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>
97 * Build xmit assembly blocks
106 struct icmphdr icmph;
110 struct ip_options replyopts;
111 unsigned char optbuf[40];
117 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
118 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
120 /* An array of errno for error messages from dest unreach. */
121 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
123 struct icmp_err icmp_err_convert[] = {
125 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
129 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
133 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
137 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
141 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
145 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
149 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
153 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
157 .errno = ENONET, /* ICMP_HOST_ISOLATED */
161 .errno = ENETUNREACH, /* ICMP_NET_ANO */
165 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
169 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
173 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
177 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
181 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
185 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
190 /* Control parameters for ECHO replies. */
191 int sysctl_icmp_echo_ignore_all __read_mostly;
192 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
194 /* Control parameter - ignore bogus broadcast responses? */
195 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
198 * Configurable global rate limit.
200 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
201 * ratemask defines which icmp types are ratelimited by setting
205 * dest unreachable (3), source quench (4),
206 * time exceeded (11), parameter problem (12)
209 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
210 int sysctl_icmp_ratemask __read_mostly = 0x1818;
211 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
214 * ICMP control array. This specifies what to do with each ICMP.
217 struct icmp_control {
218 void (*handler)(struct sk_buff *skb);
219 short error; /* This ICMP is classed as an error message */
222 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
225 * The ICMP socket(s). This is the most convenient way to flow control
226 * our ICMP output as well as maintain a clean interface throughout
227 * all layers. All Socketless IP sends will soon be gone.
229 * On SMP we have one ICMP socket per-cpu.
231 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
232 #define icmp_socket __get_cpu_var(__icmp_socket)
234 static __inline__ int icmp_xmit_lock(void)
238 if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
239 /* This can happen if the output path signals a
240 * dst_link_failure() for an outgoing ICMP packet.
248 static void icmp_xmit_unlock(void)
250 spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
254 * Send an ICMP frame.
258 * Check transmit rate limitation for given message.
259 * The rate information is held in the destination cache now.
260 * This function is generic and could be used for other purposes
261 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
263 * Note that the same dst_entry fields are modified by functions in
264 * route.c too, but these work for packet destinations while xrlim_allow
265 * works for icmp destinations. This means the rate limiting information
266 * for one "ip object" is shared - and these ICMPs are twice limited:
267 * by source and by destination.
269 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
270 * SHOULD allow setting of rate limits
272 * Shared between ICMPv4 and ICMPv6.
274 #define XRLIM_BURST_FACTOR 6
275 int xrlim_allow(struct dst_entry *dst, int timeout)
281 dst->rate_tokens += now - dst->rate_last;
282 dst->rate_last = now;
283 if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
284 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
285 if (dst->rate_tokens >= timeout) {
286 dst->rate_tokens -= timeout;
292 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
294 struct dst_entry *dst = &rt->u.dst;
297 if (type > NR_ICMP_TYPES)
300 /* Don't limit PMTU discovery. */
301 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
304 /* No rate limit on loopback */
305 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
308 /* Limit if icmp type is enabled in ratemask. */
309 if ((1 << type) & sysctl_icmp_ratemask)
310 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
316 * Maintain the counters used in the SNMP statistics for outgoing ICMP
318 void icmp_out_count(unsigned char type)
320 ICMPMSGOUT_INC_STATS(type);
321 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
325 * Checksum each fragment, and on the first include the headers and final
328 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
331 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
334 csum = skb_copy_and_csum_bits(icmp_param->skb,
335 icmp_param->offset + offset,
338 skb->csum = csum_block_add(skb->csum, csum, odd);
339 if (icmp_pointers[icmp_param->data.icmph.type].error)
340 nf_ct_attach(skb, icmp_param->skb);
344 static void icmp_push_reply(struct icmp_bxm *icmp_param,
345 struct ipcm_cookie *ipc, struct rtable *rt)
349 if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
350 icmp_param->data_len+icmp_param->head_len,
351 icmp_param->head_len,
352 ipc, rt, MSG_DONTWAIT) < 0)
353 ip_flush_pending_frames(icmp_socket->sk);
354 else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
355 struct icmphdr *icmph = icmp_hdr(skb);
357 struct sk_buff *skb1;
359 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
360 csum = csum_add(csum, skb1->csum);
362 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
364 icmp_param->head_len, csum);
365 icmph->checksum = csum_fold(csum);
366 skb->ip_summed = CHECKSUM_NONE;
367 ip_push_pending_frames(icmp_socket->sk);
372 * Driving logic for building and sending ICMP messages.
