2 * NET3: Implementation of the ICMP protocol layer.
4 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 * Some of the function names and the icmp unreach table for this
12 * module were derived from [icmp.c 1.0.11 06/02/93] by
13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14 * Other than that this module is a complete rewrite.
17 * Clemens Fruhwirth : introduce global icmp rate limiting
18 * with icmp type masking ability instead
19 * of broken per type icmp timeouts.
20 * Mike Shaver : RFC1122 checks.
21 * Alan Cox : Multicast ping reply as self.
22 * Alan Cox : Fix atomicity lockup in ip_build_xmit
24 * Alan Cox : Added 216,128 byte paths to the MTU
26 * Martin Mares : RFC1812 checks.
27 * Martin Mares : Can be configured to follow redirects
28 * if acting as a router _without_ a
29 * routing protocol (RFC 1812).
30 * Martin Mares : Echo requests may be configured to
31 * be ignored (RFC 1812).
32 * Martin Mares : Limitation of ICMP error message
33 * transmit rate (RFC 1812).
34 * Martin Mares : TOS and Precedence set correctly
36 * Martin Mares : Now copying as much data from the
37 * original packet as we can without
38 * exceeding 576 bytes (RFC 1812).
39 * Willy Konynenberg : Transparent proxying support.
40 * Keith Owens : RFC1191 correction for 4.2BSD based
42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
44 * Andi Kleen : Check all packet lengths properly
45 * and moved all kfree_skb() up to
47 * Andi Kleen : Move the rate limit bookkeeping
48 * into the dest entry and use a token
49 * bucket filter (thanks to ANK). Make
50 * the rates sysctl configurable.
51 * Yu Tianli : Fixed two ugly bugs in icmp_send
52 * - IP option length was accounted wrongly
53 * - ICMP header length was not accounted
55 * Tristan Greaves : Added sysctl option to ignore bogus
56 * broadcast responses from broken routers.
60 * - Should use skb_pull() instead of all the manual checking.
61 * This would also greatly simply some upper layer error handlers. --AK
65 #include <linux/module.h>
66 #include <linux/types.h>
67 #include <linux/jiffies.h>
68 #include <linux/kernel.h>
69 #include <linux/fcntl.h>
70 #include <linux/socket.h>
72 #include <linux/inet.h>
73 #include <linux/inetdevice.h>
74 #include <linux/netdevice.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_ipv4.h>
77 #include <linux/slab.h>
80 #include <net/route.h>
81 #include <net/protocol.h>
86 #include <linux/skbuff.h>
88 #include <linux/errno.h>
89 #include <linux/timer.h>
90 #include <linux/init.h>
91 #include <asm/system.h>
92 #include <asm/uaccess.h>
93 #include <net/checksum.h>
95 #include <net/inet_common.h>
98 * Build xmit assembly blocks
107 struct icmphdr icmph;
111 struct ip_options replyopts;
112 unsigned char optbuf[40];
115 /* An array of errno for error messages from dest unreach. */
116 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
118 const struct icmp_err icmp_err_convert[] = {
120 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
124 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
128 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
132 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
136 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
140 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
144 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
148 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
152 .errno = ENONET, /* ICMP_HOST_ISOLATED */
156 .errno = ENETUNREACH, /* ICMP_NET_ANO */
160 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
164 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
168 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
172 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
176 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
180 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
184 EXPORT_SYMBOL(icmp_err_convert);
187 * ICMP control array. This specifies what to do with each ICMP.
190 struct icmp_control {
191 void (*handler)(struct sk_buff *skb);
192 short error; /* This ICMP is classed as an error message */
195 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198 * The ICMP socket(s). This is the most convenient way to flow control
199 * our ICMP output as well as maintain a clean interface throughout
200 * all layers. All Socketless IP sends will soon be gone.
202 * On SMP we have one ICMP socket per-cpu.
204 static struct sock *icmp_sk(struct net *net)
206 return net->ipv4.icmp_sk[smp_processor_id()];
209 static inline struct sock *icmp_xmit_lock(struct net *net)
217 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
218 /* This can happen if the output path signals a
219 * dst_link_failure() for an outgoing ICMP packet.
227 static inline void icmp_xmit_unlock(struct sock *sk)
229 spin_unlock_bh(&sk->sk_lock.slock);
233 * Send an ICMP frame.
