2 * NET3: Implementation of the ICMP protocol layer.
4 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 * Some of the function names and the icmp unreach table for this
12 * module were derived from [icmp.c 1.0.11 06/02/93] by
13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14 * Other than that this module is a complete rewrite.
17 * Clemens Fruhwirth : introduce global icmp rate limiting
18 * with icmp type masking ability instead
19 * of broken per type icmp timeouts.
20 * Mike Shaver : RFC1122 checks.
21 * Alan Cox : Multicast ping reply as self.
22 * Alan Cox : Fix atomicity lockup in ip_build_xmit
24 * Alan Cox : Added 216,128 byte paths to the MTU
26 * Martin Mares : RFC1812 checks.
27 * Martin Mares : Can be configured to follow redirects
28 * if acting as a router _without_ a
29 * routing protocol (RFC 1812).
30 * Martin Mares : Echo requests may be configured to
31 * be ignored (RFC 1812).
32 * Martin Mares : Limitation of ICMP error message
33 * transmit rate (RFC 1812).
34 * Martin Mares : TOS and Precedence set correctly
36 * Martin Mares : Now copying as much data from the
37 * original packet as we can without
38 * exceeding 576 bytes (RFC 1812).
39 * Willy Konynenberg : Transparent proxying support.
40 * Keith Owens : RFC1191 correction for 4.2BSD based
42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
44 * Andi Kleen : Check all packet lengths properly
45 * and moved all kfree_skb() up to
47 * Andi Kleen : Move the rate limit bookkeeping
48 * into the dest entry and use a token
49 * bucket filter (thanks to ANK). Make
50 * the rates sysctl configurable.
51 * Yu Tianli : Fixed two ugly bugs in icmp_send
52 * - IP option length was accounted wrongly
53 * - ICMP header length was not accounted
55 * Tristan Greaves : Added sysctl option to ignore bogus
56 * broadcast responses from broken routers.
60 * - Should use skb_pull() instead of all the manual checking.
61 * This would also greatly simply some upper layer error handlers. --AK
65 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
67 #include <linux/module.h>
68 #include <linux/types.h>
69 #include <linux/jiffies.h>
70 #include <linux/kernel.h>
71 #include <linux/fcntl.h>
72 #include <linux/socket.h>
74 #include <linux/inet.h>
75 #include <linux/inetdevice.h>
76 #include <linux/netdevice.h>
77 #include <linux/string.h>
78 #include <linux/netfilter_ipv4.h>
79 #include <linux/slab.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
89 #include <linux/skbuff.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/init.h>
94 #include <asm/uaccess.h>
95 #include <net/checksum.h>
97 #include <net/inet_common.h>
98 #include <net/ip_fib.h>
99 #include <net/l3mdev.h>
102 * Build xmit assembly blocks
111 struct icmphdr icmph;
115 struct ip_options_data replyopts;
118 /* An array of errno for error messages from dest unreach. */
119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
121 const struct icmp_err icmp_err_convert[] = {
123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
155 .errno = ENONET, /* ICMP_HOST_ISOLATED */
159 .errno = ENETUNREACH, /* ICMP_NET_ANO */
163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
187 EXPORT_SYMBOL(icmp_err_convert);
190 * ICMP control array. This specifies what to do with each ICMP.
193 struct icmp_control {
194 bool (*handler)(struct sk_buff *skb);
195 short error; /* This ICMP is classed as an error message */
198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
201 * The ICMP socket(s). This is the most convenient way to flow control
202 * our ICMP output as well as maintain a clean interface throughout
203 * all layers. All Socketless IP sends will soon be gone.
205 * On SMP we have one ICMP socket per-cpu.
207 static struct sock *icmp_sk(struct net *net)
209 return *this_cpu_ptr(net->ipv4.icmp_sk);
212 static inline struct sock *icmp_xmit_lock(struct net *net)
220 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
221 /* This can happen if the output path signals a
222 * dst_link_failure() for an outgoing ICMP packet.
230 static inline void icmp_xmit_unlock(struct sock *sk)
232 spin_unlock_bh(&sk->sk_lock.slock);
235 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
236 int sysctl_icmp_msgs_burst __read_mostly = 50;
243 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
247 * icmp_global_allow - Are we allowed to send one more ICMP message ?
249 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
250 * Returns false if we reached the limit and can not send another packet.
