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>
101 * Build xmit assembly blocks
110 struct icmphdr icmph;
114 struct ip_options_data replyopts;
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
120 const struct icmp_err icmp_err_convert[] = {
122 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
126 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
130 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
134 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
138 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
142 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
146 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
150 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
154 .errno = ENONET, /* ICMP_HOST_ISOLATED */
158 .errno = ENETUNREACH, /* ICMP_NET_ANO */
162 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
166 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
170 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
174 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
178 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
182 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
186 EXPORT_SYMBOL(icmp_err_convert);
189 * ICMP control array. This specifies what to do with each ICMP.
192 struct icmp_control {
193 void (*handler)(struct sk_buff *skb);
194 short error; /* This ICMP is classed as an error message */
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
200 * The ICMP socket(s). This is the most convenient way to flow control
201 * our ICMP output as well as maintain a clean interface throughout
202 * all layers. All Socketless IP sends will soon be gone.
204 * On SMP we have one ICMP socket per-cpu.
206 static struct sock *icmp_sk(struct net *net)
208 return net->ipv4.icmp_sk[smp_processor_id()];
211 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.
229 static inline void icmp_xmit_unlock(struct sock *sk)
231 spin_unlock_bh(&sk->sk_lock.slock);
235 * Send an ICMP frame.
238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 struct flowi4 *fl4, int type, int code)
241 struct dst_entry *dst = &rt->dst;
244 if (type > NR_ICMP_TYPES)
247 /* Don't limit PMTU discovery. */
248 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
251 /* No rate limit on loopback */
252 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
255 /* Limit if icmp type is enabled in ratemask. */
256 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(peer,
259 net->ipv4.sysctl_icmp_ratelimit);
268 * Maintain the counters used in the SNMP statistics for outgoing ICMP
270 void icmp_out_count(struct net *net, unsigned char type)
272 ICMPMSGOUT_INC_STATS(net, type);
273 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
277 * Checksum each fragment, and on the first include the headers and final
280 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
283 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
286 csum = skb_copy_and_csum_bits(icmp_param->skb,
287 icmp_param->offset + offset,
290 skb->csum = csum_block_add(skb->csum, csum, odd);
291 if (icmp_pointers[icmp_param->data.icmph.type].error)
292 nf_ct_attach(skb, icmp_param->skb);
296 static void icmp_push_reply(struct icmp_bxm *icmp_param,
298 struct ipcm_cookie *ipc, struct rtable **rt)
303 sk = icmp_sk(dev_net((*rt)->dst.dev));
304 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
305 icmp_param->data_len+icmp_param->head_len,
306 icmp_param->head_len,
307 ipc, rt, MSG_DONTWAIT) < 0) {
308 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
309 ip_flush_pending_frames(sk);
310 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
311 struct icmphdr *icmph = icmp_hdr(skb);
313 struct sk_buff *skb1;
315 skb_queue_walk(&sk->sk_write_queue, skb1) {
316 csum = csum_add(csum, skb1->csum);
318 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
320 icmp_param->head_len, csum);
321 icmph->checksum = csum_fold(csum);
322 skb->ip_summed = CHECKSUM_NONE;
323 ip_push_pending_frames(sk, fl4);
328 * Driving logic for building and sending ICMP messages.
331 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
333 struct ipcm_cookie ipc;
334 struct rtable *rt = skb_rtable(skb);
335 struct net *net = dev_net(rt->dst.dev);
338 struct inet_sock *inet;
340 u32 mark = IP4_REPLY_MARK(net, skb->mark);
342 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
345 sk = icmp_xmit_lock(net);
350 icmp_param->data.icmph.checksum = 0;
352 inet->tos = ip_hdr(skb)->tos;
354 daddr = ipc.addr = ip_hdr(skb)->saddr;
355 saddr = fib_compute_spec_dst(skb);
361 if (icmp_param->replyopts.opt.opt.optlen) {
362 ipc.opt = &icmp_param->replyopts.opt;
363 if (ipc.opt->opt.srr)
364 daddr = icmp_param->replyopts.opt.opt.faddr;
366 memset(&fl4, 0, sizeof(fl4));
369 fl4.flowi4_mark = mark;
370 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
371 fl4.flowi4_proto = IPPROTO_ICMP;
372 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
373 rt = ip_route_output_key(net, &fl4);
376 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
377 icmp_param->data.icmph.code))
378 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
381 icmp_xmit_unlock(sk);
384 static struct rtable *icmp_route_lookup(struct net *net,
386 struct sk_buff *skb_in,
387 const struct iphdr *iph,
388 __be32 saddr, u8 tos, u32 mark,
390 struct icmp_bxm *param)
392 struct rtable *rt, *rt2;
393 struct flowi4 fl4_dec;
396 memset(fl4, 0, sizeof(*fl4));
397 fl4->daddr = (param->replyopts.opt.opt.srr ?