236 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 struct dst_entry *dst = &rt->dst;
242 if (type > NR_ICMP_TYPES)
245 /* Don't limit PMTU discovery. */
246 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 /* No rate limit on loopback */
250 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 /* Limit if icmp type is enabled in ratemask. */
254 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 rc = inet_peer_xrlim_allow(rt->peer,
258 net->ipv4.sysctl_icmp_ratelimit);
265 * Maintain the counters used in the SNMP statistics for outgoing ICMP
267 void icmp_out_count(struct net *net, unsigned char type)
269 ICMPMSGOUT_INC_STATS(net, type);
270 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
274 * Checksum each fragment, and on the first include the headers and final
277 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
280 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
283 csum = skb_copy_and_csum_bits(icmp_param->skb,
284 icmp_param->offset + offset,
287 skb->csum = csum_block_add(skb->csum, csum, odd);
288 if (icmp_pointers[icmp_param->data.icmph.type].error)
289 nf_ct_attach(skb, icmp_param->skb);
293 static void icmp_push_reply(struct icmp_bxm *icmp_param,
294 struct ipcm_cookie *ipc, struct rtable **rt)
299 sk = icmp_sk(dev_net((*rt)->dst.dev));
300 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
301 icmp_param->data_len+icmp_param->head_len,
302 icmp_param->head_len,
303 ipc, rt, MSG_DONTWAIT) < 0) {
304 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
305 ip_flush_pending_frames(sk);
306 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
307 struct icmphdr *icmph = icmp_hdr(skb);
309 struct sk_buff *skb1;
311 skb_queue_walk(&sk->sk_write_queue, skb1) {
312 csum = csum_add(csum, skb1->csum);
314 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
316 icmp_param->head_len, csum);
317 icmph->checksum = csum_fold(csum);
318 skb->ip_summed = CHECKSUM_NONE;
319 ip_push_pending_frames(sk);
324 * Driving logic for building and sending ICMP messages.
327 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
329 struct ipcm_cookie ipc;
330 struct rtable *rt = skb_rtable(skb);
331 struct net *net = dev_net(rt->dst.dev);
333 struct inet_sock *inet;
336 if (ip_options_echo(&icmp_param->replyopts, skb))
339 sk = icmp_xmit_lock(net);
344 icmp_param->data.icmph.checksum = 0;
346 inet->tos = ip_hdr(skb)->tos;
347 daddr = ipc.addr = rt->rt_src;
350 if (icmp_param->replyopts.optlen) {
351 ipc.opt = &icmp_param->replyopts;
353 daddr = icmp_param->replyopts.faddr;
356 struct flowi fl = { .fl4_dst= daddr,
357 .fl4_src = rt->rt_spec_dst,
358 .fl4_tos = RT_TOS(ip_hdr(skb)->tos),
359 .proto = IPPROTO_ICMP };
360 security_skb_classify_flow(skb, &fl);
361 if (ip_route_output_key(net, &rt, &fl))
364 if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
365 icmp_param->data.icmph.code))
366 icmp_push_reply(icmp_param, &ipc, &rt);
369 icmp_xmit_unlock(sk);
374 * Send an ICMP message in response to a situation
376 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
377 * MAY send more (we do).
378 * MUST NOT change this header information.
379 * MUST NOT reply to a multicast/broadcast IP address.
380 * MUST NOT reply to a multicast/broadcast MAC address.
381 * MUST reply to only the first fragment.
384 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
388 struct icmp_bxm icmp_param;
389 struct rtable *rt = skb_rtable(skb_in);
390 struct ipcm_cookie ipc;
398 net = dev_net(rt->dst.dev);
401 * Find the original header. It is expected to be valid, of course.
402 * Check this, icmp_send is called from the most obscure devices
405 iph = ip_hdr(skb_in);
407 if ((u8 *)iph < skb_in->head ||
408 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
412 * No replies to physical multicast/broadcast
414 if (skb_in->pkt_type != PACKET_HOST)
418 * Now check at the protocol level
420 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
424 * Only reply to fragment 0. We byte re-order the constant
425 * mask for efficiency.
427 if (iph->frag_off & htons(IP_OFFSET))
431 * If we send an ICMP error to an ICMP error a mess would result..