251 * Note: called with BH disabled
253 bool icmp_global_allow(void)
255 u32 credit, delta, incr = 0, now = (u32)jiffies;
258 /* Check if token bucket is empty and cannot be refilled
259 * without taking the spinlock. The READ_ONCE() are paired
260 * with the following WRITE_ONCE() in this same function.
262 if (!READ_ONCE(icmp_global.credit)) {
263 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
268 spin_lock(&icmp_global.lock);
269 delta = min_t(u32, now - icmp_global.stamp, HZ);
270 if (delta >= HZ / 50) {
271 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
273 WRITE_ONCE(icmp_global.stamp, now);
275 credit = min_t(u32, icmp_global.credit + incr,
276 READ_ONCE(sysctl_icmp_msgs_burst));
278 /* We want to use a credit of one in average, but need to randomize
279 * it for security reasons.
281 credit = max_t(int, credit - prandom_u32_max(3), 0);
284 WRITE_ONCE(icmp_global.credit, credit);
285 spin_unlock(&icmp_global.lock);
288 EXPORT_SYMBOL(icmp_global_allow);
291 * Send an ICMP frame.
294 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
295 struct flowi4 *fl4, int type, int code)
297 struct dst_entry *dst = &rt->dst;
300 if (type > NR_ICMP_TYPES)
303 /* Don't limit PMTU discovery. */
304 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
307 /* No rate limit on loopback */
308 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
311 /* Limit if icmp type is enabled in ratemask. */
312 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
316 if (icmp_global_allow()) {
317 int vif = l3mdev_master_ifindex(dst->dev);
318 struct inet_peer *peer;
320 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
321 rc = inet_peer_xrlim_allow(peer,
322 net->ipv4.sysctl_icmp_ratelimit);
331 * Maintain the counters used in the SNMP statistics for outgoing ICMP
333 void icmp_out_count(struct net *net, unsigned char type)
335 ICMPMSGOUT_INC_STATS(net, type);
336 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
340 * Checksum each fragment, and on the first include the headers and final
343 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
346 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
349 csum = skb_copy_and_csum_bits(icmp_param->skb,
350 icmp_param->offset + offset,
353 skb->csum = csum_block_add(skb->csum, csum, odd);
354 if (icmp_pointers[icmp_param->data.icmph.type].error)
355 nf_ct_attach(skb, icmp_param->skb);
359 static void icmp_push_reply(struct icmp_bxm *icmp_param,
361 struct ipcm_cookie *ipc, struct rtable **rt)
366 sk = icmp_sk(dev_net((*rt)->dst.dev));
367 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
368 icmp_param->data_len+icmp_param->head_len,
369 icmp_param->head_len,
370 ipc, rt, MSG_DONTWAIT) < 0) {
371 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
372 ip_flush_pending_frames(sk);
373 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
374 struct icmphdr *icmph = icmp_hdr(skb);
376 struct sk_buff *skb1;
378 skb_queue_walk(&sk->sk_write_queue, skb1) {
379 csum = csum_add(csum, skb1->csum);
381 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
383 icmp_param->head_len, csum);
384 icmph->checksum = csum_fold(csum);
385 skb->ip_summed = CHECKSUM_NONE;
386 ip_push_pending_frames(sk, fl4);
391 * Driving logic for building and sending ICMP messages.
394 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
396 struct ipcm_cookie ipc;
397 struct rtable *rt = skb_rtable(skb);
398 struct net *net = dev_net(rt->dst.dev);
401 struct inet_sock *inet;
403 u32 mark = IP4_REPLY_MARK(net, skb->mark);
405 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
408 sk = icmp_xmit_lock(net);
413 icmp_param->data.icmph.checksum = 0;
415 inet->tos = ip_hdr(skb)->tos;
417 daddr = ipc.addr = ip_hdr(skb)->saddr;
418 saddr = fib_compute_spec_dst(skb);
424 if (icmp_param->replyopts.opt.opt.optlen) {
425 ipc.opt = &icmp_param->replyopts.opt;
426 if (ipc.opt->opt.srr)
427 daddr = icmp_param->replyopts.opt.opt.faddr;
429 memset(&fl4, 0, sizeof(fl4));
432 fl4.flowi4_mark = mark;
433 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
434 fl4.flowi4_proto = IPPROTO_ICMP;
435 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
436 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
437 rt = ip_route_output_key(net, &fl4);
440 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
441 icmp_param->data.icmph.code))
442 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
445 icmp_xmit_unlock(sk);
448 #ifdef CONFIG_IP_ROUTE_MULTIPATH
450 /* Source and destination is swapped. See ip_multipath_icmp_hash */
451 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
453 const struct iphdr *iph = ip_hdr(skb);
455 return fib_multipath_hash(iph->daddr, iph->saddr);
460 #define icmp_multipath_hash_skb(skb) (-1)
465 * The device used for looking up which routing table to use for sending an ICMP
466 * error is preferably the source whenever it is set, which should ensure the
467 * icmp error can be sent to the source host, else lookup using the routing
468 * table of the destination device, else use the main routing table (index 0).
