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_off; /* Field offset for increment on output */
217 int input_off; /* Field offset 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_FIELD(icmp_pointers[type].output_off);
322 ICMP_INC_STATS(IcmpOutMsgs);
327 * Checksum each fragment, and on the first include the headers and final
330 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);
344 static void icmp_push_reply(struct icmp_bxm *icmp_param,
345 struct ipcm_cookie *ipc, struct rtable *rt)
349 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);
354 if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
355 struct icmphdr *icmph = skb->h.icmph;
356 unsigned int csum = 0;
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_opt *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;
390 icmp_out_count(icmp_param->data.icmph.type);
392 inet->tos = skb->nh.iph->tos;
393 daddr = ipc.addr = rt->rt_src;
395 if (icmp_param->replyopts.optlen) {
396 ipc.opt = &icmp_param->replyopts;
398 daddr = icmp_param->replyopts.faddr;
401 struct flowi fl = { .nl_u = { .ip4_u =
403 .saddr = rt->rt_spec_dst,
404 .tos = RT_TOS(skb->nh.iph->tos) } },
405 .proto = IPPROTO_ICMP };
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);
420 * Send an ICMP message in response to a situation
422 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
423 * MAY send more (we do).
424 * MUST NOT change this header information.
425 * MUST NOT reply to a multicast/broadcast IP address.
426 * MUST NOT reply to a multicast/broadcast MAC address.
427 * MUST reply to only the first fragment.
430 void icmp_send(struct sk_buff *skb_in, int type, int code, u32 info)
434 struct icmp_bxm icmp_param;
435 struct rtable *rt = (struct rtable *)skb_in->dst;
436 struct ipcm_cookie ipc;
444 * Find the original header. It is expected to be valid, of course.
445 * Check this, icmp_send is called from the most obscure devices
448 iph = skb_in->nh.iph;
450 if ((u8 *)iph < skb_in->head || (u8 *)(iph + 1) > 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) {
483 if (skb_copy_bits(skb_in,
484 skb_in->nh.raw + (iph->ihl << 2) +
485 offsetof(struct icmphdr, type) -
486 skb_in->data, &inner_type, 1))
490 * Assume any unknown ICMP type is an error. This
491 * isn't specified by the RFC, but think about it..
493 if (inner_type > NR_ICMP_TYPES ||
494 icmp_pointers[inner_type].error)
499 if (icmp_xmit_lock())
503 * Construct source address and options.
506 #ifdef CONFIG_IP_ROUTE_NAT
508 * Restore original addresses if packet has been translated.
510 if (rt->rt_flags & RTCF_NAT && IPCB(skb_in)->flags & IPSKB_TRANSLATED) {
511 iph->daddr = rt->fl.fl4_dst;
512 iph->saddr = rt->fl.fl4_src;
517 if (!(rt->rt_flags & RTCF_LOCAL))
520 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
521 IPTOS_PREC_INTERNETCONTROL) :
525 struct flowi fl = { .nl_u = { .ip4_u = { .daddr = iph->saddr,
527 .tos = RT_TOS(tos) } },
528 .proto = IPPROTO_ICMP };
530 if (ip_route_output_key(&rt, &fl))
533 if (ip_options_echo(&icmp_param.replyopts, skb_in))
538 * Prepare data for ICMP header.
541 icmp_param.data.icmph.type = type;
542 icmp_param.data.icmph.code = code;
543 icmp_param.data.icmph.un.gateway = info;
544 icmp_param.data.icmph.checksum = 0;
545 icmp_param.skb = skb_in;
546 icmp_param.offset = skb_in->nh.raw - skb_in->data;
547 icmp_out_count(icmp_param.data.icmph.type);
548 inet_sk(icmp_socket->sk)->tos = tos;
549 ipc.addr = iph->saddr;
550 ipc.opt = &icmp_param.replyopts;
551 if (icmp_param.replyopts.srr) {
552 struct flowi fl = { .nl_u = { .ip4_u =
553 { .daddr = icmp_param.replyopts.faddr,
555 .tos = RT_TOS(tos) } },
556 .proto = IPPROTO_ICMP };
558 if (ip_route_output_key(&rt, &fl))
562 if (!icmpv4_xrlim_allow(rt, type, code))
565 /* RFC says return as much as we can without exceeding 576 bytes. */
567 room = dst_pmtu(&rt->u.dst);
570 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
571 room -= sizeof(struct icmphdr);
573 icmp_param.data_len = skb_in->len - icmp_param.offset;
574 if (icmp_param.data_len > room)
575 icmp_param.data_len = room;
576 icmp_param.head_len = sizeof(struct icmphdr);
578 icmp_push_reply(&icmp_param, &ipc, rt);
588 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
591 static void icmp_unreach(struct sk_buff *skb)
594 struct icmphdr *icmph;
596 struct net_protocol *ipprot;
601 * Incomplete header ?
