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 */
191 /* Control parameters for ECHO replies. */
192 int sysctl_icmp_echo_ignore_all __read_mostly;
193 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
195 /* Control parameter - ignore bogus broadcast responses? */
196 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
199 * Configurable global rate limit.
201 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
202 * ratemask defines which icmp types are ratelimited by setting
206 * dest unreachable (3), source quench (4),
207 * time exceeded (11), parameter problem (12)
210 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
211 int sysctl_icmp_ratemask __read_mostly = 0x1818;
212 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
215 * ICMP control array. This specifies what to do with each ICMP.
218 struct icmp_control {
219 void (*handler)(struct sk_buff *skb);
220 short error; /* This ICMP is classed as an error message */
223 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
226 * The ICMP socket(s). This is the most convenient way to flow control
227 * our ICMP output as well as maintain a clean interface throughout
228 * all layers. All Socketless IP sends will soon be gone.
230 * On SMP we have one ICMP socket per-cpu.
232 static DEFINE_PER_CPU(struct sock *, __icmp_sk) = NULL;
233 #define icmp_sk __get_cpu_var(__icmp_sk)
235 static inline int icmp_xmit_lock(struct sock *sk)
239 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
240 /* This can happen if the output path signals a
241 * dst_link_failure() for an outgoing ICMP packet.
249 static inline void icmp_xmit_unlock(struct sock *sk)
251 spin_unlock_bh(&sk->sk_lock.slock);
255 * Send an ICMP frame.
259 * Check transmit rate limitation for given message.
260 * The rate information is held in the destination cache now.
261 * This function is generic and could be used for other purposes
262 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
264 * Note that the same dst_entry fields are modified by functions in
265 * route.c too, but these work for packet destinations while xrlim_allow
266 * works for icmp destinations. This means the rate limiting information
267 * for one "ip object" is shared - and these ICMPs are twice limited:
268 * by source and by destination.
270 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
271 * SHOULD allow setting of rate limits
273 * Shared between ICMPv4 and ICMPv6.
275 #define XRLIM_BURST_FACTOR 6
276 int xrlim_allow(struct dst_entry *dst, int timeout)
278 unsigned long now, token = dst->rate_tokens;
282 token += now - dst->rate_last;
283 dst->rate_last = now;
284 if (token > XRLIM_BURST_FACTOR * timeout)
285 token = XRLIM_BURST_FACTOR * timeout;
286 if (token >= timeout) {
290 dst->rate_tokens = token;
294 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
296 struct dst_entry *dst = &rt->u.dst;
299 if (type > NR_ICMP_TYPES)
302 /* Don't limit PMTU discovery. */
303 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
306 /* No rate limit on loopback */
307 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
310 /* Limit if icmp type is enabled in ratemask. */
311 if ((1 << type) & sysctl_icmp_ratemask)
312 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
318 * Maintain the counters used in the SNMP statistics for outgoing ICMP
320 void icmp_out_count(unsigned char type)
322 ICMPMSGOUT_INC_STATS(type);
323 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)
353 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
354 icmp_param->data_len+icmp_param->head_len,
355 icmp_param->head_len,
356 ipc, rt, MSG_DONTWAIT) < 0)
357 ip_flush_pending_frames(sk);
358 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
359 struct icmphdr *icmph = icmp_hdr(skb);
361 struct sk_buff *skb1;
363 skb_queue_walk(&sk->sk_write_queue, skb1) {
364 csum = csum_add(csum, skb1->csum);
366 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
368 icmp_param->head_len, csum);
369 icmph->checksum = csum_fold(csum);
370 skb->ip_summed = CHECKSUM_NONE;
371 ip_push_pending_frames(sk);
376 * Driving logic for building and sending ICMP messages.
