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/config.h>
68 #include <linux/module.h>
69 #include <linux/types.h>
70 #include <linux/jiffies.h>
71 #include <linux/kernel.h>
72 #include <linux/fcntl.h>
73 #include <linux/socket.h>
75 #include <linux/inet.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);
119 /* An array of errno for error messages from dest unreach. */
120 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
122 struct icmp_err icmp_err_convert[] = {
124 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
128 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
132 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
136 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
140 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
144 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
148 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
152 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
156 .errno = ENONET, /* ICMP_HOST_ISOLATED */
160 .errno = ENETUNREACH, /* ICMP_NET_ANO */
164 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
168 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
172 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
176 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
180 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
184 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
189 /* Control parameters for ECHO replies. */
190 int sysctl_icmp_echo_ignore_all;
191 int sysctl_icmp_echo_ignore_broadcasts;
193 /* Control parameter - ignore bogus broadcast responses? */
194 int sysctl_icmp_ignore_bogus_error_responses;
197 * Configurable global rate limit.
199 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
200 * ratemask defines which icmp types are ratelimited by setting
204 * dest unreachable (3), source quench (4),
205 * time exceeded (11), parameter problem (12)
208 int sysctl_icmp_ratelimit = 1 * HZ;
209 int sysctl_icmp_ratemask = 0x1818;
212 * ICMP control array. This specifies what to do with each ICMP.
215 struct icmp_control {
216 int output_entry; /* Field for increment on output */
217 int input_entry; /* Field for increment on input */
218 void (*handler)(struct sk_buff *skb);
219 short error; /* This ICMP is classed as an error message */
222 static 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 static void icmp_out_count(int type)
320 if (type <= NR_ICMP_TYPES) {
321 ICMP_INC_STATS(icmp_pointers[type].output_entry);
322 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
327 * Checksum each fragment, and on the first include the headers and final
330 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
333 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
336 csum = skb_copy_and_csum_bits(icmp_param->skb,
337 icmp_param->offset + offset,
340 skb->csum = csum_block_add(skb->csum, csum, odd);
341 if (icmp_pointers[icmp_param->data.icmph.type].error)
342 nf_ct_attach(skb, icmp_param->skb);
346 static void icmp_push_reply(struct icmp_bxm *icmp_param,
347 struct ipcm_cookie *ipc, struct rtable *rt)
351 ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
352 icmp_param->data_len+icmp_param->head_len,
353 icmp_param->head_len,
354 ipc, rt, MSG_DONTWAIT);
356 if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
357 struct icmphdr *icmph = skb->h.icmph;
358 unsigned int csum = 0;
359 struct sk_buff *skb1;
361 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
362 csum = csum_add(csum, skb1->csum);
364 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
366 icmp_param->head_len, csum);
367 icmph->checksum = csum_fold(csum);
368 skb->ip_summed = CHECKSUM_NONE;
369 ip_push_pending_frames(icmp_socket->sk);
374 * Driving logic for building and sending ICMP messages.
377 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
379 struct sock *sk = icmp_socket->sk;
380 struct inet_sock *inet = inet_sk(sk);
381 struct ipcm_cookie ipc;
382 struct rtable *rt = (struct rtable *)skb->dst;
385 if (ip_options_echo(&icmp_param->replyopts, skb))
388 if (icmp_xmit_lock())
391 icmp_param->data.icmph.checksum = 0;
392 icmp_out_count(icmp_param->data.icmph.type);
394 inet->tos = skb->nh.iph->tos;
395 daddr = ipc.addr = rt->rt_src;
397 if (icmp_param->replyopts.optlen) {
398 ipc.opt = &icmp_param->replyopts;
400 daddr = icmp_param->replyopts.faddr;
403 struct flowi fl = { .nl_u = { .ip4_u =
405 .saddr = rt->rt_spec_dst,
406 .tos = RT_TOS(skb->nh.iph->tos) } },
407 .proto = IPPROTO_ICMP };
408 if (ip_route_output_key(&rt, &fl))
411 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
412 icmp_param->data.icmph.code))
413 icmp_push_reply(icmp_param, &ipc, rt);
422 * Send an ICMP message in response to a situation
424 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
425 * MAY send more (we do).