470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
472 struct net_device *route_lookup_dev = NULL;
475 route_lookup_dev = skb->dev;
476 else if (skb_dst(skb))
477 route_lookup_dev = skb_dst(skb)->dev;
478 return route_lookup_dev;
481 static struct rtable *icmp_route_lookup(struct net *net,
483 struct sk_buff *skb_in,
484 const struct iphdr *iph,
485 __be32 saddr, u8 tos, u32 mark,
487 struct icmp_bxm *param)
489 struct net_device *route_lookup_dev;
490 struct rtable *rt, *rt2;
491 struct flowi4 fl4_dec;
494 memset(fl4, 0, sizeof(*fl4));
495 fl4->daddr = (param->replyopts.opt.opt.srr ?
496 param->replyopts.opt.opt.faddr : iph->saddr);
498 fl4->flowi4_mark = mark;
499 fl4->flowi4_tos = RT_TOS(tos);
500 fl4->flowi4_proto = IPPROTO_ICMP;
501 fl4->fl4_icmp_type = type;
502 fl4->fl4_icmp_code = code;
503 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
504 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
506 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
507 rt = __ip_route_output_key_hash(net, fl4,
508 icmp_multipath_hash_skb(skb_in));
512 /* No need to clone since we're just using its address. */
515 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
516 flowi4_to_flowi(fl4), NULL, 0);
520 } else if (PTR_ERR(rt) == -EPERM) {
525 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
527 goto relookup_failed;
529 if (inet_addr_type_dev_table(net, route_lookup_dev,
530 fl4_dec.saddr) == RTN_LOCAL) {
531 rt2 = __ip_route_output_key(net, &fl4_dec);
535 struct flowi4 fl4_2 = {};
536 unsigned long orefdst;
538 fl4_2.daddr = fl4_dec.saddr;
539 rt2 = ip_route_output_key(net, &fl4_2);
542 goto relookup_failed;
545 orefdst = skb_in->_skb_refdst; /* save old refdst */
546 skb_dst_set(skb_in, NULL);
547 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
548 RT_TOS(tos), rt2->dst.dev);
550 dst_release(&rt2->dst);
551 rt2 = skb_rtable(skb_in);
552 skb_in->_skb_refdst = orefdst; /* restore old refdst */
556 goto relookup_failed;
558 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
559 flowi4_to_flowi(&fl4_dec), NULL,
562 dst_release(&rt->dst);
563 memcpy(fl4, &fl4_dec, sizeof(*fl4));
565 } else if (PTR_ERR(rt2) == -EPERM) {
567 dst_release(&rt->dst);
571 goto relookup_failed;
582 * Send an ICMP message in response to a situation
584 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
585 * MAY send more (we do).
586 * MUST NOT change this header information.
587 * MUST NOT reply to a multicast/broadcast IP address.
588 * MUST NOT reply to a multicast/broadcast MAC address.
589 * MUST reply to only the first fragment.
592 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
593 const struct ip_options *opt)
597 struct icmp_bxm *icmp_param;
598 struct rtable *rt = skb_rtable(skb_in);
599 struct ipcm_cookie ipc;
609 net = dev_net(rt->dst.dev);
612 * Find the original header. It is expected to be valid, of course.
613 * Check this, icmp_send is called from the most obscure devices
616 iph = ip_hdr(skb_in);
618 if ((u8 *)iph < skb_in->head ||
619 (skb_network_header(skb_in) + sizeof(*iph)) >
620 skb_tail_pointer(skb_in))
624 * No replies to physical multicast/broadcast
626 if (skb_in->pkt_type != PACKET_HOST)
630 * Now check at the protocol level
632 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
636 * Only reply to fragment 0. We byte re-order the constant
637 * mask for efficiency.
