2 * NET3: Implementation of the ICMP protocol layer.
4 * Alan Cox, <alan@redhat.com>
6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Some of the function names and the icmp unreach table for this
14 * module were derived from [icmp.c 1.0.11 06/02/93] by
15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16 * Other than that this module is a complete rewrite.
19 * Clemens Fruhwirth : introduce global icmp rate limiting
20 * with icmp type masking ability instead
21 * of broken per type icmp timeouts.
22 * Mike Shaver : RFC1122 checks.
23 * Alan Cox : Multicast ping reply as self.
24 * Alan Cox : Fix atomicity lockup in ip_build_xmit
26 * Alan Cox : Added 216,128 byte paths to the MTU
28 * Martin Mares : RFC1812 checks.
29 * Martin Mares : Can be configured to follow redirects
30 * if acting as a router _without_ a
31 * routing protocol (RFC 1812).
32 * Martin Mares : Echo requests may be configured to
33 * be ignored (RFC 1812).
34 * Martin Mares : Limitation of ICMP error message
35 * transmit rate (RFC 1812).
36 * Martin Mares : TOS and Precedence set correctly
38 * Martin Mares : Now copying as much data from the
39 * original packet as we can without
40 * exceeding 576 bytes (RFC 1812).
41 * Willy Konynenberg : Transparent proxying support.
42 * Keith Owens : RFC1191 correction for 4.2BSD based
44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
46 * Andi Kleen : Check all packet lengths properly
47 * and moved all kfree_skb() up to
49 * Andi Kleen : Move the rate limit bookkeeping
50 * into the dest entry and use a token
51 * bucket filter (thanks to ANK). Make
52 * the rates sysctl configurable.
53 * Yu Tianli : Fixed two ugly bugs in icmp_send
54 * - IP option length was accounted wrongly
55 * - ICMP header length was not accounted
57 * Tristan Greaves : Added sysctl option to ignore bogus
58 * broadcast responses from broken routers.
62 * - Should use skb_pull() instead of all the manual checking.
63 * This would also greatly simply some upper layer error handlers. --AK
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>
81 #include <net/route.h>
82 #include <net/protocol.h>
87 #include <linux/skbuff.h>
89 #include <linux/errno.h>
90 #include <linux/timer.h>
91 #include <linux/init.h>
92 #include <asm/system.h>
93 #include <asm/uaccess.h>
94 #include <net/checksum.h>
98 * Build xmit assembly blocks
107 struct icmphdr icmph;
111 struct ip_options replyopts;
112 unsigned char optbuf[40];
118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
119 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
124 struct icmp_err icmp_err_convert[] = {
126 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
130 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
134 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
138 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
142 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
146 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
150 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
154 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
158 .errno = ENONET, /* ICMP_HOST_ISOLATED */
162 .errno = ENETUNREACH, /* ICMP_NET_ANO */
166 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
170 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
174 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
178 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
182 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
186 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
192 * ICMP control array. This specifies what to do with each ICMP.
195 struct icmp_control {
196 void (*handler)(struct sk_buff *skb);
197 short error; /* This ICMP is classed as an error message */
200 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
203 * The ICMP socket(s). This is the most convenient way to flow control
204 * our ICMP output as well as maintain a clean interface throughout
205 * all layers. All Socketless IP sends will soon be gone.
207 * On SMP we have one ICMP socket per-cpu.
209 static struct sock *icmp_sk(struct net *net)
211 return net->ipv4.icmp_sk[smp_processor_id()];
214 static inline int icmp_xmit_lock(struct sock *sk)
218 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 /* This can happen if the output path signals a
220 * dst_link_failure() for an outgoing ICMP packet.
228 static inline void icmp_xmit_unlock(struct sock *sk)
230 spin_unlock_bh(&sk->sk_lock.slock);
234 * Send an ICMP frame.
238 * Check transmit rate limitation for given message.
239 * The rate information is held in the destination cache now.