426 * MUST NOT change this header information.
427 * MUST NOT reply to a multicast/broadcast IP address.
428 * MUST NOT reply to a multicast/broadcast MAC address.
429 * MUST reply to only the first fragment.
432 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
436 struct icmp_bxm icmp_param;
437 struct rtable *rt = (struct rtable *)skb_in->dst;
438 struct ipcm_cookie ipc;
446 * Find the original header. It is expected to be valid, of course.
447 * Check this, icmp_send is called from the most obscure devices
450 iph = skb_in->nh.iph;
452 if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > skb_in->tail)
456 * No replies to physical multicast/broadcast
458 if (skb_in->pkt_type != PACKET_HOST)
462 * Now check at the protocol level
464 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
468 * Only reply to fragment 0. We byte re-order the constant
469 * mask for efficiency.
471 if (iph->frag_off & htons(IP_OFFSET))
475 * If we send an ICMP error to an ICMP error a mess would result..
477 if (icmp_pointers[type].error) {
479 * We are an error, check if we are replying to an
482 if (iph->protocol == IPPROTO_ICMP) {
483 u8 _inner_type, *itp;
485 itp = skb_header_pointer(skb_in,
488 offsetof(struct icmphdr,
497 * Assume any unknown ICMP type is an error. This
498 * isn't specified by the RFC, but think about it..
500 if (*itp > NR_ICMP_TYPES ||
501 icmp_pointers[*itp].error)
506 if (icmp_xmit_lock())
510 * Construct source address and options.
514 if (!(rt->rt_flags & RTCF_LOCAL))
517 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
518 IPTOS_PREC_INTERNETCONTROL) :
522 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = iph->saddr,
524 .tos = RT_TOS(tos) } },
525 .proto = IPPROTO_ICMP };
526 if (ip_route_output_key(&rt, &fl))
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_in->nh.raw - skb_in->data;
543 icmp_out_count(icmp_param.data.icmph.type);
544 inet_sk(icmp_socket->sk)->tos = tos;
545 ipc.addr = iph->saddr;
546 ipc.opt = &icmp_param.replyopts;
547 if (icmp_param.replyopts.srr) {
548 struct flowi fl = { .nl_u = { .ip4_u =
549 { .daddr = icmp_param.replyopts.faddr,
551 .tos = RT_TOS(tos) } },
552 .proto = IPPROTO_ICMP };
554 if (ip_route_output_key(&rt, &fl))
558 if (!icmpv4_xrlim_allow(rt, type, code))
561 /* RFC says return as much as we can without exceeding 576 bytes. */
563 room = dst_pmtu(&rt->u.dst);
566 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
567 room -= sizeof(struct icmphdr);
569 icmp_param.data_len = skb_in->len - icmp_param.offset;
570 if (icmp_param.data_len > room)
571 icmp_param.data_len = room;
572 icmp_param.head_len = sizeof(struct icmphdr);
574 icmp_push_reply(&icmp_param, &ipc, rt);
584 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
587 static void icmp_unreach(struct sk_buff *skb)
590 struct icmphdr *icmph;
592 struct net_protocol *ipprot;
597 * Incomplete header ?
598 * Only checks for the IP header, there should be an
599 * additional check for longer headers in upper levels.
602 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
605 icmph = skb->h.icmph;
606 iph = (struct iphdr *)skb->data;
608 if (iph->ihl < 5) /* Mangled header, drop. */
611 if (icmph->type == ICMP_DEST_UNREACH) {
612 switch (icmph->code & 15) {
613 case ICMP_NET_UNREACH:
614 case ICMP_HOST_UNREACH:
615 case ICMP_PROT_UNREACH:
616 case ICMP_PORT_UNREACH:
618 case ICMP_FRAG_NEEDED:
619 if (ipv4_config.no_pmtu_disc) {
621 printk(KERN_INFO "ICMP: %u.%u.%u.%u: "
622 "fragmentation needed "
624 NIPQUAD(iph->daddr)));
626 info = ip_rt_frag_needed(iph,
627 ntohs(icmph->un.frag.mtu));
634 printk(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
636 NIPQUAD(iph->daddr)));
641 if (icmph->code > NR_ICMP_UNREACH)
643 } else if (icmph->type == ICMP_PARAMETERPROB)
644 info = ntohl(icmph->un.gateway) >> 24;
647 * Throw it at our lower layers
649 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
651 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
653 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
658 * Check the other end isnt violating RFC 1122. Some routers send
659 * bogus responses to broadcast frames. If you see this message
660 * first check your netmask matches at both ends, if it does then
661 * get the other vendor to fix their kit.