375 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
377 struct sock *sk = icmp_socket->sk;
378 struct inet_sock *inet = inet_sk(sk);
379 struct ipcm_cookie ipc;
380 struct rtable *rt = (struct rtable *)skb->dst;
383 if (ip_options_echo(&icmp_param->replyopts, skb))
386 if (icmp_xmit_lock())
389 icmp_param->data.icmph.checksum = 0;
391 inet->tos = ip_hdr(skb)->tos;
392 daddr = ipc.addr = rt->rt_src;
394 if (icmp_param->replyopts.optlen) {
395 ipc.opt = &icmp_param->replyopts;
397 daddr = icmp_param->replyopts.faddr;
400 struct flowi fl = { .nl_u = { .ip4_u =
402 .saddr = rt->rt_spec_dst,
403 .tos = RT_TOS(ip_hdr(skb)->tos) } },
404 .proto = IPPROTO_ICMP };
405 security_skb_classify_flow(skb, &fl);
406 if (ip_route_output_key(&rt, &fl))
409 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
410 icmp_param->data.icmph.code))
411 icmp_push_reply(icmp_param, &ipc, rt);
419 * Send an ICMP message in response to a situation
421 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
422 * MAY send more (we do).
423 * MUST NOT change this header information.
424 * MUST NOT reply to a multicast/broadcast IP address.
425 * MUST NOT reply to a multicast/broadcast MAC address.
426 * MUST reply to only the first fragment.
429 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
433 struct icmp_bxm icmp_param;
434 struct rtable *rt = (struct rtable *)skb_in->dst;
435 struct ipcm_cookie ipc;
443 * Find the original header. It is expected to be valid, of course.
444 * Check this, icmp_send is called from the most obscure devices
447 iph = ip_hdr(skb_in);
449 if ((u8 *)iph < skb_in->head ||
450 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
454 * No replies to physical multicast/broadcast
456 if (skb_in->pkt_type != PACKET_HOST)
460 * Now check at the protocol level
462 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
466 * Only reply to fragment 0. We byte re-order the constant
467 * mask for efficiency.
469 if (iph->frag_off & htons(IP_OFFSET))
473 * If we send an ICMP error to an ICMP error a mess would result..
475 if (icmp_pointers[type].error) {
477 * We are an error, check if we are replying to an
480 if (iph->protocol == IPPROTO_ICMP) {
481 u8 _inner_type, *itp;
483 itp = skb_header_pointer(skb_in,
484 skb_network_header(skb_in) +
486 offsetof(struct icmphdr,
495 * Assume any unknown ICMP type is an error. This
496 * isn't specified by the RFC, but think about it..
498 if (*itp > NR_ICMP_TYPES ||
499 icmp_pointers[*itp].error)
504 if (icmp_xmit_lock())
508 * Construct source address and options.
512 if (!(rt->rt_flags & RTCF_LOCAL)) {
513 struct net_device *dev = NULL;
515 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
516 dev = dev_get_by_index(&init_net, rt->fl.iif);
519 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
525 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
526 IPTOS_PREC_INTERNETCONTROL) :
529 if (ip_options_echo(&icmp_param.replyopts, skb_in))
534 * Prepare data for ICMP header.