602 * Only checks for the IP header, there should be an
603 * additional check for longer headers in upper levels.
606 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
609 icmph = skb->h.icmph;
610 iph = (struct iphdr *)skb->data;
612 if (iph->ihl < 5) /* Mangled header, drop. */
615 if (icmph->type == ICMP_DEST_UNREACH) {
616 switch (icmph->code & 15) {
617 case ICMP_NET_UNREACH:
618 case ICMP_HOST_UNREACH:
619 case ICMP_PROT_UNREACH:
620 case ICMP_PORT_UNREACH:
622 case ICMP_FRAG_NEEDED:
623 if (ipv4_config.no_pmtu_disc) {
625 printk(KERN_INFO "ICMP: %u.%u.%u.%u: "
626 "fragmentation needed "
628 NIPQUAD(iph->daddr)));
630 info = ip_rt_frag_needed(iph,
631 ntohs(icmph->un.frag.mtu));
638 printk(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
640 NIPQUAD(iph->daddr)));
645 if (icmph->code > NR_ICMP_UNREACH)
647 } else if (icmph->type == ICMP_PARAMETERPROB)
648 info = ntohl(icmph->un.gateway) >> 24;
651 * Throw it at our lower layers
653 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
655 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
657 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
662 * Check the other end isnt violating RFC 1122. Some routers send
663 * bogus responses to broadcast frames. If you see this message
664 * first check your netmask matches at both ends, if it does then
665 * get the other vendor to fix their kit.
668 if (!sysctl_icmp_ignore_bogus_error_responses &&
669 inet_addr_type(iph->daddr) == RTN_BROADCAST) {
671 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
673 "error to a broadcast: %u.%u.%u.%u on %s\n",
674 NIPQUAD(skb->nh.iph->saddr),
675 icmph->type, icmph->code,
681 /* Checkin full IP header plus 8 bytes of protocol to
682 * avoid additional coding at protocol handlers.
684 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
687 iph = (struct iphdr *)skb->data;
688 protocol = iph->protocol;
691 * Deliver ICMP message to raw sockets. Pretty useless feature?
694 /* Note: See raw.c and net/raw.h, RAWV4_HTABLE_SIZE==MAX_INET_PROTOS */
695 hash = protocol & (MAX_INET_PROTOS - 1);
696 read_lock(&raw_v4_lock);
697 if ((raw_sk = sk_head(&raw_v4_htable[hash])) != NULL) {
698 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
700 skb->dev->ifindex)) != NULL) {
701 raw_err(raw_sk, skb, info);
702 raw_sk = sk_next(raw_sk);
703 iph = (struct iphdr *)skb->data;
706 read_unlock(&raw_v4_lock);
709 ipprot = inet_protos[hash];
710 smp_read_barrier_depends();
711 if (ipprot && ipprot->err_handler)
712 ipprot->err_handler(skb, info);
718 ICMP_INC_STATS_BH(IcmpInErrors);
724 * Handle ICMP_REDIRECT.
727 static void icmp_redirect(struct sk_buff *skb)
732 if (skb->len < sizeof(struct iphdr))
736 * Get the copied header of the packet that caused the redirect
738 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
741 iph = (struct iphdr *)skb->data;
744 switch (skb->h.icmph->code & 7) {
746 case ICMP_REDIR_NETTOS:
748 * As per RFC recommendations now handle it as a host redirect.
750 case ICMP_REDIR_HOST:
751 case ICMP_REDIR_HOSTTOS:
752 ip_rt_redirect(skb->nh.iph->saddr, ip, skb->h.icmph->un.gateway,
753 iph->saddr, iph->tos, skb->dev);
759 ICMP_INC_STATS_BH(IcmpInErrors);
764 * Handle ICMP_ECHO ("ping") requests.
766 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
768 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
769 * included in the reply.
770 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
771 * echo requests, MUST have default=NOT.
