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 <linux/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 /* Called with BH disabled */
213 static inline struct sock *icmp_xmit_lock(struct net *net)
219 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
220 /* This can happen if the output path signals a
221 * dst_link_failure() for an outgoing ICMP packet.
228 static inline void icmp_xmit_unlock(struct sock *sk)
230 spin_unlock(&sk->sk_lock.slock);
233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
234 int sysctl_icmp_msgs_burst __read_mostly = 50;
241 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
245 * icmp_global_allow - Are we allowed to send one more ICMP message ?
247 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
248 * Returns false if we reached the limit and can not send another packet.
249 * Note: called with BH disabled
251 bool icmp_global_allow(void)
253 u32 credit, delta, incr = 0, now = (u32)jiffies;
256 /* Check if token bucket is empty and cannot be refilled
257 * without taking the spinlock. The READ_ONCE() are paired
258 * with the following WRITE_ONCE() in this same function.
260 if (!READ_ONCE(icmp_global.credit)) {
261 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
266 spin_lock(&icmp_global.lock);
267 delta = min_t(u32, now - icmp_global.stamp, HZ);
268 if (delta >= HZ / 50) {
269 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
271 WRITE_ONCE(icmp_global.stamp, now);
273 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
275 /* We want to use a credit of one in average, but need to randomize
276 * it for security reasons.
278 credit = max_t(int, credit - prandom_u32_max(3), 0);
281 WRITE_ONCE(icmp_global.credit, credit);
282 spin_unlock(&icmp_global.lock);
285 EXPORT_SYMBOL(icmp_global_allow);
287 static bool icmpv4_mask_allow(struct net *net, int type, int code)
289 if (type > NR_ICMP_TYPES)
292 /* Don't limit PMTU discovery. */
293 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
296 /* Limit if icmp type is enabled in ratemask. */
297 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
303 static bool icmpv4_global_allow(struct net *net, int type, int code)
305 if (icmpv4_mask_allow(net, type, code))
308 if (icmp_global_allow())
315 * Send an ICMP frame.
318 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
319 struct flowi4 *fl4, int type, int code)
321 struct dst_entry *dst = &rt->dst;
322 struct inet_peer *peer;
326 if (icmpv4_mask_allow(net, type, code))
329 /* No rate limit on loopback */
330 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
333 vif = l3mdev_master_ifindex(dst->dev);
334 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
335 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
343 * Maintain the counters used in the SNMP statistics for outgoing ICMP
345 void icmp_out_count(struct net *net, unsigned char type)
347 ICMPMSGOUT_INC_STATS(net, type);
348 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
352 * Checksum each fragment, and on the first include the headers and final
355 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
358 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
361 csum = skb_copy_and_csum_bits(icmp_param->skb,
362 icmp_param->offset + offset,
365 skb->csum = csum_block_add(skb->csum, csum, odd);
366 if (icmp_pointers[icmp_param->data.icmph.type].error)
367 nf_ct_attach(skb, icmp_param->skb);
371 static void icmp_push_reply(struct icmp_bxm *icmp_param,
373 struct ipcm_cookie *ipc, struct rtable **rt)
378 sk = icmp_sk(dev_net((*rt)->dst.dev));
379 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
380 icmp_param->data_len+icmp_param->head_len,
381 icmp_param->head_len,
382 ipc, rt, MSG_DONTWAIT) < 0) {
383 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
384 ip_flush_pending_frames(sk);
385 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
386 struct icmphdr *icmph = icmp_hdr(skb);
388 struct sk_buff *skb1;
390 skb_queue_walk(&sk->sk_write_queue, skb1) {
391 csum = csum_add(csum, skb1->csum);
393 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
395 icmp_param->head_len, csum);
396 icmph->checksum = csum_fold(csum);
397 skb->ip_summed = CHECKSUM_NONE;
398 ip_push_pending_frames(sk, fl4);
403 * Driving logic for building and sending ICMP messages.