379 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
381 struct sock *sk = icmp_sk;
382 struct inet_sock *inet = inet_sk(sk);
383 struct ipcm_cookie ipc;
384 struct rtable *rt = (struct rtable *)skb->dst;
387 if (ip_options_echo(&icmp_param->replyopts, skb))
390 if (icmp_xmit_lock(sk))
393 icmp_param->data.icmph.checksum = 0;
395 inet->tos = ip_hdr(skb)->tos;
396 daddr = ipc.addr = rt->rt_src;
398 if (icmp_param->replyopts.optlen) {
399 ipc.opt = &icmp_param->replyopts;
401 daddr = icmp_param->replyopts.faddr;
404 struct flowi fl = { .nl_u = { .ip4_u =
406 .saddr = rt->rt_spec_dst,
407 .tos = RT_TOS(ip_hdr(skb)->tos) } },
408 .proto = IPPROTO_ICMP };
409 security_skb_classify_flow(skb, &fl);
410 if (ip_route_output_key(rt->u.dst.dev->nd_net, &rt, &fl))
413 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
414 icmp_param->data.icmph.code))
415 icmp_push_reply(icmp_param, &ipc, rt);
418 icmp_xmit_unlock(sk);
423 * Send an ICMP message in response to a situation
425 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
426 * MAY send more (we do).
427 * MUST NOT change this header information.
428 * MUST NOT reply to a multicast/broadcast IP address.
429 * MUST NOT reply to a multicast/broadcast MAC address.
430 * MUST reply to only the first fragment.
433 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
437 struct icmp_bxm icmp_param;
438 struct rtable *rt = (struct rtable *)skb_in->dst;
439 struct ipcm_cookie ipc;
443 struct sock *sk = icmp_sk;
447 net = rt->u.dst.dev->nd_net;
450 * Find the original header. It is expected to be valid, of course.
451 * Check this, icmp_send is called from the most obscure devices
454 iph = ip_hdr(skb_in);
456 if ((u8 *)iph < skb_in->head ||
457 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
461 * No replies to physical multicast/broadcast
463 if (skb_in->pkt_type != PACKET_HOST)
467 * Now check at the protocol level
469 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
473 * Only reply to fragment 0. We byte re-order the constant
474 * mask for efficiency.
476 if (iph->frag_off & htons(IP_OFFSET))
480 * If we send an ICMP error to an ICMP error a mess would result..
482 if (icmp_pointers[type].error) {
484 * We are an error, check if we are replying to an
487 if (iph->protocol == IPPROTO_ICMP) {
488 u8 _inner_type, *itp;
490 itp = skb_header_pointer(skb_in,
491 skb_network_header(skb_in) +
493 offsetof(struct icmphdr,
502 * Assume any unknown ICMP type is an error. This
503 * isn't specified by the RFC, but think about it..
505 if (*itp > NR_ICMP_TYPES ||
506 icmp_pointers[*itp].error)
511 if (icmp_xmit_lock(sk))
515 * Construct source address and options.
519 if (!(rt->rt_flags & RTCF_LOCAL)) {
520 struct net_device *dev = NULL;
522 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
523 dev = dev_get_by_index(net, rt->fl.iif);
526 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
532 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
533 IPTOS_PREC_INTERNETCONTROL) :
536 if (ip_options_echo(&icmp_param.replyopts, skb_in))
541 * Prepare data for ICMP header.
544 icmp_param.data.icmph.type = type;
545 icmp_param.data.icmph.code = code;
546 icmp_param.data.icmph.un.gateway = info;
547 icmp_param.data.icmph.checksum = 0;
548 icmp_param.skb = skb_in;
549 icmp_param.offset = skb_network_offset(skb_in);
550 inet_sk(sk)->tos = tos;
551 ipc.addr = iph->saddr;
552 ipc.opt = &icmp_param.replyopts;
558 .daddr = icmp_param.replyopts.srr ?