398 param->replyopts.opt.opt.faddr : iph->saddr);
400 fl4->flowi4_mark = mark;
401 fl4->flowi4_tos = RT_TOS(tos);
402 fl4->flowi4_proto = IPPROTO_ICMP;
403 fl4->fl4_icmp_type = type;
404 fl4->fl4_icmp_code = code;
405 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
406 rt = __ip_route_output_key(net, fl4);
410 /* No need to clone since we're just using its address. */
413 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
414 flowi4_to_flowi(fl4), NULL, 0);
418 } else if (PTR_ERR(rt) == -EPERM) {
423 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
425 goto relookup_failed;
427 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
428 rt2 = __ip_route_output_key(net, &fl4_dec);
432 struct flowi4 fl4_2 = {};
433 unsigned long orefdst;
435 fl4_2.daddr = fl4_dec.saddr;
436 rt2 = ip_route_output_key(net, &fl4_2);
439 goto relookup_failed;
442 orefdst = skb_in->_skb_refdst; /* save old refdst */
443 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
444 RT_TOS(tos), rt2->dst.dev);
446 dst_release(&rt2->dst);
447 rt2 = skb_rtable(skb_in);
448 skb_in->_skb_refdst = orefdst; /* restore old refdst */
452 goto relookup_failed;
454 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
455 flowi4_to_flowi(&fl4_dec), NULL,
458 dst_release(&rt->dst);
459 memcpy(fl4, &fl4_dec, sizeof(*fl4));
461 } else if (PTR_ERR(rt2) == -EPERM) {
463 dst_release(&rt->dst);
467 goto relookup_failed;
478 * Send an ICMP message in response to a situation
480 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
481 * MAY send more (we do).
482 * MUST NOT change this header information.
483 * MUST NOT reply to a multicast/broadcast IP address.
484 * MUST NOT reply to a multicast/broadcast MAC address.
485 * MUST reply to only the first fragment.
488 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
492 struct icmp_bxm *icmp_param;
493 struct rtable *rt = skb_rtable(skb_in);
494 struct ipcm_cookie ipc;
504 net = dev_net(rt->dst.dev);
507 * Find the original header. It is expected to be valid, of course.
508 * Check this, icmp_send is called from the most obscure devices
511 iph = ip_hdr(skb_in);
513 if ((u8 *)iph < skb_in->head ||
514 (skb_network_header(skb_in) + sizeof(*iph)) >
515 skb_tail_pointer(skb_in))
519 * No replies to physical multicast/broadcast
521 if (skb_in->pkt_type != PACKET_HOST)
525 * Now check at the protocol level
527 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
531 * Only reply to fragment 0. We byte re-order the constant
532 * mask for efficiency.
534 if (iph->frag_off & htons(IP_OFFSET))
538 * If we send an ICMP error to an ICMP error a mess would result..
540 if (icmp_pointers[type].error) {
542 * We are an error, check if we are replying to an
545 if (iph->protocol == IPPROTO_ICMP) {
546 u8 _inner_type, *itp;
548 itp = skb_header_pointer(skb_in,
549 skb_network_header(skb_in) +
551 offsetof(struct icmphdr,
560 * Assume any unknown ICMP type is an error. This
561 * isn't specified by the RFC, but think about it..
563 if (*itp > NR_ICMP_TYPES ||
564 icmp_pointers[*itp].error)
569 icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
573 sk = icmp_xmit_lock(net);
578 * Construct source address and options.
582 if (!(rt->rt_flags & RTCF_LOCAL)) {
583 struct net_device *dev = NULL;
586 if (rt_is_input_route(rt) &&
587 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
588 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
591 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
597 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
598 IPTOS_PREC_INTERNETCONTROL) :
600 mark = IP4_REPLY_MARK(net, skb_in->mark);
602 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
607 * Prepare data for ICMP header.
610 icmp_param->data.icmph.type = type;
611 icmp_param->data.icmph.code = code;
612 icmp_param->data.icmph.un.gateway = info;
613 icmp_param->data.icmph.checksum = 0;
614 icmp_param->skb = skb_in;
615 icmp_param->offset = skb_network_offset(skb_in);
616 inet_sk(sk)->tos = tos;
618 ipc.addr = iph->saddr;
619 ipc.opt = &icmp_param->replyopts.opt;
624 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
625 type, code, icmp_param);
629 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
632 /* RFC says return as much as we can without exceeding 576 bytes. */
634 room = dst_mtu(&rt->dst);
637 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
638 room -= sizeof(struct icmphdr);
640 icmp_param->data_len = skb_in->len - icmp_param->offset;
641 if (icmp_param->data_len > room)
642 icmp_param->data_len = room;
643 icmp_param->head_len = sizeof(struct icmphdr);
645 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
649 icmp_xmit_unlock(sk);
654 EXPORT_SYMBOL(icmp_send);
657 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
659 const struct iphdr *iph = (const struct iphdr *) skb->data;
660 const struct net_protocol *ipprot;
661 int protocol = iph->protocol;
663 /* Checkin full IP header plus 8 bytes of protocol to
664 * avoid additional coding at protocol handlers.