772 * See also WRT handling of options once they are done and working.
775 static void icmp_echo(struct sk_buff *skb)
777 if (!sysctl_icmp_echo_ignore_all) {
778 struct icmp_bxm icmp_param;
780 icmp_param.data.icmph = *skb->h.icmph;
781 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
782 icmp_param.skb = skb;
783 icmp_param.offset = 0;
784 icmp_param.data_len = skb->len;
785 icmp_param.head_len = sizeof(struct icmphdr);
786 icmp_reply(&icmp_param, skb);
791 * Handle ICMP Timestamp requests.
792 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
793 * SHOULD be in the kernel for minimum random latency.
794 * MUST be accurate to a few minutes.
795 * MUST be updated at least at 15Hz.
797 static void icmp_timestamp(struct sk_buff *skb)
800 struct icmp_bxm icmp_param;
808 * Fill in the current time as ms since midnight UT:
810 do_gettimeofday(&tv);
811 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
813 icmp_param.data.times[2] = icmp_param.data.times[1];
814 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
816 icmp_param.data.icmph = *skb->h.icmph;
817 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
818 icmp_param.data.icmph.code = 0;
819 icmp_param.skb = skb;
820 icmp_param.offset = 0;
821 icmp_param.data_len = 0;
822 icmp_param.head_len = sizeof(struct icmphdr) + 12;
823 icmp_reply(&icmp_param, skb);
827 ICMP_INC_STATS_BH(IcmpInErrors);
833 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
835 * RFC1122 (3.2.2.9). A host MUST only send replies to
836 * ADDRESS_MASK requests if it's been configured as an address mask
837 * agent. Receiving a request doesn't constitute implicit permission to
838 * act as one. Of course, implementing this correctly requires (SHOULD)
839 * a way to turn the functionality on and off. Another one for sysctl(),
842 * RFC1812 (4.3.3.9). A router MUST implement it.
843 * A router SHOULD have switch turning it on/off.
844 * This switch MUST be ON by default.
846 * Gratuitous replies, zero-source replies are not implemented,
847 * that complies with RFC. DO NOT implement them!!! All the idea
848 * of broadcast addrmask replies as specified in RFC950 is broken.
849 * The problem is that it is not uncommon to have several prefixes
850 * on one physical interface. Moreover, addrmask agent can even be
851 * not aware of existing another prefixes.
852 * If source is zero, addrmask agent cannot choose correct prefix.
853 * Gratuitous mask announcements suffer from the same problem.
854 * RFC1812 explains it, but still allows to use ADDRMASK,
855 * that is pretty silly. --ANK
857 * All these rules are so bizarre, that I removed kernel addrmask
858 * support at all. It is wrong, it is obsolete, nobody uses it in
861 * Furthermore you can do it with a usermode address agent program
865 static void icmp_address(struct sk_buff *skb)
869 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
874 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
875 * loudly if an inconsistency is found.
878 static void icmp_address_reply(struct sk_buff *skb)
880 struct rtable *rt = (struct rtable *)skb->dst;
881 struct net_device *dev = skb->dev;
882 struct in_device *in_dev;
883 struct in_ifaddr *ifa;
886 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
889 in_dev = in_dev_get(dev);
892 read_lock(&in_dev->lock);
893 if (in_dev->ifa_list &&
894 IN_DEV_LOG_MARTIANS(in_dev) &&
895 IN_DEV_FORWARD(in_dev)) {
896 if (skb_copy_bits(skb, 0, &mask, 4))
898 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
899 if (mask == ifa->ifa_mask &&
900 inet_ifa_match(rt->rt_src, ifa))
903 if (!ifa && net_ratelimit()) {
904 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
906 NIPQUAD(mask), dev->name, NIPQUAD(rt->rt_src));
909 read_unlock(&in_dev->lock);
914 #ifdef CONFIG_ICMP_IPOD
915 #include <linux/reboot.h>
917 int sysctl_icmp_ipod_version = 2;
918 int sysctl_icmp_ipod_enabled = 0;
919 u32 sysctl_icmp_ipod_host = 0xffffffff;
920 u32 sysctl_icmp_ipod_mask = 0xffffffff;
921 char sysctl_icmp_ipod_key[32+1] = { "SETMETOSOMETHINGTHIRTYTWOBYTES!!" };
922 #define IPOD_CHECK_KEY \
923 (sysctl_icmp_ipod_key[0] != 0)
924 #define IPOD_VALID_KEY(d) \
925 (strncmp(sysctl_icmp_ipod_key, (char *)(d), strlen(sysctl_icmp_ipod_key)) == 0)
927 static void icmp_ping_of_death(struct sk_buff *skb)
929 struct icmphdr *icmph = skb->h.icmph;
930 struct iphdr *iph = skb->nh.iph;
934 printk(KERN_INFO "IPOD: got type=6, code=%d, host=%u.%u.%u.%u\n", icmph->code, ntohs(iph->tot_len), NIPQUAD(iph->saddr));
938 * If IPOD not enabled or wrong ICMP code, ignore.