240 * This function is generic and could be used for other purposes
241 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
243 * Note that the same dst_entry fields are modified by functions in
244 * route.c too, but these work for packet destinations while xrlim_allow
245 * works for icmp destinations. This means the rate limiting information
246 * for one "ip object" is shared - and these ICMPs are twice limited:
247 * by source and by destination.
249 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
250 * SHOULD allow setting of rate limits
252 * Shared between ICMPv4 and ICMPv6.
254 #define XRLIM_BURST_FACTOR 6
255 int xrlim_allow(struct dst_entry *dst, int timeout)
257 unsigned long now, token = dst->rate_tokens;
261 token += now - dst->rate_last;
262 dst->rate_last = now;
263 if (token > XRLIM_BURST_FACTOR * timeout)
264 token = XRLIM_BURST_FACTOR * timeout;
265 if (token >= timeout) {
269 dst->rate_tokens = token;
273 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
275 struct dst_entry *dst = &rt->u.dst;
278 if (type > NR_ICMP_TYPES)
281 /* Don't limit PMTU discovery. */
282 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 /* No rate limit on loopback */
286 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
289 /* Limit if icmp type is enabled in ratemask. */
290 if ((1 << type) & init_net.ipv4.sysctl_icmp_ratemask)
291 rc = xrlim_allow(dst, init_net.ipv4.sysctl_icmp_ratelimit);
297 * Maintain the counters used in the SNMP statistics for outgoing ICMP
299 void icmp_out_count(unsigned char type)
301 ICMPMSGOUT_INC_STATS(type);
302 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
306 * Checksum each fragment, and on the first include the headers and final
309 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
312 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
315 csum = skb_copy_and_csum_bits(icmp_param->skb,
316 icmp_param->offset + offset,
319 skb->csum = csum_block_add(skb->csum, csum, odd);
320 if (icmp_pointers[icmp_param->data.icmph.type].error)
321 nf_ct_attach(skb, icmp_param->skb);
325 static void icmp_push_reply(struct icmp_bxm *icmp_param,
326 struct ipcm_cookie *ipc, struct rtable *rt)
331 sk = icmp_sk(dev_net(rt->u.dst.dev));
332 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
333 icmp_param->data_len+icmp_param->head_len,
334 icmp_param->head_len,
335 ipc, rt, MSG_DONTWAIT) < 0)
336 ip_flush_pending_frames(sk);
337 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
338 struct icmphdr *icmph = icmp_hdr(skb);
340 struct sk_buff *skb1;
342 skb_queue_walk(&sk->sk_write_queue, skb1) {
343 csum = csum_add(csum, skb1->csum);
345 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
347 icmp_param->head_len, csum);
348 icmph->checksum = csum_fold(csum);
349 skb->ip_summed = CHECKSUM_NONE;
350 ip_push_pending_frames(sk);
355 * Driving logic for building and sending ICMP messages.
358 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
360 struct ipcm_cookie ipc;
361 struct rtable *rt = skb->rtable;
362 struct net *net = dev_net(rt->u.dst.dev);
363 struct sock *sk = icmp_sk(net);
364 struct inet_sock *inet = inet_sk(sk);
367 if (ip_options_echo(&icmp_param->replyopts, skb))
370 if (icmp_xmit_lock(sk))
373 icmp_param->data.icmph.checksum = 0;
375 inet->tos = ip_hdr(skb)->tos;
376 daddr = ipc.addr = rt->rt_src;
378 if (icmp_param->replyopts.optlen) {
379 ipc.opt = &icmp_param->replyopts;
381 daddr = icmp_param->replyopts.faddr;
384 struct flowi fl = { .nl_u = { .ip4_u =
386 .saddr = rt->rt_spec_dst,
387 .tos = RT_TOS(ip_hdr(skb)->tos) } },
388 .proto = IPPROTO_ICMP };
389 security_skb_classify_flow(skb, &fl);
390 if (ip_route_output_key(net, &rt, &fl))
393 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
394 icmp_param->data.icmph.code))
395 icmp_push_reply(icmp_param, &ipc, rt);
398 icmp_xmit_unlock(sk);
403 * Send an ICMP message in response to a situation
405 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
406 * MAY send more (we do).