664 if (!sysctl_icmp_ignore_bogus_error_responses &&
665 inet_addr_type(iph->daddr) == RTN_BROADCAST) {
667 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
669 "error to a broadcast: %u.%u.%u.%u on %s\n",
670 NIPQUAD(skb->nh.iph->saddr),
671 icmph->type, icmph->code,
677 /* Checkin full IP header plus 8 bytes of protocol to
678 * avoid additional coding at protocol handlers.
680 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
683 iph = (struct iphdr *)skb->data;
684 protocol = iph->protocol;
687 * Deliver ICMP message to raw sockets. Pretty useless feature?
690 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
691 hash = protocol & (MAX_INET_PROTOS - 1);
692 read_lock(&raw_v4_lock);
693 if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
694 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
696 skb->dev->ifindex)) != NULL) {
697 raw_err(raw_sk, skb, info);
698 raw_sk = sk_next(raw_sk);
699 iph = (struct iphdr *)skb->data;
702 read_unlock(&raw_v4_lock);
705 ipprot = rcu_dereference(inet_protos[hash]);
706 if (ipprot && ipprot->err_handler)
707 ipprot->err_handler(skb, info);
713 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
719 * Handle ICMP_REDIRECT.
722 static void icmp_redirect(struct sk_buff *skb)
727 if (skb->len < sizeof(struct iphdr))
731 * Get the copied header of the packet that caused the redirect
733 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
736 iph = (struct iphdr *)skb->data;
739 switch (skb->h.icmph->code & 7) {
741 case ICMP_REDIR_NETTOS:
743 * As per RFC recommendations now handle it as a host redirect.
745 case ICMP_REDIR_HOST:
746 case ICMP_REDIR_HOSTTOS:
747 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway,
748 iph->saddr, iph->tos, skb->dev);
754 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
759 * Handle ICMP_ECHO ("ping") requests.
761 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
763 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
764 * included in the reply.
765 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
766 * echo requests, MUST have default=NOT.
767 * See also WRT handling of options once they are done and working.
770 static void icmp_echo(struct sk_buff *skb)
772 if (!sysctl_icmp_echo_ignore_all) {
773 struct icmp_bxm icmp_param;
775 icmp_param.data.icmph = *skb->h.icmph;
776 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
777 icmp_param.skb = skb;
778 icmp_param.offset = 0;
779 icmp_param.data_len = skb->len;
780 icmp_param.head_len = sizeof(struct icmphdr);
781 icmp_reply(&icmp_param, skb);
786 * Handle ICMP Timestamp requests.
787 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
788 * SHOULD be in the kernel for minimum random latency.
789 * MUST be accurate to a few minutes.
790 * MUST be updated at least at 15Hz.
792 static void icmp_timestamp(struct sk_buff *skb)
795 struct icmp_bxm icmp_param;
803 * Fill in the current time as ms since midnight UT:
805 do_gettimeofday(&tv);
806 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
808 icmp_param.data.times[2] = icmp_param.data.times[1];
809 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
811 icmp_param.data.icmph = *skb->h.icmph;
812 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
813 icmp_param.data.icmph.code = 0;
814 icmp_param.skb = skb;
815 icmp_param.offset = 0;
816 icmp_param.data_len = 0;
817 icmp_param.head_len = sizeof(struct icmphdr) + 12;
818 icmp_reply(&icmp_param, skb);
822 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
828 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
830 * RFC1122 (3.2.2.9). A host MUST only send replies to
831 * ADDRESS_MASK requests if it's been configured as an address mask
832 * agent. Receiving a request doesn't constitute implicit permission to
833 * act as one. Of course, implementing this correctly requires (SHOULD)
834 * a way to turn the functionality on and off. Another one for sysctl(),
837 * RFC1812 (4.3.3.9). A router MUST implement it.