940 if (!sysctl_icmp_ipod_enabled || icmph->code != 6)
944 * First check the source address info.
945 * If host not set, ignore.
947 if (sysctl_icmp_ipod_host != 0xffffffff &&
948 (ntohl(iph->saddr) & sysctl_icmp_ipod_mask) == sysctl_icmp_ipod_host) {
950 * Now check the key if enabled.
951 * If packet doesn't contain enough data or key
952 * is otherwise invalid, ignore.
954 if (IPOD_CHECK_KEY) {
955 if (pskb_may_pull(skb, sizeof(sysctl_icmp_ipod_key)-1) &&
956 IPOD_VALID_KEY(skb->data))
964 sysctl_icmp_ipod_enabled = 0;
965 printk(KERN_CRIT "IPOD: reboot forced by %u.%u.%u.%u...\n",
966 NIPQUAD(iph->saddr));
967 machine_restart(NULL);
969 printk(KERN_WARNING "IPOD: from %u.%u.%u.%u rejected\n",
970 NIPQUAD(iph->saddr));
975 static void icmp_discard(struct sk_buff *skb)
980 * Deal with incoming ICMP packets.
982 int icmp_rcv(struct sk_buff *skb)
984 struct icmphdr *icmph;
985 struct rtable *rt = (struct rtable *)skb->dst;
987 ICMP_INC_STATS_BH(IcmpInMsgs);
989 switch (skb->ip_summed) {
991 if (!(u16)csum_fold(skb->csum))
993 NETDEBUG(if (net_ratelimit())
994 printk(KERN_DEBUG "icmp v4 hw csum failure\n"));
996 if ((u16)csum_fold(skb_checksum(skb, 0, skb->len, 0)))
1001 if (!pskb_pull(skb, sizeof(struct icmphdr)))
1004 icmph = skb->h.icmph;
1007 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1009 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1012 if (icmph->type > NR_ICMP_TYPES)
1017 * Parse the ICMP message
1020 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1022 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1023 * silently ignored (we let user decide with a sysctl).
1024 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1025 * discarded if to broadcast/multicast.
1027 if (icmph->type == ICMP_ECHO &&
1028 sysctl_icmp_echo_ignore_broadcasts) {
1031 if (icmph->type != ICMP_ECHO &&
1032 icmph->type != ICMP_TIMESTAMP &&
1033 icmph->type != ICMP_ADDRESS &&
1034 icmph->type != ICMP_ADDRESSREPLY) {
1039 ICMP_INC_STATS_BH_FIELD(icmp_pointers[icmph->type].input_off);
1040 icmp_pointers[icmph->type].handler(skb);
1046 ICMP_INC_STATS_BH(IcmpInErrors);
1051 * This table is the definition of how we handle ICMP.