407 * MUST NOT change this header information.
408 * MUST NOT reply to a multicast/broadcast IP address.
409 * MUST NOT reply to a multicast/broadcast MAC address.
410 * MUST reply to only the first fragment.
413 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
417 struct icmp_bxm icmp_param;
418 struct rtable *rt = skb_in->rtable;
419 struct ipcm_cookie ipc;
427 net = dev_net(rt->u.dst.dev);
431 * Find the original header. It is expected to be valid, of course.
432 * Check this, icmp_send is called from the most obscure devices
435 iph = ip_hdr(skb_in);
437 if ((u8 *)iph < skb_in->head ||
438 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
442 * No replies to physical multicast/broadcast
444 if (skb_in->pkt_type != PACKET_HOST)
448 * Now check at the protocol level
450 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
454 * Only reply to fragment 0. We byte re-order the constant
455 * mask for efficiency.
457 if (iph->frag_off & htons(IP_OFFSET))
461 * If we send an ICMP error to an ICMP error a mess would result..
463 if (icmp_pointers[type].error) {
465 * We are an error, check if we are replying to an
468 if (iph->protocol == IPPROTO_ICMP) {
469 u8 _inner_type, *itp;
471 itp = skb_header_pointer(skb_in,
472 skb_network_header(skb_in) +
474 offsetof(struct icmphdr,
483 * Assume any unknown ICMP type is an error. This
484 * isn't specified by the RFC, but think about it..
486 if (*itp > NR_ICMP_TYPES ||
487 icmp_pointers[*itp].error)
492 if (icmp_xmit_lock(sk))
496 * Construct source address and options.
500 if (!(rt->rt_flags & RTCF_LOCAL)) {
501 struct net_device *dev = NULL;
504 init_net.ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
505 dev = dev_get_by_index(net, rt->fl.iif);
508 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
514 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
515 IPTOS_PREC_INTERNETCONTROL) :
518 if (ip_options_echo(&icmp_param.replyopts, skb_in))
523 * Prepare data for ICMP header.
526 icmp_param.data.icmph.type = type;
527 icmp_param.data.icmph.code = code;
528 icmp_param.data.icmph.un.gateway = info;
529 icmp_param.data.icmph.checksum = 0;
530 icmp_param.skb = skb_in;
531 icmp_param.offset = skb_network_offset(skb_in);
532 inet_sk(sk)->tos = tos;
533 ipc.addr = iph->saddr;
534 ipc.opt = &icmp_param.replyopts;
540 .daddr = icmp_param.replyopts.srr ?
541 icmp_param.replyopts.faddr :
547 .proto = IPPROTO_ICMP,
558 security_skb_classify_flow(skb_in, &fl);
559 if (__ip_route_output_key(net, &rt, &fl))
562 /* No need to clone since we're just using its address. */
565 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
578 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
581 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
582 err = __ip_route_output_key(net, &rt2, &fl);
584 struct flowi fl2 = {};
585 struct dst_entry *odst;
587 fl2.fl4_dst = fl.fl4_src;
588 if (ip_route_output_key(net, &rt2, &fl2))
593 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
594 RT_TOS(tos), rt2->u.dst.dev);
596 dst_release(&rt2->u.dst);
597 rt2 = skb_in->rtable;
604 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
606 if (err == -ENOENT) {
612 dst_release(&rt->u.dst);
620 if (!icmpv4_xrlim_allow(rt, type, code))
623 /* RFC says return as much as we can without exceeding 576 bytes. */
625 room = dst_mtu(&rt->u.dst);
628 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
629 room -= sizeof(struct icmphdr);
631 icmp_param.data_len = skb_in->len - icmp_param.offset;
632 if (icmp_param.data_len > room)
633 icmp_param.data_len = room;
634 icmp_param.head_len = sizeof(struct icmphdr);
636 icmp_push_reply(&icmp_param, &ipc, rt);
640 icmp_xmit_unlock(sk);
646 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
649 static void icmp_unreach(struct sk_buff *skb)
652 struct icmphdr *icmph;
654 struct net_protocol *ipprot;
658 net = dev_net(skb->dst->dev);
661 * Incomplete header ?
