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 = sysctl_icmp_msgs_per_sec * delta / HZ ;
273 WRITE_ONCE(icmp_global.stamp, now);
275 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
277 /* We want to use a credit of one in average, but need to randomize
278 * it for security reasons.
280 credit = max_t(int, credit - prandom_u32_max(3), 0);
283 WRITE_ONCE(icmp_global.credit, credit);
284 spin_unlock(&icmp_global.lock);
287 EXPORT_SYMBOL(icmp_global_allow);
290 * Send an ICMP frame.
293 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
294 struct flowi4 *fl4, int type, int code)
296 struct dst_entry *dst = &rt->dst;
299 if (type > NR_ICMP_TYPES)
302 /* Don't limit PMTU discovery. */
303 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
306 /* No rate limit on loopback */
307 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
310 /* Limit if icmp type is enabled in ratemask. */
311 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
315 if (icmp_global_allow()) {
316 int vif = l3mdev_master_ifindex(dst->dev);
317 struct inet_peer *peer;
319 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
320 rc = inet_peer_xrlim_allow(peer,
321 net->ipv4.sysctl_icmp_ratelimit);
330 * Maintain the counters used in the SNMP statistics for outgoing ICMP
332 void icmp_out_count(struct net *net, unsigned char type)
334 ICMPMSGOUT_INC_STATS(net, type);
335 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
339 * Checksum each fragment, and on the first include the headers and final
342 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
345 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
348 csum = skb_copy_and_csum_bits(icmp_param->skb,
349 icmp_param->offset + offset,
352 skb->csum = csum_block_add(skb->csum, csum, odd);
353 if (icmp_pointers[icmp_param->data.icmph.type].error)
354 nf_ct_attach(skb, icmp_param->skb);
358 static void icmp_push_reply(struct icmp_bxm *icmp_param,
360 struct ipcm_cookie *ipc, struct rtable **rt)
365 sk = icmp_sk(dev_net((*rt)->dst.dev));
366 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
367 icmp_param->data_len+icmp_param->head_len,
368 icmp_param->head_len,
369 ipc, rt, MSG_DONTWAIT) < 0) {
370 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
371 ip_flush_pending_frames(sk);
372 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
373 struct icmphdr *icmph = icmp_hdr(skb);
375 struct sk_buff *skb1;
377 skb_queue_walk(&sk->sk_write_queue, skb1) {
378 csum = csum_add(csum, skb1->csum);
380 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
382 icmp_param->head_len, csum);
383 icmph->checksum = csum_fold(csum);
384 skb->ip_summed = CHECKSUM_NONE;
385 ip_push_pending_frames(sk, fl4);
390 * Driving logic for building and sending ICMP messages.
393 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
395 struct ipcm_cookie ipc;
396 struct rtable *rt = skb_rtable(skb);
397 struct net *net = dev_net(rt->dst.dev);
400 struct inet_sock *inet;
402 u32 mark = IP4_REPLY_MARK(net, skb->mark);
404 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
407 sk = icmp_xmit_lock(net);
412 icmp_param->data.icmph.checksum = 0;
414 inet->tos = ip_hdr(skb)->tos;
416 daddr = ipc.addr = ip_hdr(skb)->saddr;
417 saddr = fib_compute_spec_dst(skb);
423 if (icmp_param->replyopts.opt.opt.optlen) {
424 ipc.opt = &icmp_param->replyopts.opt;
425 if (ipc.opt->opt.srr)
426 daddr = icmp_param->replyopts.opt.opt.faddr;
428 memset(&fl4, 0, sizeof(fl4));
431 fl4.flowi4_mark = mark;
432 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
433 fl4.flowi4_proto = IPPROTO_ICMP;
434 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
435 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
436 rt = ip_route_output_key(net, &fl4);
439 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
440 icmp_param->data.icmph.code))
441 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
444 icmp_xmit_unlock(sk);
447 #ifdef CONFIG_IP_ROUTE_MULTIPATH
449 /* Source and destination is swapped. See ip_multipath_icmp_hash */
450 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
452 const struct iphdr *iph = ip_hdr(skb);
454 return fib_multipath_hash(iph->daddr, iph->saddr);
459 #define icmp_multipath_hash_skb(skb) (-1)
464 * The device used for looking up which routing table to use for sending an ICMP
465 * error is preferably the source whenever it is set, which should ensure the
466 * icmp error can be sent to the source host, else lookup using the routing
467 * table of the destination device, else use the main routing table (index 0).