537 icmp_param.data.icmph.type = type;
538 icmp_param.data.icmph.code = code;
539 icmp_param.data.icmph.un.gateway = info;
540 icmp_param.data.icmph.checksum = 0;
541 icmp_param.skb = skb_in;
542 icmp_param.offset = skb_network_offset(skb_in);
543 inet_sk(icmp_socket->sk)->tos = tos;
544 ipc.addr = iph->saddr;
545 ipc.opt = &icmp_param.replyopts;
551 .daddr = icmp_param.replyopts.srr ?
552 icmp_param.replyopts.faddr :
558 .proto = IPPROTO_ICMP,
566 security_skb_classify_flow(skb_in, &fl);
567 if (ip_route_output_key(&rt, &fl))
571 if (!icmpv4_xrlim_allow(rt, type, code))
574 /* RFC says return as much as we can without exceeding 576 bytes. */
576 room = dst_mtu(&rt->u.dst);
579 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
580 room -= sizeof(struct icmphdr);
582 icmp_param.data_len = skb_in->len - icmp_param.offset;
583 if (icmp_param.data_len > room)
584 icmp_param.data_len = room;
585 icmp_param.head_len = sizeof(struct icmphdr);
587 icmp_push_reply(&icmp_param, &ipc, rt);
597 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
600 static void icmp_unreach(struct sk_buff *skb)
603 struct icmphdr *icmph;
605 struct net_protocol *ipprot;
609 * Incomplete header ?
610 * Only checks for the IP header, there should be an
611 * additional check for longer headers in upper levels.
614 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
617 icmph = icmp_hdr(skb);
618 iph = (struct iphdr *)skb->data;
620 if (iph->ihl < 5) /* Mangled header, drop. */
623 if (icmph->type == ICMP_DEST_UNREACH) {
624 switch (icmph->code & 15) {
625 case ICMP_NET_UNREACH:
626 case ICMP_HOST_UNREACH:
627 case ICMP_PROT_UNREACH:
628 case ICMP_PORT_UNREACH:
630 case ICMP_FRAG_NEEDED:
631 if (ipv4_config.no_pmtu_disc) {
632 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
633 "fragmentation needed "
635 NIPQUAD(iph->daddr));
637 info = ip_rt_frag_needed(iph,
638 ntohs(icmph->un.frag.mtu));
644 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
646 NIPQUAD(iph->daddr));
651 if (icmph->code > NR_ICMP_UNREACH)
653 } else if (icmph->type == ICMP_PARAMETERPROB)
654 info = ntohl(icmph->un.gateway) >> 24;
657 * Throw it at our lower layers
659 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
661 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
663 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
668 * Check the other end isnt violating RFC 1122. Some routers send
669 * bogus responses to broadcast frames. If you see this message
670 * first check your netmask matches at both ends, if it does then
671 * get the other vendor to fix their kit.
674 if (!sysctl_icmp_ignore_bogus_error_responses &&
675 inet_addr_type(iph->daddr) == RTN_BROADCAST) {
677 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
679 "error to a broadcast: %u.%u.%u.%u on %s\n",
680 NIPQUAD(ip_hdr(skb)->saddr),
681 icmph->type, icmph->code,
687 /* Checkin full IP header plus 8 bytes of protocol to
688 * avoid additional coding at protocol handlers.
690 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
693 iph = (struct iphdr *)skb->data;
694 protocol = iph->protocol;
697 * Deliver ICMP message to raw sockets. Pretty useless feature?
699 raw_icmp_error(skb, protocol, info);
701 hash = protocol & (MAX_INET_PROTOS - 1);
703 ipprot = rcu_dereference(inet_protos[hash]);
704 if (ipprot && ipprot->err_handler)
705 ipprot->err_handler(skb, info);
711 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
717 * Handle ICMP_REDIRECT.
