec_path(skb);
1193 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1197 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1200 nh = skb_network_offset(skb);
1201 skb_set_network_header(skb, sizeof(*icmph));
1203 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1206 skb_set_network_header(skb, nh);
1209 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1211 if (skb_checksum_simple_validate(skb))
1214 if (!pskb_pull(skb, sizeof(*icmph)))
1217 icmph = icmp_hdr(skb);
1219 ICMPMSGIN_INC_STATS(net, icmph->type);
1221 /* Check for ICMP Extended Echo (PROBE) messages */
1222 if (icmph->type == ICMP_EXT_ECHO) {
1223 /* We can't use icmp_pointers[].handler() because it is an array of
1224 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1226 success = icmp_echo(skb);
1230 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1231 success = ping_rcv(skb);
1236 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1238 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1241 if (icmph->type > NR_ICMP_TYPES)
1245 * Parse the ICMP message
1248 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1250 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1251 * silently ignored (we let user decide with a sysctl).
1252 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1253 * discarded if to broadcast/multicast.
1255 if ((icmph->type == ICMP_ECHO ||
1256 icmph->type == ICMP_TIMESTAMP) &&
1257 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1260 if (icmph->type != ICMP_ECHO &&
1261 icmph->type != ICMP_TIMESTAMP &&
1262 icmph->type != ICMP_ADDRESS &&
1263 icmph->type != ICMP_ADDRESSREPLY) {
1268 success = icmp_pointers[icmph->type].handler(skb);
1272 return NET_RX_SUCCESS;
1279 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1281 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1285 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1287 struct icmp_extobj_hdr *objh, _objh;
1288 struct icmp_ext_hdr *exth, _exth;
1291 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1294 if (exth->version != 2)
1297 if (exth->checksum &&
1298 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1301 off += sizeof(_exth);
1302 while (off < skb->len) {
1303 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1307 olen = ntohs(objh->length);
1308 if (olen < sizeof(_objh))
1319 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1320 struct sock_ee_data_rfc4884 *out,
1325 /* original datagram headers: end of icmph to payload (skb->data) */
1326 hlen = -skb_transport_offset(skb) - thlen;
1328 /* per rfc 4884: minimal datagram length of 128 bytes */
1329 if (off < 128 || off < hlen)
1332 /* kernel has stripped headers: return payload offset in bytes */
1334 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1339 if (!ip_icmp_error_rfc4884_validate(skb, off))
1340 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1342 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1344 int icmp_err(struct sk_buff *skb, u32 info)
1346 struct iphdr *iph = (struct iphdr *)skb->data;
1347 int offset = iph->ihl<<2;
1348 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1349 int type = icmp_hdr(skb)->type;
1350 int code = icmp_hdr(skb)->code;
1351 struct net *net = dev_net(skb->dev);
1354 * Use ping_err to handle all icmp errors except those
1355 * triggered by ICMP_ECHOREPLY which sent from kernel.
1357 if (icmph->type != ICMP_ECHOREPLY) {
1358 ping_err(skb, offset, info);
1362 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1363 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1364 else if (type == ICMP_REDIRECT)
1365 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1371 * This table is the definition of how we handle ICMP.
1373 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1374 [ICMP_ECHOREPLY] = {
1375 .handler = ping_rcv,
1378 .handler = icmp_discard,
1382 .handler = icmp_discard,
1385 [ICMP_DEST_UNREACH] = {
1386 .handler = icmp_unreach,
1389 [ICMP_SOURCE_QUENCH] = {
1390 .handler = icmp_unreach,
1394 .handler = icmp_redirect,
1398 .handler = icmp_discard,
1402 .handler = icmp_discard,
1406 .handler = icmp_echo,
1409 .handler = icmp_discard,
1413 .handler = icmp_discard,
1416 [ICMP_TIME_EXCEEDED] = {
1417 .handler = icmp_unreach,
1420 [ICMP_PARAMETERPROB] = {
1421 .handler = icmp_unreach,
1424 [ICMP_TIMESTAMP] = {
1425 .handler = icmp_timestamp,
1427 [ICMP_TIMESTAMPREPLY] = {
1428 .handler = icmp_discard,
1430 [ICMP_INFO_REQUEST] = {
1431 .handler = icmp_discard,
1433 [ICMP_INFO_REPLY] = {
1434 .handler = icmp_discard,
1437 .handler = icmp_discard,
1439 [ICMP_ADDRESSREPLY] = {
1440 .handler = icmp_discard,
1444 static void __net_exit icmp_sk_exit(struct net *net)
1448 for_each_possible_cpu(i)
1449 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1450 free_percpu(net->ipv4.icmp_sk);
1451 net->ipv4.icmp_sk = NULL;
1454 static int __net_init icmp_sk_init(struct net *net)
1458 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1459 if (!net->ipv4.icmp_sk)
1462 for_each_possible_cpu(i) {
1465 err = inet_ctl_sock_create(&sk, PF_INET,
1466 SOCK_RAW, IPPROTO_ICMP, net);
1470 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1472 /* Enough space for 2 64K ICMP packets, including
1473 * sk_buff/skb_shared_info struct overhead.
1475 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1478 * Speedup sock_wfree()
1480 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1481 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1484 /* Control parameters for ECHO replies. */
1485 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1486 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1487 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1489 /* Control parameter - ignore bogus broadcast responses? */
1490 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1493 * Configurable global rate limit.
1495 * ratelimit defines tokens/packet consumed for dst->rate_token
1496 * bucket ratemask defines which icmp types are ratelimited by
1497 * setting it's bit position.
1500 * dest unreachable (3), source quench (4),
1501 * time exceeded (11), parameter problem (12)
1504 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1505 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1506 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1515 static struct pernet_operations __net_initdata icmp_sk_ops = {
1516 .init = icmp_sk_init,
1517 .exit = icmp_sk_exit,
1520 int __init icmp_init(void)
1522 return register_pernet_subsys(&icmp_sk_ops);