469 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
471 struct net_device *route_lookup_dev = NULL;
474 route_lookup_dev = skb->dev;
475 else if (skb_dst(skb))
476 route_lookup_dev = skb_dst(skb)->dev;
477 return route_lookup_dev;
480 static struct rtable *icmp_route_lookup(struct net *net,
482 struct sk_buff *skb_in,
483 const struct iphdr *iph,
484 __be32 saddr, u8 tos, u32 mark,
486 struct icmp_bxm *param)
488 struct net_device *route_lookup_dev;
489 struct rtable *rt, *rt2;
490 struct flowi4 fl4_dec;
493 memset(fl4, 0, sizeof(*fl4));
494 fl4->daddr = (param->replyopts.opt.opt.srr ?
495 param->replyopts.opt.opt.faddr : iph->saddr);
497 fl4->flowi4_mark = mark;
498 fl4->flowi4_tos = RT_TOS(tos);
499 fl4->flowi4_proto = IPPROTO_ICMP;
500 fl4->fl4_icmp_type = type;
501 fl4->fl4_icmp_code = code;
502 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
503 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
505 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
506 rt = __ip_route_output_key_hash(net, fl4,
507 icmp_multipath_hash_skb(skb_in));
511 /* No need to clone since we're just using its address. */
514 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
515 flowi4_to_flowi(fl4), NULL, 0);
519 } else if (PTR_ERR(rt) == -EPERM) {
524 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
526 goto relookup_failed;
528 if (inet_addr_type_dev_table(net, route_lookup_dev,
529 fl4_dec.saddr) == RTN_LOCAL) {
530 rt2 = __ip_route_output_key(net, &fl4_dec);
534 struct flowi4 fl4_2 = {};
535 unsigned long orefdst;
537 fl4_2.daddr = fl4_dec.saddr;
538 rt2 = ip_route_output_key(net, &fl4_2);
541 goto relookup_failed;
544 orefdst = skb_in->_skb_refdst; /* save old refdst */
545 skb_dst_set(skb_in, NULL);
546 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
547 RT_TOS(tos), rt2->dst.dev);
549 dst_release(&rt2->dst);
550 rt2 = skb_rtable(skb_in);
551 skb_in->_skb_refdst = orefdst; /* restore old refdst */
555 goto relookup_failed;
557 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
558 flowi4_to_flowi(&fl4_dec), NULL,
561 dst_release(&rt->dst);
562 memcpy(fl4, &fl4_dec, sizeof(*fl4));
564 } else if (PTR_ERR(rt2) == -EPERM) {
566 dst_release(&rt->dst);
570 goto relookup_failed;
581 * Send an ICMP message in response to a situation
583 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
584 * MAY send more (we do).
585 * MUST NOT change this header information.
586 * MUST NOT reply to a multicast/broadcast IP address.
587 * MUST NOT reply to a multicast/broadcast MAC address.
588 * MUST reply to only the first fragment.
591 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
592 const struct ip_options *opt)
596 struct icmp_bxm *icmp_param;
597 struct rtable *rt = skb_rtable(skb_in);
598 struct ipcm_cookie ipc;
608 net = dev_net(rt->dst.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 icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
677 sk = icmp_xmit_lock(net);
682 * Construct source address and options.