720 static void icmp_redirect(struct sk_buff *skb)
724 if (skb->len < sizeof(struct iphdr))
728 * Get the copied header of the packet that caused the redirect
730 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
733 iph = (struct iphdr *)skb->data;
735 switch (icmp_hdr(skb)->code & 7) {
737 case ICMP_REDIR_NETTOS:
739 * As per RFC recommendations now handle it as a host redirect.
741 case ICMP_REDIR_HOST:
742 case ICMP_REDIR_HOSTTOS:
743 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
744 icmp_hdr(skb)->un.gateway,
745 iph->saddr, skb->dev);
751 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
756 * Handle ICMP_ECHO ("ping") requests.
758 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
760 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
761 * included in the reply.
762 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
763 * echo requests, MUST have default=NOT.
764 * See also WRT handling of options once they are done and working.
767 static void icmp_echo(struct sk_buff *skb)
769 if (!sysctl_icmp_echo_ignore_all) {
770 struct icmp_bxm icmp_param;
772 icmp_param.data.icmph = *icmp_hdr(skb);
773 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
774 icmp_param.skb = skb;
775 icmp_param.offset = 0;
776 icmp_param.data_len = skb->len;
777 icmp_param.head_len = sizeof(struct icmphdr);
778 icmp_reply(&icmp_param, skb);
783 * Handle ICMP Timestamp requests.
784 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
785 * SHOULD be in the kernel for minimum random latency.
786 * MUST be accurate to a few minutes.
787 * MUST be updated at least at 15Hz.
789 static void icmp_timestamp(struct sk_buff *skb)
792 struct icmp_bxm icmp_param;
800 * Fill in the current time as ms since midnight UT:
802 do_gettimeofday(&tv);
803 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
805 icmp_param.data.times[2] = icmp_param.data.times[1];
806 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
808 icmp_param.data.icmph = *icmp_hdr(skb);
809 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
810 icmp_param.data.icmph.code = 0;
811 icmp_param.skb = skb;
812 icmp_param.offset = 0;
813 icmp_param.data_len = 0;
814 icmp_param.head_len = sizeof(struct icmphdr) + 12;
815 icmp_reply(&icmp_param, skb);
819 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
825 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
827 * RFC1122 (3.2.2.9). A host MUST only send replies to
828 * ADDRESS_MASK requests if it's been configured as an address mask
829 * agent. Receiving a request doesn't constitute implicit permission to
830 * act as one. Of course, implementing this correctly requires (SHOULD)
831 * a way to turn the functionality on and off. Another one for sysctl(),
834 * RFC1812 (4.3.3.9). A router MUST implement it.
835 * A router SHOULD have switch turning it on/off.
836 * This switch MUST be ON by default.
838 * Gratuitous replies, zero-source replies are not implemented,
839 * that complies with RFC. DO NOT implement them!!! All the idea
840 * of broadcast addrmask replies as specified in RFC950 is broken.
841 * The problem is that it is not uncommon to have several prefixes
842 * on one physical interface. Moreover, addrmask agent can even be
843 * not aware of existing another prefixes.
844 * If source is zero, addrmask agent cannot choose correct prefix.
845 * Gratuitous mask announcements suffer from the same problem.
846 * RFC1812 explains it, but still allows to use ADDRMASK,
847 * that is pretty silly. --ANK
849 * All these rules are so bizarre, that I removed kernel addrmask
850 * support at all. It is wrong, it is obsolete, nobody uses it in
853 * Furthermore you can do it with a usermode address agent program
857 static void icmp_address(struct sk_buff *skb)
861 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
866 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
867 * loudly if an inconsistency is found.