639 if (iph->frag_off & htons(IP_OFFSET))
643 * If we send an ICMP error to an ICMP error a mess would result..
645 if (icmp_pointers[type].error) {
647 * We are an error, check if we are replying to an
650 if (iph->protocol == IPPROTO_ICMP) {
651 u8 _inner_type, *itp;
653 itp = skb_header_pointer(skb_in,
654 skb_network_header(skb_in) +
656 offsetof(struct icmphdr,
665 * Assume any unknown ICMP type is an error. This
666 * isn't specified by the RFC, but think about it..
668 if (*itp > NR_ICMP_TYPES ||
669 icmp_pointers[*itp].error)
674 icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
678 sk = icmp_xmit_lock(net);
683 * Construct source address and options.
687 if (!(rt->rt_flags & RTCF_LOCAL)) {
688 struct net_device *dev = NULL;
691 if (rt_is_input_route(rt) &&
692 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
693 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
696 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
702 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
703 IPTOS_PREC_INTERNETCONTROL) :
705 mark = IP4_REPLY_MARK(net, skb_in->mark);
707 if (__ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in, opt))
712 * Prepare data for ICMP header.
715 icmp_param->data.icmph.type = type;
716 icmp_param->data.icmph.code = code;
717 icmp_param->data.icmph.un.gateway = info;
718 icmp_param->data.icmph.checksum = 0;
719 icmp_param->skb = skb_in;
720 icmp_param->offset = skb_network_offset(skb_in);
721 inet_sk(sk)->tos = tos;
723 ipc.addr = iph->saddr;
724 ipc.opt = &icmp_param->replyopts.opt;
729 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
730 type, code, icmp_param);
734 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
737 /* RFC says return as much as we can without exceeding 576 bytes. */
739 room = dst_mtu(&rt->dst);
742 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
743 room -= sizeof(struct icmphdr);
745 icmp_param->data_len = skb_in->len - icmp_param->offset;
746 if (icmp_param->data_len > room)
747 icmp_param->data_len = room;
748 icmp_param->head_len = sizeof(struct icmphdr);
750 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
754 icmp_xmit_unlock(sk);
759 EXPORT_SYMBOL(__icmp_send);
761 #if IS_ENABLED(CONFIG_NF_NAT)
762 #include <net/netfilter/nf_conntrack.h>
763 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
765 struct sk_buff *cloned_skb = NULL;
766 struct ip_options opts = { 0 };
767 enum ip_conntrack_info ctinfo;
771 ct = nf_ct_get(skb_in, &ctinfo);
772 if (!ct || !(ct->status & IPS_SRC_NAT)) {
773 __icmp_send(skb_in, type, code, info, &opts);
777 if (skb_shared(skb_in))
778 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
780 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
781 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
782 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
783 skb_network_offset(skb_in) + sizeof(struct iphdr))))
786 orig_ip = ip_hdr(skb_in)->saddr;
787 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
788 __icmp_send(skb_in, type, code, info, &opts);
789 ip_hdr(skb_in)->saddr = orig_ip;
791 consume_skb(cloned_skb);
793 EXPORT_SYMBOL(icmp_ndo_send);
796 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
798 const struct iphdr *iph = (const struct iphdr *) skb->data;
799 const struct net_protocol *ipprot;
800 int protocol = iph->protocol;
802 /* Checkin full IP header plus 8 bytes of protocol to
803 * avoid additional coding at protocol handlers.
805 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
806 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
810 raw_icmp_error(skb, protocol, info);
812 ipprot = rcu_dereference(inet_protos[protocol]);
813 if (ipprot && ipprot->err_handler)
814 ipprot->err_handler(skb, info);
817 static bool icmp_tag_validation(int proto)
822 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
828 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
829 * ICMP_PARAMETERPROB.
832 static bool icmp_unreach(struct sk_buff *skb)
834 const struct iphdr *iph;
835 struct icmphdr *icmph;
839 net = dev_net(skb_dst(skb)->dev);
842 * Incomplete header ?
843 * Only checks for the IP header, there should be an
844 * additional check for longer headers in upper levels.