662 * Only checks for the IP header, there should be an
663 * additional check for longer headers in upper levels.
666 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
669 icmph = icmp_hdr(skb);
670 iph = (struct iphdr *)skb->data;
672 if (iph->ihl < 5) /* Mangled header, drop. */
675 if (icmph->type == ICMP_DEST_UNREACH) {
676 switch (icmph->code & 15) {
677 case ICMP_NET_UNREACH:
678 case ICMP_HOST_UNREACH:
679 case ICMP_PROT_UNREACH:
680 case ICMP_PORT_UNREACH:
682 case ICMP_FRAG_NEEDED:
683 if (ipv4_config.no_pmtu_disc) {
684 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
685 "fragmentation needed "
687 NIPQUAD(iph->daddr));
689 info = ip_rt_frag_needed(net, iph,
690 ntohs(icmph->un.frag.mtu));
696 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
698 NIPQUAD(iph->daddr));
703 if (icmph->code > NR_ICMP_UNREACH)
705 } else if (icmph->type == ICMP_PARAMETERPROB)
706 info = ntohl(icmph->un.gateway) >> 24;
709 * Throw it at our lower layers
711 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
713 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
715 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
720 * Check the other end isnt violating RFC 1122. Some routers send
721 * bogus responses to broadcast frames. If you see this message
722 * first check your netmask matches at both ends, if it does then
723 * get the other vendor to fix their kit.
726 if (!init_net.ipv4.sysctl_icmp_ignore_bogus_error_responses &&
727 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
729 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
731 "error to a broadcast: %u.%u.%u.%u on %s\n",
732 NIPQUAD(ip_hdr(skb)->saddr),
733 icmph->type, icmph->code,
739 /* Checkin full IP header plus 8 bytes of protocol to
740 * avoid additional coding at protocol handlers.
742 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
745 iph = (struct iphdr *)skb->data;
746 protocol = iph->protocol;
749 * Deliver ICMP message to raw sockets. Pretty useless feature?
751 raw_icmp_error(skb, protocol, info);
753 hash = protocol & (MAX_INET_PROTOS - 1);
755 ipprot = rcu_dereference(inet_protos[hash]);
756 if (ipprot && ipprot->err_handler)
757 ipprot->err_handler(skb, info);
763 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
769 * Handle ICMP_REDIRECT.
772 static void icmp_redirect(struct sk_buff *skb)
776 if (skb->len < sizeof(struct iphdr))
780 * Get the copied header of the packet that caused the redirect
782 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
785 iph = (struct iphdr *)skb->data;
787 switch (icmp_hdr(skb)->code & 7) {
789 case ICMP_REDIR_NETTOS:
791 * As per RFC recommendations now handle it as a host redirect.
793 case ICMP_REDIR_HOST:
794 case ICMP_REDIR_HOSTTOS:
795 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
796 icmp_hdr(skb)->un.gateway,
797 iph->saddr, skb->dev);
803 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
808 * Handle ICMP_ECHO ("ping") requests.
810 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
812 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
813 * included in the reply.
814 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
815 * echo requests, MUST have default=NOT.
816 * See also WRT handling of options once they are done and working.
819 static void icmp_echo(struct sk_buff *skb)
821 if (!init_net.ipv4.sysctl_icmp_echo_ignore_all) {
822 struct icmp_bxm icmp_param;
824 icmp_param.data.icmph = *icmp_hdr(skb);
825 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
826 icmp_param.skb = skb;
827 icmp_param.offset = 0;
828 icmp_param.data_len = skb->len;
829 icmp_param.head_len = sizeof(struct icmphdr);
830 icmp_reply(&icmp_param, skb);
835 * Handle ICMP Timestamp requests.
836 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
837 * SHOULD be in the kernel for minimum random latency.