686 if (!(rt->rt_flags & RTCF_LOCAL)) {
687 struct net_device *dev = NULL;
690 if (rt_is_input_route(rt) &&
691 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
692 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
695 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
701 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
702 IPTOS_PREC_INTERNETCONTROL) :
704 mark = IP4_REPLY_MARK(net, skb_in->mark);
706 if (__ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in, opt))
711 * Prepare data for ICMP header.
714 icmp_param->data.icmph.type = type;
715 icmp_param->data.icmph.code = code;
716 icmp_param->data.icmph.un.gateway = info;
717 icmp_param->data.icmph.checksum = 0;
718 icmp_param->skb = skb_in;
719 icmp_param->offset = skb_network_offset(skb_in);
720 inet_sk(sk)->tos = tos;
722 ipc.addr = iph->saddr;
723 ipc.opt = &icmp_param->replyopts.opt;
728 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
729 type, code, icmp_param);
733 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
736 /* RFC says return as much as we can without exceeding 576 bytes. */
738 room = dst_mtu(&rt->dst);
741 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
742 room -= sizeof(struct icmphdr);
744 icmp_param->data_len = skb_in->len - icmp_param->offset;
745 if (icmp_param->data_len > room)
746 icmp_param->data_len = room;
747 icmp_param->head_len = sizeof(struct icmphdr);
749 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
753 icmp_xmit_unlock(sk);
758 EXPORT_SYMBOL(__icmp_send);
760 #if IS_ENABLED(CONFIG_NF_NAT)
761 #include <net/netfilter/nf_conntrack.h>
762 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
764 struct sk_buff *cloned_skb = NULL;
765 struct ip_options opts = { 0 };
766 enum ip_conntrack_info ctinfo;
770 ct = nf_ct_get(skb_in, &ctinfo);
771 if (!ct || !(ct->status & IPS_SRC_NAT)) {
772 __icmp_send(skb_in, type, code, info, &opts);
776 if (skb_shared(skb_in))
777 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
779 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
780 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
781 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
782 skb_network_offset(skb_in) + sizeof(struct iphdr))))
785 orig_ip = ip_hdr(skb_in)->saddr;
786 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
787 __icmp_send(skb_in, type, code, info, &opts);
788 ip_hdr(skb_in)->saddr = orig_ip;
790 consume_skb(cloned_skb);
792 EXPORT_SYMBOL(icmp_ndo_send);
795 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
797 const struct iphdr *iph = (const struct iphdr *) skb->data;
798 const struct net_protocol *ipprot;
799 int protocol = iph->protocol;
801 /* Checkin full IP header plus 8 bytes of protocol to
802 * avoid additional coding at protocol handlers.
804 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
805 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
809 raw_icmp_error(skb, protocol, info);
811 ipprot = rcu_dereference(inet_protos[protocol]);
812 if (ipprot && ipprot->err_handler)
813 ipprot->err_handler(skb, info);
816 static bool icmp_tag_validation(int proto)
821 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
827 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
828 * ICMP_PARAMETERPROB.
831 static bool icmp_unreach(struct sk_buff *skb)
833 const struct iphdr *iph;
834 struct icmphdr *icmph;
838 net = dev_net(skb_dst(skb)->dev);
841 * Incomplete header ?
842 * Only checks for the IP header, there should be an
843 * additional check for longer headers in upper levels.
846 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
849 icmph = icmp_hdr(skb);
850 iph = (const struct iphdr *)skb->data;
852 if (iph->ihl < 5) /* Mangled header, drop. */
855 switch (icmph->type) {
856 case ICMP_DEST_UNREACH:
857 switch (icmph->code & 15) {
858 case ICMP_NET_UNREACH:
859 case ICMP_HOST_UNREACH:
860 case ICMP_PROT_UNREACH:
861 case ICMP_PORT_UNREACH:
863 case ICMP_FRAG_NEEDED:
864 /* for documentation of the ip_no_pmtu_disc
866 * Documentation/networking/ip-sysctl.txt
868 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
870 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
876 if (!icmp_tag_validation(iph->protocol))
880 info = ntohs(icmph->un.frag.mtu);
884 net_dbg_ratelimited("%pI4: Source Route Failed\n",
890 if (icmph->code > NR_ICMP_UNREACH)
893 case ICMP_PARAMETERPROB:
894 info = ntohl(icmph->un.gateway) >> 24;
896 case ICMP_TIME_EXCEEDED:
897 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
898 if (icmph->code == ICMP_EXC_FRAGTIME)
904 * Throw it at our lower layers
906 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
908 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
910 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
915 * Check the other end isn't violating RFC 1122. Some routers send
916 * bogus responses to broadcast frames. If you see this message
917 * first check your netmask matches at both ends, if it does then
918 * get the other vendor to fix their kit.