847 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
850 icmph = icmp_hdr(skb);
851 iph = (const struct iphdr *)skb->data;
853 if (iph->ihl < 5) /* Mangled header, drop. */
856 switch (icmph->type) {
857 case ICMP_DEST_UNREACH:
858 switch (icmph->code & 15) {
859 case ICMP_NET_UNREACH:
860 case ICMP_HOST_UNREACH:
861 case ICMP_PROT_UNREACH:
862 case ICMP_PORT_UNREACH:
864 case ICMP_FRAG_NEEDED:
865 /* for documentation of the ip_no_pmtu_disc
867 * Documentation/networking/ip-sysctl.txt
869 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
871 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
877 if (!icmp_tag_validation(iph->protocol))
881 info = ntohs(icmph->un.frag.mtu);
885 net_dbg_ratelimited("%pI4: Source Route Failed\n",
891 if (icmph->code > NR_ICMP_UNREACH)
894 case ICMP_PARAMETERPROB:
895 info = ntohl(icmph->un.gateway) >> 24;
897 case ICMP_TIME_EXCEEDED:
898 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
899 if (icmph->code == ICMP_EXC_FRAGTIME)
905 * Throw it at our lower layers
907 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
909 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
911 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
916 * Check the other end isn't violating RFC 1122. Some routers send
917 * bogus responses to broadcast frames. If you see this message
918 * first check your netmask matches at both ends, if it does then
919 * get the other vendor to fix their kit.
922 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
923 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
924 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
926 icmph->type, icmph->code,
927 &iph->daddr, skb->dev->name);
931 icmp_socket_deliver(skb, info);
936 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
942 * Handle ICMP_REDIRECT.
945 static bool icmp_redirect(struct sk_buff *skb)
947 if (skb->len < sizeof(struct iphdr)) {
948 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
952 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
953 /* there aught to be a stat */
957 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
962 * Handle ICMP_ECHO ("ping") requests.
964 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
966 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
967 * included in the reply.
968 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
969 * echo requests, MUST have default=NOT.
970 * See also WRT handling of options once they are done and working.
973 static bool icmp_echo(struct sk_buff *skb)
977 net = dev_net(skb_dst(skb)->dev);
978 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
979 struct icmp_bxm icmp_param;
981 icmp_param.data.icmph = *icmp_hdr(skb);
982 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
983 icmp_param.skb = skb;
984 icmp_param.offset = 0;
985 icmp_param.data_len = skb->len;
986 icmp_param.head_len = sizeof(struct icmphdr);
987 icmp_reply(&icmp_param, skb);
989 /* should there be an ICMP stat for ignored echos? */
994 * Handle ICMP Timestamp requests.
995 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
996 * SHOULD be in the kernel for minimum random latency.
997 * MUST be accurate to a few minutes.
998 * MUST be updated at least at 15Hz.
1000 static bool icmp_timestamp(struct sk_buff *skb)
1002 struct icmp_bxm icmp_param;
1010 * Fill in the current time as ms since midnight UT:
1012 icmp_param.data.times[1] = inet_current_timestamp();
1013 icmp_param.data.times[2] = icmp_param.data.times[1];
1014 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
1016 icmp_param.data.icmph = *icmp_hdr(skb);
1017 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1018 icmp_param.data.icmph.code = 0;
1019 icmp_param.skb = skb;
1020 icmp_param.offset = 0;
1021 icmp_param.data_len = 0;
1022 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1023 icmp_reply(&icmp_param, skb);
1027 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1031 static bool icmp_discard(struct sk_buff *skb)
1033 /* pretend it was a success */
1038 * Deal with incoming ICMP packets.
1040 int icmp_rcv(struct sk_buff *skb)
1042 struct icmphdr *icmph;
1043 struct rtable *rt = skb_rtable(skb);
1044 struct net *net = dev_net(rt->dst.dev);
1047 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1048 struct sec_path *sp = skb_sec_path(skb);
1051 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1055 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1058 nh = skb_network_offset(skb);
1059 skb_set_network_header(skb, sizeof(*icmph));
1061 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1064 skb_set_network_header(skb, nh);
1067 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1069 if (skb_checksum_simple_validate(skb))
1072 if (!pskb_pull(skb, sizeof(*icmph)))
1075 icmph = icmp_hdr(skb);
1077 ICMPMSGIN_INC_STATS(net, icmph->type);
1079 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1081 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1084 if (icmph->type > NR_ICMP_TYPES)
1089 * Parse the ICMP message
1092 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1094 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1095 * silently ignored (we let user decide with a sysctl).