406 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
408 struct ipcm_cookie ipc;
409 struct rtable *rt = skb_rtable(skb);
410 struct net *net = dev_net(rt->dst.dev);
413 struct inet_sock *inet;
415 u32 mark = IP4_REPLY_MARK(net, skb->mark);
416 int type = icmp_param->data.icmph.type;
417 int code = icmp_param->data.icmph.code;
419 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
422 /* Needed by both icmp_global_allow and icmp_xmit_lock */
425 /* global icmp_msgs_per_sec */
426 if (!icmpv4_global_allow(net, type, code))
429 sk = icmp_xmit_lock(net);
434 icmp_param->data.icmph.checksum = 0;
437 inet->tos = ip_hdr(skb)->tos;
439 daddr = ipc.addr = ip_hdr(skb)->saddr;
440 saddr = fib_compute_spec_dst(skb);
442 if (icmp_param->replyopts.opt.opt.optlen) {
443 ipc.opt = &icmp_param->replyopts.opt;
444 if (ipc.opt->opt.srr)
445 daddr = icmp_param->replyopts.opt.opt.faddr;
447 memset(&fl4, 0, sizeof(fl4));
450 fl4.flowi4_mark = mark;
451 fl4.flowi4_uid = sock_net_uid(net, NULL);
452 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
453 fl4.flowi4_proto = IPPROTO_ICMP;
454 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
455 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
456 rt = ip_route_output_key(net, &fl4);
459 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
460 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
463 icmp_xmit_unlock(sk);
469 * The device used for looking up which routing table to use for sending an ICMP
470 * error is preferably the source whenever it is set, which should ensure the
471 * icmp error can be sent to the source host, else lookup using the routing
472 * table of the destination device, else use the main routing table (index 0).
474 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
476 struct net_device *route_lookup_dev = NULL;
479 route_lookup_dev = skb->dev;
480 else if (skb_dst(skb))
481 route_lookup_dev = skb_dst(skb)->dev;
482 return route_lookup_dev;
485 static struct rtable *icmp_route_lookup(struct net *net,
487 struct sk_buff *skb_in,
488 const struct iphdr *iph,
489 __be32 saddr, u8 tos, u32 mark,
491 struct icmp_bxm *param)
493 struct net_device *route_lookup_dev;
494 struct rtable *rt, *rt2;
495 struct flowi4 fl4_dec;
498 memset(fl4, 0, sizeof(*fl4));
499 fl4->daddr = (param->replyopts.opt.opt.srr ?
500 param->replyopts.opt.opt.faddr : iph->saddr);
502 fl4->flowi4_mark = mark;
503 fl4->flowi4_uid = sock_net_uid(net, NULL);
504 fl4->flowi4_tos = RT_TOS(tos);
505 fl4->flowi4_proto = IPPROTO_ICMP;
506 fl4->fl4_icmp_type = type;
507 fl4->fl4_icmp_code = code;
508 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
509 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
511 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
512 rt = ip_route_output_key_hash(net, fl4, skb_in);
516 /* No need to clone since we're just using its address. */
519 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
520 flowi4_to_flowi(fl4), NULL, 0);
524 } else if (PTR_ERR(rt) == -EPERM) {
529 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
531 goto relookup_failed;
533 if (inet_addr_type_dev_table(net, route_lookup_dev,
534 fl4_dec.saddr) == RTN_LOCAL) {
535 rt2 = __ip_route_output_key(net, &fl4_dec);
539 struct flowi4 fl4_2 = {};
540 unsigned long orefdst;
542 fl4_2.daddr = fl4_dec.saddr;
543 rt2 = ip_route_output_key(net, &fl4_2);
546 goto relookup_failed;
549 orefdst = skb_in->_skb_refdst; /* save old refdst */
550 skb_dst_set(skb_in, NULL);
551 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
552 RT_TOS(tos), rt2->dst.dev);
554 dst_release(&rt2->dst);
555 rt2 = skb_rtable(skb_in);
556 skb_in->_skb_refdst = orefdst; /* restore old refdst */
560 goto relookup_failed;
562 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
563 flowi4_to_flowi(&fl4_dec), NULL,
566 dst_release(&rt->dst);
567 memcpy(fl4, &fl4_dec, sizeof(*fl4));
569 } else if (PTR_ERR(rt2) == -EPERM) {
571 dst_release(&rt->dst);
575 goto relookup_failed;
586 * Send an ICMP message in response to a situation
588 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
589 * MAY send more (we do).
590 * MUST NOT change this header information.
591 * MUST NOT reply to a multicast/broadcast IP address.
592 * MUST NOT reply to a multicast/broadcast MAC address.
593 * MUST reply to only the first fragment.
596 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
597 const struct ip_options *opt)
601 struct icmp_bxm icmp_param;
602 struct rtable *rt = skb_rtable(skb_in);
603 struct ipcm_cookie ipc;
615 net = dev_net(rt->dst.dev);
616 else if (skb_in->dev)
617 net = dev_net(skb_in->dev);
622 * Find the original header. It is expected to be valid, of course.