921 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
922 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
923 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
925 icmph->type, icmph->code,
926 &iph->daddr, skb->dev->name);
930 icmp_socket_deliver(skb, info);
935 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
941 * Handle ICMP_REDIRECT.
944 static bool icmp_redirect(struct sk_buff *skb)
946 if (skb->len < sizeof(struct iphdr)) {
947 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
951 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
952 /* there aught to be a stat */
956 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
961 * Handle ICMP_ECHO ("ping") requests.
963 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
965 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
966 * included in the reply.
967 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
968 * echo requests, MUST have default=NOT.
969 * See also WRT handling of options once they are done and working.
972 static bool icmp_echo(struct sk_buff *skb)
976 net = dev_net(skb_dst(skb)->dev);
977 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
978 struct icmp_bxm icmp_param;
980 icmp_param.data.icmph = *icmp_hdr(skb);
981 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
982 icmp_param.skb = skb;
983 icmp_param.offset = 0;
984 icmp_param.data_len = skb->len;
985 icmp_param.head_len = sizeof(struct icmphdr);
986 icmp_reply(&icmp_param, skb);
988 /* should there be an ICMP stat for ignored echos? */
993 * Handle ICMP Timestamp requests.
994 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
995 * SHOULD be in the kernel for minimum random latency.
996 * MUST be accurate to a few minutes.
997 * MUST be updated at least at 15Hz.
999 static bool icmp_timestamp(struct sk_buff *skb)
1001 struct icmp_bxm icmp_param;
1009 * Fill in the current time as ms since midnight UT:
1011 icmp_param.data.times[1] = inet_current_timestamp();
1012 icmp_param.data.times[2] = icmp_param.data.times[1];
1013 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
1015 icmp_param.data.icmph = *icmp_hdr(skb);
1016 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1017 icmp_param.data.icmph.code = 0;
1018 icmp_param.skb = skb;
1019 icmp_param.offset = 0;
1020 icmp_param.data_len = 0;
1021 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1022 icmp_reply(&icmp_param, skb);
1026 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1030 static bool icmp_discard(struct sk_buff *skb)
1032 /* pretend it was a success */
1037 * Deal with incoming ICMP packets.
1039 int icmp_rcv(struct sk_buff *skb)
1041 struct icmphdr *icmph;
1042 struct rtable *rt = skb_rtable(skb);
1043 struct net *net = dev_net(rt->dst.dev);
1046 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1047 struct sec_path *sp = skb_sec_path(skb);
1050 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1054 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1057 nh = skb_network_offset(skb);
1058 skb_set_network_header(skb, sizeof(*icmph));
1060 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1063 skb_set_network_header(skb, nh);
1066 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1068 if (skb_checksum_simple_validate(skb))
1071 if (!pskb_pull(skb, sizeof(*icmph)))
1074 icmph = icmp_hdr(skb);
1076 ICMPMSGIN_INC_STATS(net, icmph->type);
1078 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1080 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1083 if (icmph->type > NR_ICMP_TYPES)
1088 * Parse the ICMP message
1091 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1093 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1094 * silently ignored (we let user decide with a sysctl).
1095 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1096 * discarded if to broadcast/multicast.