666 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
667 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
671 raw_icmp_error(skb, protocol, info);
673 ipprot = rcu_dereference(inet_protos[protocol]);
674 if (ipprot && ipprot->err_handler)
675 ipprot->err_handler(skb, info);
678 static bool icmp_tag_validation(int proto)
683 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
689 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
690 * ICMP_PARAMETERPROB.
693 static void icmp_unreach(struct sk_buff *skb)
695 const struct iphdr *iph;
696 struct icmphdr *icmph;
700 net = dev_net(skb_dst(skb)->dev);
703 * Incomplete header ?
704 * Only checks for the IP header, there should be an
705 * additional check for longer headers in upper levels.
708 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
711 icmph = icmp_hdr(skb);
712 iph = (const struct iphdr *)skb->data;
714 if (iph->ihl < 5) /* Mangled header, drop. */
717 if (icmph->type == ICMP_DEST_UNREACH) {
718 switch (icmph->code & 15) {
719 case ICMP_NET_UNREACH:
720 case ICMP_HOST_UNREACH:
721 case ICMP_PROT_UNREACH:
722 case ICMP_PORT_UNREACH:
724 case ICMP_FRAG_NEEDED:
725 /* for documentation of the ip_no_pmtu_disc
727 * Documentation/networking/ip-sysctl.txt
729 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
731 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
737 if (!icmp_tag_validation(iph->protocol))
741 info = ntohs(icmph->un.frag.mtu);
745 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
751 if (icmph->code > NR_ICMP_UNREACH)
753 } else if (icmph->type == ICMP_PARAMETERPROB)
754 info = ntohl(icmph->un.gateway) >> 24;
757 * Throw it at our lower layers
759 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
761 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
763 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
768 * Check the other end isn't violating RFC 1122. Some routers send
769 * bogus responses to broadcast frames. If you see this message
770 * first check your netmask matches at both ends, if it does then
771 * get the other vendor to fix their kit.
774 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
775 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
776 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
778 icmph->type, icmph->code,
779 &iph->daddr, skb->dev->name);
783 icmp_socket_deliver(skb, info);
788 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
794 * Handle ICMP_REDIRECT.
797 static void icmp_redirect(struct sk_buff *skb)
799 if (skb->len < sizeof(struct iphdr)) {
800 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
804 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
807 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
811 * Handle ICMP_ECHO ("ping") requests.
813 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
815 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
816 * included in the reply.
817 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
818 * echo requests, MUST have default=NOT.
819 * See also WRT handling of options once they are done and working.
822 static void icmp_echo(struct sk_buff *skb)
826 net = dev_net(skb_dst(skb)->dev);
827 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
828 struct icmp_bxm icmp_param;
830 icmp_param.data.icmph = *icmp_hdr(skb);
831 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
832 icmp_param.skb = skb;
833 icmp_param.offset = 0;
834 icmp_param.data_len = skb->len;
835 icmp_param.head_len = sizeof(struct icmphdr);
836 icmp_reply(&icmp_param, skb);
841 * Handle ICMP Timestamp requests.
842 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
843 * SHOULD be in the kernel for minimum random latency.
844 * MUST be accurate to a few minutes.
845 * MUST be updated at least at 15Hz.
847 static void icmp_timestamp(struct sk_buff *skb)
850 struct icmp_bxm icmp_param;
858 * Fill in the current time as ms since midnight UT:
861 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
862 tv.tv_nsec / NSEC_PER_MSEC);
863 icmp_param.data.times[2] = icmp_param.data.times[1];
864 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
866 icmp_param.data.icmph = *icmp_hdr(skb);
867 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
868 icmp_param.data.icmph.code = 0;
869 icmp_param.skb = skb;
870 icmp_param.offset = 0;
871 icmp_param.data_len = 0;
872 icmp_param.head_len = sizeof(struct icmphdr) + 12;
873 icmp_reply(&icmp_param, skb);
877 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
881 static void icmp_discard(struct sk_buff *skb)
886 * Deal with incoming ICMP packets.