433 if (icmp_pointers[type].error) {
435 * We are an error, check if we are replying to an
438 if (iph->protocol == IPPROTO_ICMP) {
439 u8 _inner_type, *itp;
441 itp = skb_header_pointer(skb_in,
442 skb_network_header(skb_in) +
444 offsetof(struct icmphdr,
453 * Assume any unknown ICMP type is an error. This
454 * isn't specified by the RFC, but think about it..
456 if (*itp > NR_ICMP_TYPES ||
457 icmp_pointers[*itp].error)
462 sk = icmp_xmit_lock(net);
467 * Construct source address and options.
471 if (!(rt->rt_flags & RTCF_LOCAL)) {
472 struct net_device *dev = NULL;
475 if (rt_is_input_route(rt) &&
476 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
477 dev = dev_get_by_index_rcu(net, rt->fl.iif);
480 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
486 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
487 IPTOS_PREC_INTERNETCONTROL) :
490 if (ip_options_echo(&icmp_param.replyopts, skb_in))
495 * Prepare data for ICMP header.
498 icmp_param.data.icmph.type = type;
499 icmp_param.data.icmph.code = code;
500 icmp_param.data.icmph.un.gateway = info;
501 icmp_param.data.icmph.checksum = 0;
502 icmp_param.skb = skb_in;
503 icmp_param.offset = skb_network_offset(skb_in);
504 inet_sk(sk)->tos = tos;
505 ipc.addr = iph->saddr;
506 ipc.opt = &icmp_param.replyopts;
511 .fl4_dst = icmp_param.replyopts.srr ?
512 icmp_param.replyopts.faddr : iph->saddr,
514 .fl4_tos = RT_TOS(tos),
515 .proto = IPPROTO_ICMP,
516 .fl_icmp_type = type,
517 .fl_icmp_code = code,
522 security_skb_classify_flow(skb_in, &fl);
523 if (__ip_route_output_key(net, &rt, &fl))
526 /* No need to clone since we're just using its address. */
529 if (!fl.nl_u.ip4_u.saddr)
530 fl.nl_u.ip4_u.saddr = rt->rt_src;
532 err = xfrm_lookup(net, (struct dst_entry **)&rt, &fl, NULL, 0);
545 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
546 goto relookup_failed;
548 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
549 err = __ip_route_output_key(net, &rt2, &fl);
551 struct flowi fl2 = {};
552 unsigned long orefdst;
554 fl2.fl4_dst = fl.fl4_src;
555 if (ip_route_output_key(net, &rt2, &fl2))
556 goto relookup_failed;
559 orefdst = skb_in->_skb_refdst; /* save old refdst */
560 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
561 RT_TOS(tos), rt2->dst.dev);
563 dst_release(&rt2->dst);
564 rt2 = skb_rtable(skb_in);
565 skb_in->_skb_refdst = orefdst; /* restore old refdst */
569 goto relookup_failed;
571 err = xfrm_lookup(net, (struct dst_entry **)&rt2, &fl, NULL,
575 dst_release(&rt->dst);
589 if (!icmpv4_xrlim_allow(net, rt, type, code))
592 /* RFC says return as much as we can without exceeding 576 bytes. */
594 room = dst_mtu(&rt->dst);
597 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
598 room -= sizeof(struct icmphdr);
600 icmp_param.data_len = skb_in->len - icmp_param.offset;
601 if (icmp_param.data_len > room)
602 icmp_param.data_len = room;
603 icmp_param.head_len = sizeof(struct icmphdr);
605 icmp_push_reply(&icmp_param, &ipc, &rt);
609 icmp_xmit_unlock(sk);
612 EXPORT_SYMBOL(icmp_send);
616 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
619 static void icmp_unreach(struct sk_buff *skb)
622 struct icmphdr *icmph;
624 const struct net_protocol *ipprot;
628 net = dev_net(skb_dst(skb)->dev);
631 * Incomplete header ?
632 * Only checks for the IP header, there should be an
633 * additional check for longer headers in upper levels.