838 * MUST be accurate to a few minutes.
839 * MUST be updated at least at 15Hz.
841 static void icmp_timestamp(struct sk_buff *skb)
844 struct icmp_bxm icmp_param;
852 * Fill in the current time as ms since midnight UT:
854 do_gettimeofday(&tv);
855 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
857 icmp_param.data.times[2] = icmp_param.data.times[1];
858 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
860 icmp_param.data.icmph = *icmp_hdr(skb);
861 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
862 icmp_param.data.icmph.code = 0;
863 icmp_param.skb = skb;
864 icmp_param.offset = 0;
865 icmp_param.data_len = 0;
866 icmp_param.head_len = sizeof(struct icmphdr) + 12;
867 icmp_reply(&icmp_param, skb);
871 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
877 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
879 * RFC1122 (3.2.2.9). A host MUST only send replies to
880 * ADDRESS_MASK requests if it's been configured as an address mask
881 * agent. Receiving a request doesn't constitute implicit permission to
882 * act as one. Of course, implementing this correctly requires (SHOULD)
883 * a way to turn the functionality on and off. Another one for sysctl(),
886 * RFC1812 (4.3.3.9). A router MUST implement it.
887 * A router SHOULD have switch turning it on/off.
888 * This switch MUST be ON by default.
890 * Gratuitous replies, zero-source replies are not implemented,
891 * that complies with RFC. DO NOT implement them!!! All the idea
892 * of broadcast addrmask replies as specified in RFC950 is broken.
893 * The problem is that it is not uncommon to have several prefixes
894 * on one physical interface. Moreover, addrmask agent can even be
895 * not aware of existing another prefixes.
896 * If source is zero, addrmask agent cannot choose correct prefix.
897 * Gratuitous mask announcements suffer from the same problem.
898 * RFC1812 explains it, but still allows to use ADDRMASK,
899 * that is pretty silly. --ANK
901 * All these rules are so bizarre, that I removed kernel addrmask
902 * support at all. It is wrong, it is obsolete, nobody uses it in
905 * Furthermore you can do it with a usermode address agent program
909 static void icmp_address(struct sk_buff *skb)
913 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
918 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
919 * loudly if an inconsistency is found.
922 static void icmp_address_reply(struct sk_buff *skb)
924 struct rtable *rt = skb->rtable;
925 struct net_device *dev = skb->dev;
926 struct in_device *in_dev;
927 struct in_ifaddr *ifa;
929 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
932 in_dev = in_dev_get(dev);
936 if (in_dev->ifa_list &&
937 IN_DEV_LOG_MARTIANS(in_dev) &&
938 IN_DEV_FORWARD(in_dev)) {
941 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
943 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
944 if (*mp == ifa->ifa_mask &&
945 inet_ifa_match(rt->rt_src, ifa))
948 if (!ifa && net_ratelimit()) {
949 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
951 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
959 static void icmp_discard(struct sk_buff *skb)
964 * Deal with incoming ICMP packets.
966 int icmp_rcv(struct sk_buff *skb)
968 struct icmphdr *icmph;
969 struct rtable *rt = skb->rtable;
971 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
974 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
978 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
981 nh = skb_network_offset(skb);
982 skb_set_network_header(skb, sizeof(*icmph));
984 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
987 skb_set_network_header(skb, nh);
990 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
992 switch (skb->ip_summed) {
993 case CHECKSUM_COMPLETE:
994 if (!csum_fold(skb->csum))
999 if (__skb_checksum_complete(skb))
1003 if (!pskb_pull(skb, sizeof(*icmph)))
1006 icmph = icmp_hdr(skb);
1008 ICMPMSGIN_INC_STATS_BH(icmph->type);
1010 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1012 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1015 if (icmph->type > NR_ICMP_TYPES)
1020 * Parse the ICMP message
1023 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1025 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1026 * silently ignored (we let user decide with a sysctl).
1027 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1028 * discarded if to broadcast/multicast.