870 static void icmp_address_reply(struct sk_buff *skb)
872 struct rtable *rt = (struct rtable *)skb->dst;
873 struct net_device *dev = skb->dev;
874 struct in_device *in_dev;
875 struct in_ifaddr *ifa;
877 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
880 in_dev = in_dev_get(dev);
884 if (in_dev->ifa_list &&
885 IN_DEV_LOG_MARTIANS(in_dev) &&
886 IN_DEV_FORWARD(in_dev)) {
889 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
891 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
892 if (*mp == ifa->ifa_mask &&
893 inet_ifa_match(rt->rt_src, ifa))
896 if (!ifa && net_ratelimit()) {
897 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
899 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
907 static void icmp_discard(struct sk_buff *skb)
912 * Deal with incoming ICMP packets.
914 int icmp_rcv(struct sk_buff *skb)
916 struct icmphdr *icmph;
917 struct rtable *rt = (struct rtable *)skb->dst;
919 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
921 switch (skb->ip_summed) {
922 case CHECKSUM_COMPLETE:
923 if (!csum_fold(skb->csum))
928 if (__skb_checksum_complete(skb))
932 if (!pskb_pull(skb, sizeof(struct icmphdr)))
935 icmph = icmp_hdr(skb);
937 ICMPMSGIN_INC_STATS_BH(icmph->type);
939 * 18 is the highest 'known' ICMP type. Anything else is a mystery
941 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
944 if (icmph->type > NR_ICMP_TYPES)
949 * Parse the ICMP message
952 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
954 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
955 * silently ignored (we let user decide with a sysctl).
956 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
957 * discarded if to broadcast/multicast.
959 if ((icmph->type == ICMP_ECHO ||
960 icmph->type == ICMP_TIMESTAMP) &&
961 sysctl_icmp_echo_ignore_broadcasts) {
964 if (icmph->type != ICMP_ECHO &&
965 icmph->type != ICMP_TIMESTAMP &&
966 icmph->type != ICMP_ADDRESS &&
967 icmph->type != ICMP_ADDRESSREPLY) {
972 icmp_pointers[icmph->type].handler(skb);
978 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
983 * This table is the definition of how we handle ICMP.
985 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
987 .handler = icmp_discard,
990 .handler = icmp_discard,
994 .handler = icmp_discard,
997 [ICMP_DEST_UNREACH] = {
998 .handler = icmp_unreach,
1001 [ICMP_SOURCE_QUENCH] = {
1002 .handler = icmp_unreach,
1006 .handler = icmp_redirect,
1010 .handler = icmp_discard,
1014 .handler = icmp_discard,
1018 .handler = icmp_echo,
1021 .handler = icmp_discard,
1025 .handler = icmp_discard,
1028 [ICMP_TIME_EXCEEDED] = {
1029 .handler = icmp_unreach,
1032 [ICMP_PARAMETERPROB] = {
1033 .handler = icmp_unreach,
1036 [ICMP_TIMESTAMP] = {
1037 .handler = icmp_timestamp,
1039 [ICMP_TIMESTAMPREPLY] = {
1040 .handler = icmp_discard,
1042 [ICMP_INFO_REQUEST] = {
1043 .handler = icmp_discard,
1045 [ICMP_INFO_REPLY] = {
1046 .handler = icmp_discard,
1049 .handler = icmp_address,
1051 [ICMP_ADDRESSREPLY] = {
1052 .handler = icmp_address_reply,
1056 void __init icmp_init(struct net_proto_family *ops)
1058 struct inet_sock *inet;
1061 for_each_possible_cpu(i) {
1064 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1065 &per_cpu(__icmp_socket, i));
1068 panic("Failed to create the ICMP control socket.\n");
1070 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1072 /* Enough space for 2 64K ICMP packets, including
1073 * sk_buff struct overhead.
1075 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1076 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1078 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1080 inet->pmtudisc = IP_PMTUDISC_DONT;
1082 /* Unhash it so that IP input processing does not even
1083 * see it, we do not wish this socket to see incoming
1086 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1090 EXPORT_SYMBOL(icmp_err_convert);
1091 EXPORT_SYMBOL(icmp_send);
1092 EXPORT_SYMBOL(icmp_statistics);
1093 EXPORT_SYMBOL(xrlim_allow);