559 icmp_param.replyopts.faddr :
565 .proto = IPPROTO_ICMP,
576 security_skb_classify_flow(skb_in, &fl);
577 if (__ip_route_output_key(net, &rt, &fl))
580 /* No need to clone since we're just using its address. */
583 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
596 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
599 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
600 err = __ip_route_output_key(net, &rt2, &fl);
602 struct flowi fl2 = {};
603 struct dst_entry *odst;
605 fl2.fl4_dst = fl.fl4_src;
606 if (ip_route_output_key(net, &rt2, &fl2))
611 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
612 RT_TOS(tos), rt2->u.dst.dev);
614 dst_release(&rt2->u.dst);
615 rt2 = (struct rtable *)skb_in->dst;
622 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
624 if (err == -ENOENT) {
630 dst_release(&rt->u.dst);
638 if (!icmpv4_xrlim_allow(rt, type, code))
641 /* RFC says return as much as we can without exceeding 576 bytes. */
643 room = dst_mtu(&rt->u.dst);
646 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
647 room -= sizeof(struct icmphdr);
649 icmp_param.data_len = skb_in->len - icmp_param.offset;
650 if (icmp_param.data_len > room)
651 icmp_param.data_len = room;
652 icmp_param.head_len = sizeof(struct icmphdr);
654 icmp_push_reply(&icmp_param, &ipc, rt);
658 icmp_xmit_unlock(sk);
664 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
667 static void icmp_unreach(struct sk_buff *skb)
670 struct icmphdr *icmph;
672 struct net_protocol *ipprot;
676 net = skb->dst->dev->nd_net;
679 * Incomplete header ?
680 * Only checks for the IP header, there should be an
681 * additional check for longer headers in upper levels.
684 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
687 icmph = icmp_hdr(skb);
688 iph = (struct iphdr *)skb->data;
690 if (iph->ihl < 5) /* Mangled header, drop. */
693 if (icmph->type == ICMP_DEST_UNREACH) {
694 switch (icmph->code & 15) {
695 case ICMP_NET_UNREACH:
696 case ICMP_HOST_UNREACH:
697 case ICMP_PROT_UNREACH:
698 case ICMP_PORT_UNREACH:
700 case ICMP_FRAG_NEEDED:
701 if (ipv4_config.no_pmtu_disc) {
702 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
703 "fragmentation needed "
705 NIPQUAD(iph->daddr));
707 info = ip_rt_frag_needed(net, iph,
708 ntohs(icmph->un.frag.mtu));
714 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
716 NIPQUAD(iph->daddr));
721 if (icmph->code > NR_ICMP_UNREACH)
723 } else if (icmph->type == ICMP_PARAMETERPROB)
724 info = ntohl(icmph->un.gateway) >> 24;
727 * Throw it at our lower layers
729 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
731 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
733 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
738 * Check the other end isnt violating RFC 1122. Some routers send
739 * bogus responses to broadcast frames. If you see this message
740 * first check your netmask matches at both ends, if it does then
741 * get the other vendor to fix their kit.
744 if (!sysctl_icmp_ignore_bogus_error_responses &&
745 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
747 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
749 "error to a broadcast: %u.%u.%u.%u on %s\n",
750 NIPQUAD(ip_hdr(skb)->saddr),
751 icmph->type, icmph->code,
757 /* Checkin full IP header plus 8 bytes of protocol to
758 * avoid additional coding at protocol handlers.
760 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
763 iph = (struct iphdr *)skb->data;
764 protocol = iph->protocol;
767 * Deliver ICMP message to raw sockets. Pretty useless feature?
769 raw_icmp_error(skb, protocol, info);
771 hash = protocol & (MAX_INET_PROTOS - 1);
773 ipprot = rcu_dereference(inet_protos[hash]);
774 if (ipprot && ipprot->err_handler)
775 ipprot->err_handler(skb, info);
781 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
787 * Handle ICMP_REDIRECT.
790 static void icmp_redirect(struct sk_buff *skb)
794 if (skb->len < sizeof(struct iphdr))
798 * Get the copied header of the packet that caused the redirect
800 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
803 iph = (struct iphdr *)skb->data;
805 switch (icmp_hdr(skb)->code & 7) {
807 case ICMP_REDIR_NETTOS:
809 * As per RFC recommendations now handle it as a host redirect.