1030 if ((icmph->type == ICMP_ECHO ||
1031 icmph->type == ICMP_TIMESTAMP) &&
1032 init_net.ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1035 if (icmph->type != ICMP_ECHO &&
1036 icmph->type != ICMP_TIMESTAMP &&
1037 icmph->type != ICMP_ADDRESS &&
1038 icmph->type != ICMP_ADDRESSREPLY) {
1043 icmp_pointers[icmph->type].handler(skb);
1049 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1054 * This table is the definition of how we handle ICMP.
1056 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1057 [ICMP_ECHOREPLY] = {
1058 .handler = icmp_discard,
1061 .handler = icmp_discard,
1065 .handler = icmp_discard,
1068 [ICMP_DEST_UNREACH] = {
1069 .handler = icmp_unreach,
1072 [ICMP_SOURCE_QUENCH] = {
1073 .handler = icmp_unreach,
1077 .handler = icmp_redirect,
1081 .handler = icmp_discard,
1085 .handler = icmp_discard,
1089 .handler = icmp_echo,
1092 .handler = icmp_discard,
1096 .handler = icmp_discard,
1099 [ICMP_TIME_EXCEEDED] = {
1100 .handler = icmp_unreach,
1103 [ICMP_PARAMETERPROB] = {
1104 .handler = icmp_unreach,
1107 [ICMP_TIMESTAMP] = {
1108 .handler = icmp_timestamp,
1110 [ICMP_TIMESTAMPREPLY] = {
1111 .handler = icmp_discard,
1113 [ICMP_INFO_REQUEST] = {
1114 .handler = icmp_discard,
1116 [ICMP_INFO_REPLY] = {
1117 .handler = icmp_discard,
1120 .handler = icmp_address,
1122 [ICMP_ADDRESSREPLY] = {
1123 .handler = icmp_address_reply,
1127 static void __net_exit icmp_sk_exit(struct net *net)
1131 for_each_possible_cpu(i)
1132 sk_release_kernel(net->ipv4.icmp_sk[i]);
1133 kfree(net->ipv4.icmp_sk);
1134 net->ipv4.icmp_sk = NULL;
1137 int __net_init icmp_sk_init(struct net *net)
1142 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1143 if (net->ipv4.icmp_sk == NULL)
1146 for_each_possible_cpu(i) {
1148 struct socket *sock;
1149 struct inet_sock *inet;
1151 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, &sock);
1155 net->ipv4.icmp_sk[i] = sk = sock->sk;
1156 sk_change_net(sk, net);
1158 sk->sk_allocation = GFP_ATOMIC;
1160 /* Enough space for 2 64K ICMP packets, including
1161 * sk_buff struct overhead.
1164 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1168 inet->pmtudisc = IP_PMTUDISC_DONT;
1170 /* Unhash it so that IP input processing does not even
1171 * see it, we do not wish this socket to see incoming
1174 sk->sk_prot->unhash(sk);
1177 /* Control parameters for ECHO replies. */
1178 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1179 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1181 /* Control parameter - ignore bogus broadcast responses? */
1182 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1185 * Configurable global rate limit.
1187 * ratelimit defines tokens/packet consumed for dst->rate_token
1188 * bucket ratemask defines which icmp types are ratelimited by
1189 * setting it's bit position.
1192 * dest unreachable (3), source quench (4),
1193 * time exceeded (11), parameter problem (12)
1196 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1197 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1198 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1203 for_each_possible_cpu(i)
1204 sk_release_kernel(net->ipv4.icmp_sk[i]);
1205 kfree(net->ipv4.icmp_sk);
1209 static struct pernet_operations __net_initdata icmp_sk_ops = {
1210 .init = icmp_sk_init,
1211 .exit = icmp_sk_exit,
1214 int __init icmp_init(void)
1216 return register_pernet_device(&icmp_sk_ops);
1219 EXPORT_SYMBOL(icmp_err_convert);
1220 EXPORT_SYMBOL(icmp_send);
1221 EXPORT_SYMBOL(icmp_statistics);
1222 EXPORT_SYMBOL(xrlim_allow);