636 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
639 icmph = icmp_hdr(skb);
640 iph = (struct iphdr *)skb->data;
642 if (iph->ihl < 5) /* Mangled header, drop. */
645 if (icmph->type == ICMP_DEST_UNREACH) {
646 switch (icmph->code & 15) {
647 case ICMP_NET_UNREACH:
648 case ICMP_HOST_UNREACH:
649 case ICMP_PROT_UNREACH:
650 case ICMP_PORT_UNREACH:
652 case ICMP_FRAG_NEEDED:
653 if (ipv4_config.no_pmtu_disc) {
654 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: fragmentation needed and DF set.\n",
657 info = ip_rt_frag_needed(net, iph,
658 ntohs(icmph->un.frag.mtu),
665 LIMIT_NETDEBUG(KERN_INFO "ICMP: %pI4: Source Route Failed.\n",
671 if (icmph->code > NR_ICMP_UNREACH)
673 } else if (icmph->type == ICMP_PARAMETERPROB)
674 info = ntohl(icmph->un.gateway) >> 24;
677 * Throw it at our lower layers
679 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
681 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
683 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
688 * Check the other end isnt violating RFC 1122. Some routers send
689 * bogus responses to broadcast frames. If you see this message
690 * first check your netmask matches at both ends, if it does then
691 * get the other vendor to fix their kit.
694 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
695 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
697 printk(KERN_WARNING "%pI4 sent an invalid ICMP "
699 "error to a broadcast: %pI4 on %s\n",
701 icmph->type, icmph->code,
707 /* Checkin full IP header plus 8 bytes of protocol to
708 * avoid additional coding at protocol handlers.
710 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
713 iph = (struct iphdr *)skb->data;
714 protocol = iph->protocol;
717 * Deliver ICMP message to raw sockets. Pretty useless feature?
719 raw_icmp_error(skb, protocol, info);
721 hash = protocol & (MAX_INET_PROTOS - 1);
723 ipprot = rcu_dereference(inet_protos[hash]);
724 if (ipprot && ipprot->err_handler)
725 ipprot->err_handler(skb, info);
731 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
737 * Handle ICMP_REDIRECT.
740 static void icmp_redirect(struct sk_buff *skb)
744 if (skb->len < sizeof(struct iphdr))
748 * Get the copied header of the packet that caused the redirect
750 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
753 iph = (struct iphdr *)skb->data;
755 switch (icmp_hdr(skb)->code & 7) {
757 case ICMP_REDIR_NETTOS:
759 * As per RFC recommendations now handle it as a host redirect.
761 case ICMP_REDIR_HOST:
762 case ICMP_REDIR_HOSTTOS:
763 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
764 icmp_hdr(skb)->un.gateway,
765 iph->saddr, skb->dev);
771 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
776 * Handle ICMP_ECHO ("ping") requests.
778 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
780 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
781 * included in the reply.
782 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
783 * echo requests, MUST have default=NOT.
784 * See also WRT handling of options once they are done and working.
787 static void icmp_echo(struct sk_buff *skb)
791 net = dev_net(skb_dst(skb)->dev);
792 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
793 struct icmp_bxm icmp_param;
795 icmp_param.data.icmph = *icmp_hdr(skb);
796 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
797 icmp_param.skb = skb;
798 icmp_param.offset = 0;
799 icmp_param.data_len = skb->len;
800 icmp_param.head_len = sizeof(struct icmphdr);
801 icmp_reply(&icmp_param, skb);
806 * Handle ICMP Timestamp requests.
807 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
808 * SHOULD be in the kernel for minimum random latency.
809 * MUST be accurate to a few minutes.
810 * MUST be updated at least at 15Hz.
812 static void icmp_timestamp(struct sk_buff *skb)
815 struct icmp_bxm icmp_param;
823 * Fill in the current time as ms since midnight UT:
826 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
827 tv.tv_nsec / NSEC_PER_MSEC);
828 icmp_param.data.times[2] = icmp_param.data.times[1];
829 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
831 icmp_param.data.icmph = *icmp_hdr(skb);
832 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
833 icmp_param.data.icmph.code = 0;
834 icmp_param.skb = skb;
835 icmp_param.offset = 0;
836 icmp_param.data_len = 0;
837 icmp_param.head_len = sizeof(struct icmphdr) + 12;
838 icmp_reply(&icmp_param, skb);
842 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
848 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
850 * RFC1122 (3.2.2.9). A host MUST only send replies to
851 * ADDRESS_MASK requests if it's been configured as an address mask
852 * agent. Receiving a request doesn't constitute implicit permission to
853 * act as one. Of course, implementing this correctly requires (SHOULD)
854 * a way to turn the functionality on and off. Another one for sysctl(),
857 * RFC1812 (4.3.3.9). A router MUST implement it.
858 * A router SHOULD have switch turning it on/off.