811 case ICMP_REDIR_HOST:
812 case ICMP_REDIR_HOSTTOS:
813 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
814 icmp_hdr(skb)->un.gateway,
815 iph->saddr, skb->dev);
821 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
826 * Handle ICMP_ECHO ("ping") requests.
828 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
830 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
831 * included in the reply.
832 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
833 * echo requests, MUST have default=NOT.
834 * See also WRT handling of options once they are done and working.
837 static void icmp_echo(struct sk_buff *skb)
839 if (!sysctl_icmp_echo_ignore_all) {
840 struct icmp_bxm icmp_param;
842 icmp_param.data.icmph = *icmp_hdr(skb);
843 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
844 icmp_param.skb = skb;
845 icmp_param.offset = 0;
846 icmp_param.data_len = skb->len;
847 icmp_param.head_len = sizeof(struct icmphdr);
848 icmp_reply(&icmp_param, skb);
853 * Handle ICMP Timestamp requests.
854 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
855 * SHOULD be in the kernel for minimum random latency.
856 * MUST be accurate to a few minutes.
857 * MUST be updated at least at 15Hz.
859 static void icmp_timestamp(struct sk_buff *skb)
862 struct icmp_bxm icmp_param;
870 * Fill in the current time as ms since midnight UT:
872 do_gettimeofday(&tv);
873 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
875 icmp_param.data.times[2] = icmp_param.data.times[1];
876 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
878 icmp_param.data.icmph = *icmp_hdr(skb);
879 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
880 icmp_param.data.icmph.code = 0;
881 icmp_param.skb = skb;
882 icmp_param.offset = 0;
883 icmp_param.data_len = 0;
884 icmp_param.head_len = sizeof(struct icmphdr) + 12;
885 icmp_reply(&icmp_param, skb);
889 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
895 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
897 * RFC1122 (3.2.2.9). A host MUST only send replies to
898 * ADDRESS_MASK requests if it's been configured as an address mask
899 * agent. Receiving a request doesn't constitute implicit permission to
900 * act as one. Of course, implementing this correctly requires (SHOULD)
901 * a way to turn the functionality on and off. Another one for sysctl(),
904 * RFC1812 (4.3.3.9). A router MUST implement it.
905 * A router SHOULD have switch turning it on/off.
906 * This switch MUST be ON by default.
908 * Gratuitous replies, zero-source replies are not implemented,
909 * that complies with RFC. DO NOT implement them!!! All the idea
910 * of broadcast addrmask replies as specified in RFC950 is broken.
911 * The problem is that it is not uncommon to have several prefixes
912 * on one physical interface. Moreover, addrmask agent can even be
913 * not aware of existing another prefixes.
914 * If source is zero, addrmask agent cannot choose correct prefix.
915 * Gratuitous mask announcements suffer from the same problem.
916 * RFC1812 explains it, but still allows to use ADDRMASK,
917 * that is pretty silly. --ANK
919 * All these rules are so bizarre, that I removed kernel addrmask
920 * support at all. It is wrong, it is obsolete, nobody uses it in
923 * Furthermore you can do it with a usermode address agent program
927 static void icmp_address(struct sk_buff *skb)
931 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
936 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
937 * loudly if an inconsistency is found.
940 static void icmp_address_reply(struct sk_buff *skb)
942 struct rtable *rt = (struct rtable *)skb->dst;
943 struct net_device *dev = skb->dev;
944 struct in_device *in_dev;
945 struct in_ifaddr *ifa;
947 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
950 in_dev = in_dev_get(dev);
954 if (in_dev->ifa_list &&
955 IN_DEV_LOG_MARTIANS(in_dev) &&
956 IN_DEV_FORWARD(in_dev)) {
959 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
961 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
962 if (*mp == ifa->ifa_mask &&
963 inet_ifa_match(rt->rt_src, ifa))
966 if (!ifa && net_ratelimit()) {
967 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
969 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
977 static void icmp_discard(struct sk_buff *skb)
982 * Deal with incoming ICMP packets.