623 * Check this, icmp_send is called from the most obscure devices
626 iph = ip_hdr(skb_in);
628 if ((u8 *)iph < skb_in->head ||
629 (skb_network_header(skb_in) + sizeof(*iph)) >
630 skb_tail_pointer(skb_in))
634 * No replies to physical multicast/broadcast
636 if (skb_in->pkt_type != PACKET_HOST)
640 * Now check at the protocol level
642 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
646 * Only reply to fragment 0. We byte re-order the constant
647 * mask for efficiency.
649 if (iph->frag_off & htons(IP_OFFSET))
653 * If we send an ICMP error to an ICMP error a mess would result..
655 if (icmp_pointers[type].error) {
657 * We are an error, check if we are replying to an
660 if (iph->protocol == IPPROTO_ICMP) {
661 u8 _inner_type, *itp;
663 itp = skb_header_pointer(skb_in,
664 skb_network_header(skb_in) +
666 offsetof(struct icmphdr,
675 * Assume any unknown ICMP type is an error. This
676 * isn't specified by the RFC, but think about it..
678 if (*itp > NR_ICMP_TYPES ||
679 icmp_pointers[*itp].error)
684 /* Needed by both icmp_global_allow and icmp_xmit_lock */
687 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
688 * incoming dev is loopback. If outgoing dev change to not be
689 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
691 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
692 !icmpv4_global_allow(net, type, code))
695 sk = icmp_xmit_lock(net);
700 * Construct source address and options.
704 if (!(rt->rt_flags & RTCF_LOCAL)) {
705 struct net_device *dev = NULL;
708 if (rt_is_input_route(rt) &&
709 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
710 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
713 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
719 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
720 IPTOS_PREC_INTERNETCONTROL) :
722 mark = IP4_REPLY_MARK(net, skb_in->mark);
724 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
729 * Prepare data for ICMP header.
732 icmp_param.data.icmph.type = type;
733 icmp_param.data.icmph.code = code;
734 icmp_param.data.icmph.un.gateway = info;
735 icmp_param.data.icmph.checksum = 0;
736 icmp_param.skb = skb_in;
737 icmp_param.offset = skb_network_offset(skb_in);
738 inet_sk(sk)->tos = tos;
741 ipc.addr = iph->saddr;
742 ipc.opt = &icmp_param.replyopts.opt;
744 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
745 type, code, &icmp_param);
749 /* peer icmp_ratelimit */
750 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
753 /* RFC says return as much as we can without exceeding 576 bytes. */
755 room = dst_mtu(&rt->dst);
758 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
759 room -= sizeof(struct icmphdr);
761 icmp_param.data_len = skb_in->len - icmp_param.offset;
762 if (icmp_param.data_len > room)
763 icmp_param.data_len = room;
764 icmp_param.head_len = sizeof(struct icmphdr);
766 /* if we don't have a source address at this point, fall back to the
767 * dummy address instead of sending out a packet with a source address
771 fl4.saddr = htonl(INADDR_DUMMY);
773 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
777 icmp_xmit_unlock(sk);
782 EXPORT_SYMBOL(__icmp_send);
784 #if IS_ENABLED(CONFIG_NF_NAT)
785 #include <net/netfilter/nf_conntrack.h>
786 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
788 struct sk_buff *cloned_skb = NULL;
789 struct ip_options opts = { 0 };
790 enum ip_conntrack_info ctinfo;
794 ct = nf_ct_get(skb_in, &ctinfo);
795 if (!ct || !(ct->status & IPS_SRC_NAT)) {
796 __icmp_send(skb_in, type, code, info, &opts);
800 if (skb_shared(skb_in))
801 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
803 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
804 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
805 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
806 skb_network_offset(skb_in) + sizeof(struct iphdr))))
809 orig_ip = ip_hdr(skb_in)->saddr;
810 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
811 __icmp_send(skb_in, type, code, info, &opts);
812 ip_hdr(skb_in)->saddr = orig_ip;
814 consume_skb(cloned_skb);
816 EXPORT_SYMBOL(icmp_ndo_send);
819 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
821 const struct iphdr *iph = (const struct iphdr *) skb->data;
822 const struct net_protocol *ipprot;
823 int protocol = iph->protocol;
825 /* Checkin full IP header plus 8 bytes of protocol to
826 * avoid additional coding at protocol handlers.
828 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
829 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
833 raw_icmp_error(skb, protocol, info);
835 ipprot = rcu_dereference(inet_protos[protocol]);
836 if (ipprot && ipprot->err_handler)
837 ipprot->err_handler(skb, info);
840 static bool icmp_tag_validation(int proto)
845 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
851 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
852 * ICMP_PARAMETERPROB.