838 * A router SHOULD have switch turning it on/off.
839 * This switch MUST be ON by default.
841 * Gratuitous replies, zero-source replies are not implemented,
842 * that complies with RFC. DO NOT implement them!!! All the idea
843 * of broadcast addrmask replies as specified in RFC950 is broken.
844 * The problem is that it is not uncommon to have several prefixes
845 * on one physical interface. Moreover, addrmask agent can even be
846 * not aware of existing another prefixes.
847 * If source is zero, addrmask agent cannot choose correct prefix.
848 * Gratuitous mask announcements suffer from the same problem.
849 * RFC1812 explains it, but still allows to use ADDRMASK,
850 * that is pretty silly. --ANK
852 * All these rules are so bizarre, that I removed kernel addrmask
853 * support at all. It is wrong, it is obsolete, nobody uses it in
856 * Furthermore you can do it with a usermode address agent program
860 static void icmp_address(struct sk_buff *skb)
864 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
869 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
870 * loudly if an inconsistency is found.
873 static void icmp_address_reply(struct sk_buff *skb)
875 struct rtable *rt = (struct rtable *)skb->dst;
876 struct net_device *dev = skb->dev;
877 struct in_device *in_dev;
878 struct in_ifaddr *ifa;
880 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
883 in_dev = in_dev_get(dev);
887 if (in_dev->ifa_list &&
888 IN_DEV_LOG_MARTIANS(in_dev) &&
889 IN_DEV_FORWARD(in_dev)) {
892 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
895 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
896 if (*mp == ifa->ifa_mask &&
897 inet_ifa_match(rt->rt_src, ifa))
900 if (!ifa && net_ratelimit()) {
901 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
903 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
911 static void icmp_discard(struct sk_buff *skb)
916 * Deal with incoming ICMP packets.
918 int icmp_rcv(struct sk_buff *skb)
920 struct icmphdr *icmph;
921 struct rtable *rt = (struct rtable *)skb->dst;
923 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
925 switch (skb->ip_summed) {
927 if (!(u16)csum_fold(skb->csum))
929 NETDEBUG(if (net_ratelimit())
930 printk(KERN_DEBUG "icmp v4 hw csum failure\n"));
932 if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0)))
937 if (!pskb_pull(skb, sizeof(struct icmphdr)))
940 icmph = skb->h.icmph;
943 * 18 is the highest 'known' ICMP type. Anything else is a mystery
945 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
948 if (icmph->type > NR_ICMP_TYPES)
953 * Parse the ICMP message
956 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
958 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
959 * silently ignored (we let user decide with a sysctl).
960 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
961 * discarded if to broadcast/multicast.
963 if (icmph->type == ICMP_ECHO &&
964 sysctl_icmp_echo_ignore_broadcasts) {
967 if (icmph->type != ICMP_ECHO &&
968 icmph->type != ICMP_TIMESTAMP &&
969 icmph->type != ICMP_ADDRESS &&
970 icmph->type != ICMP_ADDRESSREPLY) {
975 ICMP_INC_STATS_BH(icmp_pointers[icmph->type].input_entry);
976 icmp_pointers[icmph->type].handler(skb);
982 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
987 * This table is the definition of how we handle ICMP.