984 int icmp_rcv(struct sk_buff *skb)
986 struct icmphdr *icmph;
987 struct rtable *rt = (struct rtable *)skb->dst;
989 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
992 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
996 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
999 nh = skb_network_offset(skb);
1000 skb_set_network_header(skb, sizeof(*icmph));
1002 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1005 skb_set_network_header(skb, nh);
1008 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1010 switch (skb->ip_summed) {
1011 case CHECKSUM_COMPLETE:
1012 if (!csum_fold(skb->csum))
1017 if (__skb_checksum_complete(skb))
1021 if (!pskb_pull(skb, sizeof(*icmph)))
1024 icmph = icmp_hdr(skb);
1026 ICMPMSGIN_INC_STATS_BH(icmph->type);
1028 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1030 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1033 if (icmph->type > NR_ICMP_TYPES)
1038 * Parse the ICMP message
1041 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1043 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1044 * silently ignored (we let user decide with a sysctl).
1045 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1046 * discarded if to broadcast/multicast.
1048 if ((icmph->type == ICMP_ECHO ||
1049 icmph->type == ICMP_TIMESTAMP) &&
1050 sysctl_icmp_echo_ignore_broadcasts) {
1053 if (icmph->type != ICMP_ECHO &&
1054 icmph->type != ICMP_TIMESTAMP &&
1055 icmph->type != ICMP_ADDRESS &&
1056 icmph->type != ICMP_ADDRESSREPLY) {
1061 icmp_pointers[icmph->type].handler(skb);
1067 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1072 * This table is the definition of how we handle ICMP.
1074 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1075 [ICMP_ECHOREPLY] = {
1076 .handler = icmp_discard,
1079 .handler = icmp_discard,
1083 .handler = icmp_discard,
1086 [ICMP_DEST_UNREACH] = {
1087 .handler = icmp_unreach,
1090 [ICMP_SOURCE_QUENCH] = {
1091 .handler = icmp_unreach,
1095 .handler = icmp_redirect,
1099 .handler = icmp_discard,
1103 .handler = icmp_discard,
1107 .handler = icmp_echo,
1110 .handler = icmp_discard,
1114 .handler = icmp_discard,
1117 [ICMP_TIME_EXCEEDED] = {
1118 .handler = icmp_unreach,
1121 [ICMP_PARAMETERPROB] = {
1122 .handler = icmp_unreach,
1125 [ICMP_TIMESTAMP] = {
1126 .handler = icmp_timestamp,
1128 [ICMP_TIMESTAMPREPLY] = {
1129 .handler = icmp_discard,
1131 [ICMP_INFO_REQUEST] = {
1132 .handler = icmp_discard,
1134 [ICMP_INFO_REPLY] = {
1135 .handler = icmp_discard,
1138 .handler = icmp_address,
1140 [ICMP_ADDRESSREPLY] = {
1141 .handler = icmp_address_reply,
1145 static void __exit icmp_exit(void)
1149 for_each_possible_cpu(i) {
1152 sk = per_cpu(__icmp_sk, i);
1155 per_cpu(__icmp_sk, i) = NULL;
1156 sock_release(sk->sk_socket);
1160 int __init icmp_init(void)
1164 for_each_possible_cpu(i) {
1166 struct socket *sock;
1167 struct inet_sock *inet;
1169 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, &sock);
1173 per_cpu(__icmp_sk, i) = sk = sock->sk;
1174 sk->sk_allocation = GFP_ATOMIC;
1176 /* Enough space for 2 64K ICMP packets, including
1177 * sk_buff struct overhead.
1180 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1184 inet->pmtudisc = IP_PMTUDISC_DONT;
1186 /* Unhash it so that IP input processing does not even
1187 * see it, we do not wish this socket to see incoming
1190 sk->sk_prot->unhash(sk);
1199 EXPORT_SYMBOL(icmp_err_convert);
1200 EXPORT_SYMBOL(icmp_send);
1201 EXPORT_SYMBOL(icmp_statistics);
1202 EXPORT_SYMBOL(xrlim_allow);