855 static bool icmp_unreach(struct sk_buff *skb)
857 const struct iphdr *iph;
858 struct icmphdr *icmph;
862 net = dev_net(skb_dst(skb)->dev);
865 * Incomplete header ?
866 * Only checks for the IP header, there should be an
867 * additional check for longer headers in upper levels.
870 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
873 icmph = icmp_hdr(skb);
874 iph = (const struct iphdr *)skb->data;
876 if (iph->ihl < 5) /* Mangled header, drop. */
879 switch (icmph->type) {
880 case ICMP_DEST_UNREACH:
881 switch (icmph->code & 15) {
882 case ICMP_NET_UNREACH:
883 case ICMP_HOST_UNREACH:
884 case ICMP_PROT_UNREACH:
885 case ICMP_PORT_UNREACH:
887 case ICMP_FRAG_NEEDED:
888 /* for documentation of the ip_no_pmtu_disc
890 * Documentation/networking/ip-sysctl.txt
892 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
894 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
900 if (!icmp_tag_validation(iph->protocol))
904 info = ntohs(icmph->un.frag.mtu);
908 net_dbg_ratelimited("%pI4: Source Route Failed\n",
914 if (icmph->code > NR_ICMP_UNREACH)
917 case ICMP_PARAMETERPROB:
918 info = ntohl(icmph->un.gateway) >> 24;
920 case ICMP_TIME_EXCEEDED:
921 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
922 if (icmph->code == ICMP_EXC_FRAGTIME)
928 * Throw it at our lower layers
930 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
932 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
934 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
939 * Check the other end isn't violating RFC 1122. Some routers send
940 * bogus responses to broadcast frames. If you see this message
941 * first check your netmask matches at both ends, if it does then
942 * get the other vendor to fix their kit.
945 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
946 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
947 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
949 icmph->type, icmph->code,
950 &iph->daddr, skb->dev->name);
954 icmp_socket_deliver(skb, info);
959 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
965 * Handle ICMP_REDIRECT.
968 static bool icmp_redirect(struct sk_buff *skb)
970 if (skb->len < sizeof(struct iphdr)) {
971 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
975 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
976 /* there aught to be a stat */
980 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
985 * Handle ICMP_ECHO ("ping") requests.
987 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
989 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
990 * included in the reply.
991 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
992 * echo requests, MUST have default=NOT.
993 * See also WRT handling of options once they are done and working.
996 static bool icmp_echo(struct sk_buff *skb)
1000 net = dev_net(skb_dst(skb)->dev);
1001 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
1002 struct icmp_bxm icmp_param;
1004 icmp_param.data.icmph = *icmp_hdr(skb);
1005 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1006 icmp_param.skb = skb;
1007 icmp_param.offset = 0;
1008 icmp_param.data_len = skb->len;
1009 icmp_param.head_len = sizeof(struct icmphdr);
1010 icmp_reply(&icmp_param, skb);
1012 /* should there be an ICMP stat for ignored echos? */
1017 * Handle ICMP Timestamp requests.
1018 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1019 * SHOULD be in the kernel for minimum random latency.
1020 * MUST be accurate to a few minutes.
1021 * MUST be updated at least at 15Hz.
1023 static bool icmp_timestamp(struct sk_buff *skb)
1025 struct icmp_bxm icmp_param;
1033 * Fill in the current time as ms since midnight UT:
1035 icmp_param.data.times[1] = inet_current_timestamp();
1036 icmp_param.data.times[2] = icmp_param.data.times[1];
1038 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1040 icmp_param.data.icmph = *icmp_hdr(skb);
1041 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1042 icmp_param.data.icmph.code = 0;
1043 icmp_param.skb = skb;
1044 icmp_param.offset = 0;
1045 icmp_param.data_len = 0;
1046 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1047 icmp_reply(&icmp_param, skb);
1051 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1055 static bool icmp_discard(struct sk_buff *skb)
1057 /* pretend it was a success */
1062 * Deal with incoming ICMP packets.
1064 int icmp_rcv(struct sk_buff *skb)
1066 struct icmphdr *icmph;
1067 struct rtable *rt = skb_rtable(skb);
1068 struct net *net = dev_net(rt->dst.dev);
1071 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1072 struct sec_path *sp = skb_sec_path(skb);
1075 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1079 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1082 nh = skb_network_offset(skb);
1083 skb_set_network_header(skb, sizeof(*icmph));
1085 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1088 skb_set_network_header(skb, nh);
1091 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1093 if (skb_checksum_simple_validate(skb))
1096 if (!pskb_pull(skb, sizeof(*icmph)))
1099 icmph = icmp_hdr(skb);
1101 ICMPMSGIN_INC_STATS(net, icmph->type);
1103 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1105 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1108 if (icmph->type > NR_ICMP_TYPES)
1113 * Parse the ICMP message
1116 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1118 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1119 * silently ignored (we let user decide with a sysctl).