1053 static struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1054 [ICMP_ECHOREPLY] = {
1055 .output_off = offsetof(struct icmp_mib, IcmpOutEchoReps),
1056 .input_off = offsetof(struct icmp_mib, IcmpInEchoReps),
1057 .handler = icmp_discard,
1060 .output_off = offsetof(struct icmp_mib, dummy),
1061 .input_off = offsetof(struct icmp_mib,IcmpInErrors),
1062 .handler = icmp_discard,
1066 .output_off = offsetof(struct icmp_mib, dummy),
1067 .input_off = offsetof(struct icmp_mib,IcmpInErrors),
1068 .handler = icmp_discard,
1071 [ICMP_DEST_UNREACH] = {
1072 .output_off = offsetof(struct icmp_mib, IcmpOutDestUnreachs),
1073 .input_off = offsetof(struct icmp_mib, IcmpInDestUnreachs),
1074 .handler = icmp_unreach,
1077 [ICMP_SOURCE_QUENCH] = {
1078 .output_off = offsetof(struct icmp_mib, IcmpOutSrcQuenchs),
1079 .input_off = offsetof(struct icmp_mib, IcmpInSrcQuenchs),
1080 .handler = icmp_unreach,
1084 .output_off = offsetof(struct icmp_mib, IcmpOutRedirects),
1085 .input_off = offsetof(struct icmp_mib, IcmpInRedirects),
1086 .handler = icmp_redirect,
1089 #ifdef CONFIG_ICMP_IPOD
1091 .output_off = offsetof(struct icmp_mib, dummy),
1092 .input_off = offsetof(struct icmp_mib, dummy),
1093 .handler = icmp_ping_of_death,
1098 .output_off = offsetof(struct icmp_mib, dummy),
1099 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1100 .handler = icmp_discard,
1105 .output_off = offsetof(struct icmp_mib, dummy),
1106 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1107 .handler = icmp_discard,
1111 .output_off = offsetof(struct icmp_mib, IcmpOutEchos),
1112 .input_off = offsetof(struct icmp_mib, IcmpInEchos),
1113 .handler = icmp_echo,
1116 .output_off = offsetof(struct icmp_mib, dummy),
1117 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1118 .handler = icmp_discard,
1122 .output_off = offsetof(struct icmp_mib, dummy),
1123 .input_off = offsetof(struct icmp_mib, IcmpInErrors),
1124 .handler = icmp_discard,
1127 [ICMP_TIME_EXCEEDED] = {
1128 .output_off = offsetof(struct icmp_mib, IcmpOutTimeExcds),
1129 .input_off = offsetof(struct icmp_mib,IcmpInTimeExcds),
1130 .handler = icmp_unreach,
1133 [ICMP_PARAMETERPROB] = {
1134 .output_off = offsetof(struct icmp_mib, IcmpOutParmProbs),
1135 .input_off = offsetof(struct icmp_mib, IcmpInParmProbs),
1136 .handler = icmp_unreach,
1139 [ICMP_TIMESTAMP] = {
1140 .output_off = offsetof(struct icmp_mib, IcmpOutTimestamps),
1141 .input_off = offsetof(struct icmp_mib, IcmpInTimestamps),
1142 .handler = icmp_timestamp,
1144 [ICMP_TIMESTAMPREPLY] = {
1145 .output_off = offsetof(struct icmp_mib, IcmpOutTimestampReps),
1146 .input_off = offsetof(struct icmp_mib, IcmpInTimestampReps),
1147 .handler = icmp_discard,
1149 [ICMP_INFO_REQUEST] = {
1150 .output_off = offsetof(struct icmp_mib, dummy),
1151 .input_off = offsetof(struct icmp_mib, dummy),
1152 .handler = icmp_discard,
1154 [ICMP_INFO_REPLY] = {
1155 .output_off = offsetof(struct icmp_mib, dummy),
1156 .input_off = offsetof(struct icmp_mib, dummy),
1157 .handler = icmp_discard,
1160 .output_off = offsetof(struct icmp_mib, IcmpOutAddrMasks),
1161 .input_off = offsetof(struct icmp_mib, IcmpInAddrMasks),
1162 .handler = icmp_address,
1164 [ICMP_ADDRESSREPLY] = {
1165 .output_off = offsetof(struct icmp_mib, IcmpOutAddrMaskReps),
1166 .input_off = offsetof(struct icmp_mib, IcmpInAddrMaskReps),
1167 .handler = icmp_address_reply,
1171 void __init icmp_init(struct net_proto_family *ops)
1173 struct inet_opt *inet;
1176 for (i = 0; i < NR_CPUS; i++) {
1179 if (!cpu_possible(i))
1182 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1183 &per_cpu(__icmp_socket, i));
1186 panic("Failed to create the ICMP control socket.\n");
1188 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1190 /* Enough space for 2 64K ICMP packets, including
1191 * sk_buff struct overhead.
1193 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1194 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1196 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1198 inet->pmtudisc = IP_PMTUDISC_DONT;
1200 /* Unhash it so that IP input processing does not even
1201 * see it, we do not wish this socket to see incoming
1204 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1208 EXPORT_SYMBOL(icmp_err_convert);
1209 EXPORT_SYMBOL(icmp_send);
1210 EXPORT_SYMBOL(icmp_statistics);
1211 EXPORT_SYMBOL(xrlim_allow);