1096 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1097 * discarded if to broadcast/multicast.
1099 if ((icmph->type == ICMP_ECHO ||
1100 icmph->type == ICMP_TIMESTAMP) &&
1101 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1104 if (icmph->type != ICMP_ECHO &&
1105 icmph->type != ICMP_TIMESTAMP &&
1106 icmph->type != ICMP_ADDRESS &&
1107 icmph->type != ICMP_ADDRESSREPLY) {
1112 success = icmp_pointers[icmph->type].handler(skb);
1123 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1125 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1129 void icmp_err(struct sk_buff *skb, u32 info)
1131 struct iphdr *iph = (struct iphdr *)skb->data;
1132 int offset = iph->ihl<<2;
1133 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1134 int type = icmp_hdr(skb)->type;
1135 int code = icmp_hdr(skb)->code;
1136 struct net *net = dev_net(skb->dev);
1139 * Use ping_err to handle all icmp errors except those
1140 * triggered by ICMP_ECHOREPLY which sent from kernel.
1142 if (icmph->type != ICMP_ECHOREPLY) {
1143 ping_err(skb, offset, info);
1147 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1148 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1149 else if (type == ICMP_REDIRECT)
1150 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1154 * This table is the definition of how we handle ICMP.
1156 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1157 [ICMP_ECHOREPLY] = {
1158 .handler = ping_rcv,
1161 .handler = icmp_discard,
1165 .handler = icmp_discard,
1168 [ICMP_DEST_UNREACH] = {
1169 .handler = icmp_unreach,
1172 [ICMP_SOURCE_QUENCH] = {
1173 .handler = icmp_unreach,
1177 .handler = icmp_redirect,
1181 .handler = icmp_discard,
1185 .handler = icmp_discard,
1189 .handler = icmp_echo,
1192 .handler = icmp_discard,
1196 .handler = icmp_discard,
1199 [ICMP_TIME_EXCEEDED] = {
1200 .handler = icmp_unreach,
1203 [ICMP_PARAMETERPROB] = {
1204 .handler = icmp_unreach,
1207 [ICMP_TIMESTAMP] = {
1208 .handler = icmp_timestamp,
1210 [ICMP_TIMESTAMPREPLY] = {
1211 .handler = icmp_discard,
1213 [ICMP_INFO_REQUEST] = {
1214 .handler = icmp_discard,
1216 [ICMP_INFO_REPLY] = {
1217 .handler = icmp_discard,
1220 .handler = icmp_discard,
1222 [ICMP_ADDRESSREPLY] = {
1223 .handler = icmp_discard,
1227 static void __net_exit icmp_sk_exit(struct net *net)
1231 for_each_possible_cpu(i)
1232 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1233 free_percpu(net->ipv4.icmp_sk);
1234 net->ipv4.icmp_sk = NULL;
1237 static int __net_init icmp_sk_init(struct net *net)
1241 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1242 if (!net->ipv4.icmp_sk)
1245 for_each_possible_cpu(i) {
1248 err = inet_ctl_sock_create(&sk, PF_INET,
1249 SOCK_RAW, IPPROTO_ICMP, net);
1253 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1255 /* Enough space for 2 64K ICMP packets, including
1256 * sk_buff/skb_shared_info struct overhead.
1258 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1261 * Speedup sock_wfree()
1263 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1264 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1267 /* Control parameters for ECHO replies. */
1268 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1269 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1271 /* Control parameter - ignore bogus broadcast responses? */
1272 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1275 * Configurable global rate limit.
1277 * ratelimit defines tokens/packet consumed for dst->rate_token
1278 * bucket ratemask defines which icmp types are ratelimited by
1279 * setting it's bit position.
1282 * dest unreachable (3), source quench (4),
1283 * time exceeded (11), parameter problem (12)
1286 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1287 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1288 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1293 for_each_possible_cpu(i)
1294 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1295 free_percpu(net->ipv4.icmp_sk);
1299 static struct pernet_operations __net_initdata icmp_sk_ops = {
1300 .init = icmp_sk_init,
1301 .exit = icmp_sk_exit,
1304 int __init icmp_init(void)
1306 return register_pernet_subsys(&icmp_sk_ops);