1120 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1121 * discarded if to broadcast/multicast.
1123 if ((icmph->type == ICMP_ECHO ||
1124 icmph->type == ICMP_TIMESTAMP) &&
1125 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1128 if (icmph->type != ICMP_ECHO &&
1129 icmph->type != ICMP_TIMESTAMP &&
1130 icmph->type != ICMP_ADDRESS &&
1131 icmph->type != ICMP_ADDRESSREPLY) {
1136 success = icmp_pointers[icmph->type].handler(skb);
1140 return NET_RX_SUCCESS;
1147 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1149 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1153 void icmp_err(struct sk_buff *skb, u32 info)
1155 struct iphdr *iph = (struct iphdr *)skb->data;
1156 int offset = iph->ihl<<2;
1157 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1158 int type = icmp_hdr(skb)->type;
1159 int code = icmp_hdr(skb)->code;
1160 struct net *net = dev_net(skb->dev);
1163 * Use ping_err to handle all icmp errors except those
1164 * triggered by ICMP_ECHOREPLY which sent from kernel.
1166 if (icmph->type != ICMP_ECHOREPLY) {
1167 ping_err(skb, offset, info);
1171 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1172 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1173 else if (type == ICMP_REDIRECT)
1174 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1178 * This table is the definition of how we handle ICMP.
1180 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1181 [ICMP_ECHOREPLY] = {
1182 .handler = ping_rcv,
1185 .handler = icmp_discard,
1189 .handler = icmp_discard,
1192 [ICMP_DEST_UNREACH] = {
1193 .handler = icmp_unreach,
1196 [ICMP_SOURCE_QUENCH] = {
1197 .handler = icmp_unreach,
1201 .handler = icmp_redirect,
1205 .handler = icmp_discard,
1209 .handler = icmp_discard,
1213 .handler = icmp_echo,
1216 .handler = icmp_discard,
1220 .handler = icmp_discard,
1223 [ICMP_TIME_EXCEEDED] = {
1224 .handler = icmp_unreach,
1227 [ICMP_PARAMETERPROB] = {
1228 .handler = icmp_unreach,
1231 [ICMP_TIMESTAMP] = {
1232 .handler = icmp_timestamp,
1234 [ICMP_TIMESTAMPREPLY] = {
1235 .handler = icmp_discard,
1237 [ICMP_INFO_REQUEST] = {
1238 .handler = icmp_discard,
1240 [ICMP_INFO_REPLY] = {
1241 .handler = icmp_discard,
1244 .handler = icmp_discard,
1246 [ICMP_ADDRESSREPLY] = {
1247 .handler = icmp_discard,
1251 static void __net_exit icmp_sk_exit(struct net *net)
1255 for_each_possible_cpu(i)
1256 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1257 free_percpu(net->ipv4.icmp_sk);
1258 net->ipv4.icmp_sk = NULL;
1261 static int __net_init icmp_sk_init(struct net *net)
1265 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1266 if (!net->ipv4.icmp_sk)
1269 for_each_possible_cpu(i) {
1272 err = inet_ctl_sock_create(&sk, PF_INET,
1273 SOCK_RAW, IPPROTO_ICMP, net);
1277 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1279 /* Enough space for 2 64K ICMP packets, including
1280 * sk_buff/skb_shared_info struct overhead.
1282 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1285 * Speedup sock_wfree()
1287 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1288 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1291 /* Control parameters for ECHO replies. */
1292 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1293 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1295 /* Control parameter - ignore bogus broadcast responses? */
1296 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1299 * Configurable global rate limit.
1301 * ratelimit defines tokens/packet consumed for dst->rate_token
1302 * bucket ratemask defines which icmp types are ratelimited by
1303 * setting it's bit position.
1306 * dest unreachable (3), source quench (4),
1307 * time exceeded (11), parameter problem (12)
1310 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1311 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1312 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1317 for_each_possible_cpu(i)
1318 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1319 free_percpu(net->ipv4.icmp_sk);
1323 static struct pernet_operations __net_initdata icmp_sk_ops = {
1324 .init = icmp_sk_init,
1325 .exit = icmp_sk_exit,
1328 int __init icmp_init(void)
1330 return register_pernet_subsys(&icmp_sk_ops);