888 int icmp_rcv(struct sk_buff *skb)
890 struct icmphdr *icmph;
891 struct rtable *rt = skb_rtable(skb);
892 struct net *net = dev_net(rt->dst.dev);
894 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
895 struct sec_path *sp = skb_sec_path(skb);
898 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
902 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
905 nh = skb_network_offset(skb);
906 skb_set_network_header(skb, sizeof(*icmph));
908 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
911 skb_set_network_header(skb, nh);
914 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
916 if (skb_checksum_simple_validate(skb))
919 if (!pskb_pull(skb, sizeof(*icmph)))
922 icmph = icmp_hdr(skb);
924 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
926 * 18 is the highest 'known' ICMP type. Anything else is a mystery
928 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
931 if (icmph->type > NR_ICMP_TYPES)
936 * Parse the ICMP message
939 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
941 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
942 * silently ignored (we let user decide with a sysctl).
943 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
944 * discarded if to broadcast/multicast.
946 if ((icmph->type == ICMP_ECHO ||
947 icmph->type == ICMP_TIMESTAMP) &&
948 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
951 if (icmph->type != ICMP_ECHO &&
952 icmph->type != ICMP_TIMESTAMP &&
953 icmph->type != ICMP_ADDRESS &&
954 icmph->type != ICMP_ADDRESSREPLY) {
959 icmp_pointers[icmph->type].handler(skb);
965 ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
967 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
971 void icmp_err(struct sk_buff *skb, u32 info)
973 struct iphdr *iph = (struct iphdr *)skb->data;
974 int offset = iph->ihl<<2;
975 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
976 int type = icmp_hdr(skb)->type;
977 int code = icmp_hdr(skb)->code;
978 struct net *net = dev_net(skb->dev);
981 * Use ping_err to handle all icmp errors except those
982 * triggered by ICMP_ECHOREPLY which sent from kernel.
984 if (icmph->type != ICMP_ECHOREPLY) {
985 ping_err(skb, offset, info);
989 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
990 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
991 else if (type == ICMP_REDIRECT)
992 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
996 * This table is the definition of how we handle ICMP.
998 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1000 .handler = ping_rcv,
1003 .handler = icmp_discard,
1007 .handler = icmp_discard,
1010 [ICMP_DEST_UNREACH] = {
1011 .handler = icmp_unreach,
1014 [ICMP_SOURCE_QUENCH] = {
1015 .handler = icmp_unreach,
1019 .handler = icmp_redirect,
1023 .handler = icmp_discard,
1027 .handler = icmp_discard,
1031 .handler = icmp_echo,
1034 .handler = icmp_discard,
1038 .handler = icmp_discard,
1041 [ICMP_TIME_EXCEEDED] = {
1042 .handler = icmp_unreach,
1045 [ICMP_PARAMETERPROB] = {
1046 .handler = icmp_unreach,
1049 [ICMP_TIMESTAMP] = {
1050 .handler = icmp_timestamp,
1052 [ICMP_TIMESTAMPREPLY] = {
1053 .handler = icmp_discard,
1055 [ICMP_INFO_REQUEST] = {
1056 .handler = icmp_discard,
1058 [ICMP_INFO_REPLY] = {
1059 .handler = icmp_discard,
1062 .handler = icmp_discard,
1064 [ICMP_ADDRESSREPLY] = {
1065 .handler = icmp_discard,
1069 static void __net_exit icmp_sk_exit(struct net *net)
1073 for_each_possible_cpu(i)
1074 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1075 kfree(net->ipv4.icmp_sk);
1076 net->ipv4.icmp_sk = NULL;
1079 static int __net_init icmp_sk_init(struct net *net)
1084 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1085 if (net->ipv4.icmp_sk == NULL)
1088 for_each_possible_cpu(i) {
1091 err = inet_ctl_sock_create(&sk, PF_INET,
1092 SOCK_RAW, IPPROTO_ICMP, net);
1096 net->ipv4.icmp_sk[i] = sk;
1098 /* Enough space for 2 64K ICMP packets, including
1099 * sk_buff/skb_shared_info struct overhead.
1101 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1104 * Speedup sock_wfree()
1106 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1107 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1110 /* Control parameters for ECHO replies. */
1111 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1112 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1114 /* Control parameter - ignore bogus broadcast responses? */
1115 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1118 * Configurable global rate limit.
1120 * ratelimit defines tokens/packet consumed for dst->rate_token
1121 * bucket ratemask defines which icmp types are ratelimited by
1122 * setting it's bit position.
1125 * dest unreachable (3), source quench (4),
1126 * time exceeded (11), parameter problem (12)
1129 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1130 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1131 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1136 for_each_possible_cpu(i)
1137 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1138 kfree(net->ipv4.icmp_sk);
1142 static struct pernet_operations __net_initdata icmp_sk_ops = {
1143 .init = icmp_sk_init,
1144 .exit = icmp_sk_exit,
1147 int __init icmp_init(void)
1149 return register_pernet_subsys(&icmp_sk_ops);