989 static struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
991 .output_entry = ICMP_MIB_OUTECHOREPS,
992 .input_entry = ICMP_MIB_INECHOREPS,
993 .handler = icmp_discard,
996 .output_entry = ICMP_MIB_DUMMY,
997 .input_entry = ICMP_MIB_INERRORS,
998 .handler = icmp_discard,
1002 .output_entry = ICMP_MIB_DUMMY,
1003 .input_entry = ICMP_MIB_INERRORS,
1004 .handler = icmp_discard,
1007 [ICMP_DEST_UNREACH] = {
1008 .output_entry = ICMP_MIB_OUTDESTUNREACHS,
1009 .input_entry = ICMP_MIB_INDESTUNREACHS,
1010 .handler = icmp_unreach,
1013 [ICMP_SOURCE_QUENCH] = {
1014 .output_entry = ICMP_MIB_OUTSRCQUENCHS,
1015 .input_entry = ICMP_MIB_INSRCQUENCHS,
1016 .handler = icmp_unreach,
1020 .output_entry = ICMP_MIB_OUTREDIRECTS,
1021 .input_entry = ICMP_MIB_INREDIRECTS,
1022 .handler = icmp_redirect,
1026 .output_entry = ICMP_MIB_DUMMY,
1027 .input_entry = ICMP_MIB_INERRORS,
1028 .handler = icmp_discard,
1032 .output_entry = ICMP_MIB_DUMMY,
1033 .input_entry = ICMP_MIB_INERRORS,
1034 .handler = icmp_discard,
1038 .output_entry = ICMP_MIB_OUTECHOS,
1039 .input_entry = ICMP_MIB_INECHOS,
1040 .handler = icmp_echo,
1043 .output_entry = ICMP_MIB_DUMMY,
1044 .input_entry = ICMP_MIB_INERRORS,
1045 .handler = icmp_discard,
1049 .output_entry = ICMP_MIB_DUMMY,
1050 .input_entry = ICMP_MIB_INERRORS,
1051 .handler = icmp_discard,
1054 [ICMP_TIME_EXCEEDED] = {
1055 .output_entry = ICMP_MIB_OUTTIMEEXCDS,
1056 .input_entry = ICMP_MIB_INTIMEEXCDS,
1057 .handler = icmp_unreach,
1060 [ICMP_PARAMETERPROB] = {
1061 .output_entry = ICMP_MIB_OUTPARMPROBS,
1062 .input_entry = ICMP_MIB_INPARMPROBS,
1063 .handler = icmp_unreach,
1066 [ICMP_TIMESTAMP] = {
1067 .output_entry = ICMP_MIB_OUTTIMESTAMPS,
1068 .input_entry = ICMP_MIB_INTIMESTAMPS,
1069 .handler = icmp_timestamp,
1071 [ICMP_TIMESTAMPREPLY] = {
1072 .output_entry = ICMP_MIB_OUTTIMESTAMPREPS,
1073 .input_entry = ICMP_MIB_INTIMESTAMPREPS,
1074 .handler = icmp_discard,
1076 [ICMP_INFO_REQUEST] = {
1077 .output_entry = ICMP_MIB_DUMMY,
1078 .input_entry = ICMP_MIB_DUMMY,
1079 .handler = icmp_discard,
1081 [ICMP_INFO_REPLY] = {
1082 .output_entry = ICMP_MIB_DUMMY,
1083 .input_entry = ICMP_MIB_DUMMY,
1084 .handler = icmp_discard,
1087 .output_entry = ICMP_MIB_OUTADDRMASKS,
1088 .input_entry = ICMP_MIB_INADDRMASKS,
1089 .handler = icmp_address,
1091 [ICMP_ADDRESSREPLY] = {
1092 .output_entry = ICMP_MIB_OUTADDRMASKREPS,
1093 .input_entry = ICMP_MIB_INADDRMASKREPS,
1094 .handler = icmp_address_reply,
1098 void __init icmp_init(struct net_proto_family *ops)
1100 struct inet_sock *inet;
1103 for (i = 0; i < NR_CPUS; i++) {
1106 if (!cpu_possible(i))
1109 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1110 &per_cpu(__icmp_socket, i));
1113 panic("Failed to create the ICMP control socket.\n");
1115 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1117 /* Enough space for 2 64K ICMP packets, including
1118 * sk_buff struct overhead.
1120 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1121 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1123 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1125 inet->pmtudisc = IP_PMTUDISC_DONT;
1127 /* Unhash it so that IP input processing does not even
1128 * see it, we do not wish this socket to see incoming
1131 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1135 EXPORT_SYMBOL(icmp_err_convert);
1136 EXPORT_SYMBOL(icmp_send);
1137 EXPORT_SYMBOL(icmp_statistics);
1138 EXPORT_SYMBOL(xrlim_allow);