1098 if ((icmph->type == ICMP_ECHO ||
1099 icmph->type == ICMP_TIMESTAMP) &&
1100 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1103 if (icmph->type != ICMP_ECHO &&
1104 icmph->type != ICMP_TIMESTAMP &&
1105 icmph->type != ICMP_ADDRESS &&
1106 icmph->type != ICMP_ADDRESSREPLY) {
1111 success = icmp_pointers[icmph->type].handler(skb);
1122 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1124 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1128 void icmp_err(struct sk_buff *skb, u32 info)
1130 struct iphdr *iph = (struct iphdr *)skb->data;
1131 int offset = iph->ihl<<2;
1132 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1133 int type = icmp_hdr(skb)->type;
1134 int code = icmp_hdr(skb)->code;
1135 struct net *net = dev_net(skb->dev);
1138 * Use ping_err to handle all icmp errors except those
1139 * triggered by ICMP_ECHOREPLY which sent from kernel.
1141 if (icmph->type != ICMP_ECHOREPLY) {
1142 ping_err(skb, offset, info);
1146 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1147 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1148 else if (type == ICMP_REDIRECT)
1149 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1153 * This table is the definition of how we handle ICMP.
1155 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1156 [ICMP_ECHOREPLY] = {
1157 .handler = ping_rcv,
1160 .handler = icmp_discard,
1164 .handler = icmp_discard,
1167 [ICMP_DEST_UNREACH] = {
1168 .handler = icmp_unreach,
1171 [ICMP_SOURCE_QUENCH] = {
1172 .handler = icmp_unreach,
1176 .handler = icmp_redirect,
1180 .handler = icmp_discard,
1184 .handler = icmp_discard,
1188 .handler = icmp_echo,
1191 .handler = icmp_discard,
1195 .handler = icmp_discard,
1198 [ICMP_TIME_EXCEEDED] = {
1199 .handler = icmp_unreach,
1202 [ICMP_PARAMETERPROB] = {
1203 .handler = icmp_unreach,
1206 [ICMP_TIMESTAMP] = {
1207 .handler = icmp_timestamp,
1209 [ICMP_TIMESTAMPREPLY] = {
1210 .handler = icmp_discard,
1212 [ICMP_INFO_REQUEST] = {
1213 .handler = icmp_discard,
1215 [ICMP_INFO_REPLY] = {
1216 .handler = icmp_discard,
1219 .handler = icmp_discard,
1221 [ICMP_ADDRESSREPLY] = {
1222 .handler = icmp_discard,
1226 static void __net_exit icmp_sk_exit(struct net *net)
1230 for_each_possible_cpu(i)
1231 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1232 free_percpu(net->ipv4.icmp_sk);
1233 net->ipv4.icmp_sk = NULL;
1236 static int __net_init icmp_sk_init(struct net *net)
1240 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1241 if (!net->ipv4.icmp_sk)
1244 for_each_possible_cpu(i) {
1247 err = inet_ctl_sock_create(&sk, PF_INET,
1248 SOCK_RAW, IPPROTO_ICMP, net);
1252 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1254 /* Enough space for 2 64K ICMP packets, including
1255 * sk_buff/skb_shared_info struct overhead.
1257 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1260 * Speedup sock_wfree()
1262 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1263 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1266 /* Control parameters for ECHO replies. */
1267 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1268 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1270 /* Control parameter - ignore bogus broadcast responses? */
1271 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1274 * Configurable global rate limit.
1276 * ratelimit defines tokens/packet consumed for dst->rate_token
1277 * bucket ratemask defines which icmp types are ratelimited by
1278 * setting it's bit position.
1281 * dest unreachable (3), source quench (4),
1282 * time exceeded (11), parameter problem (12)
1285 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1286 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1287 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1292 for_each_possible_cpu(i)
1293 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1294 free_percpu(net->ipv4.icmp_sk);
1298 static struct pernet_operations __net_initdata icmp_sk_ops = {
1299 .init = icmp_sk_init,
1300 .exit = icmp_sk_exit,
1303 int __init icmp_init(void)
1305 return register_pernet_subsys(&icmp_sk_ops);