859 * This switch MUST be ON by default.
861 * Gratuitous replies, zero-source replies are not implemented,
862 * that complies with RFC. DO NOT implement them!!! All the idea
863 * of broadcast addrmask replies as specified in RFC950 is broken.
864 * The problem is that it is not uncommon to have several prefixes
865 * on one physical interface. Moreover, addrmask agent can even be
866 * not aware of existing another prefixes.
867 * If source is zero, addrmask agent cannot choose correct prefix.
868 * Gratuitous mask announcements suffer from the same problem.
869 * RFC1812 explains it, but still allows to use ADDRMASK,
870 * that is pretty silly. --ANK
872 * All these rules are so bizarre, that I removed kernel addrmask
873 * support at all. It is wrong, it is obsolete, nobody uses it in
876 * Furthermore you can do it with a usermode address agent program
880 static void icmp_address(struct sk_buff *skb)
884 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
889 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
890 * loudly if an inconsistency is found.
891 * called with rcu_read_lock()
894 static void icmp_address_reply(struct sk_buff *skb)
896 struct rtable *rt = skb_rtable(skb);
897 struct net_device *dev = skb->dev;
898 struct in_device *in_dev;
899 struct in_ifaddr *ifa;
901 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
904 in_dev = __in_dev_get_rcu(dev);
908 if (in_dev->ifa_list &&
909 IN_DEV_LOG_MARTIANS(in_dev) &&
910 IN_DEV_FORWARD(in_dev)) {
913 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
915 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
916 if (*mp == ifa->ifa_mask &&
917 inet_ifa_match(rt->rt_src, ifa))
920 if (!ifa && net_ratelimit()) {
921 printk(KERN_INFO "Wrong address mask %pI4 from %s/%pI4\n",
922 mp, dev->name, &rt->rt_src);
927 static void icmp_discard(struct sk_buff *skb)
932 * Deal with incoming ICMP packets.
934 int icmp_rcv(struct sk_buff *skb)
936 struct icmphdr *icmph;
937 struct rtable *rt = skb_rtable(skb);
938 struct net *net = dev_net(rt->dst.dev);
940 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
941 struct sec_path *sp = skb_sec_path(skb);
944 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
948 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
951 nh = skb_network_offset(skb);
952 skb_set_network_header(skb, sizeof(*icmph));
954 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
957 skb_set_network_header(skb, nh);
960 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
962 switch (skb->ip_summed) {
963 case CHECKSUM_COMPLETE:
964 if (!csum_fold(skb->csum))
969 if (__skb_checksum_complete(skb))
973 if (!pskb_pull(skb, sizeof(*icmph)))
976 icmph = icmp_hdr(skb);
978 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
980 * 18 is the highest 'known' ICMP type. Anything else is a mystery
982 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
985 if (icmph->type > NR_ICMP_TYPES)
990 * Parse the ICMP message
993 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
995 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
996 * silently ignored (we let user decide with a sysctl).
997 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
998 * discarded if to broadcast/multicast.
1000 if ((icmph->type == ICMP_ECHO ||
1001 icmph->type == ICMP_TIMESTAMP) &&
1002 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1005 if (icmph->type != ICMP_ECHO &&
1006 icmph->type != ICMP_TIMESTAMP &&
1007 icmph->type != ICMP_ADDRESS &&
1008 icmph->type != ICMP_ADDRESSREPLY) {
1013 icmp_pointers[icmph->type].handler(skb);
1019 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1024 * This table is the definition of how we handle ICMP.
1026 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1027 [ICMP_ECHOREPLY] = {
1028 .handler = icmp_discard,
1031 .handler = icmp_discard,
1035 .handler = icmp_discard,
1038 [ICMP_DEST_UNREACH] = {
1039 .handler = icmp_unreach,
1042 [ICMP_SOURCE_QUENCH] = {
1043 .handler = icmp_unreach,
1047 .handler = icmp_redirect,
1051 .handler = icmp_discard,
1055 .handler = icmp_discard,
1059 .handler = icmp_echo,
1062 .handler = icmp_discard,
1066 .handler = icmp_discard,
1069 [ICMP_TIME_EXCEEDED] = {
1070 .handler = icmp_unreach,
1073 [ICMP_PARAMETERPROB] = {
1074 .handler = icmp_unreach,
1077 [ICMP_TIMESTAMP] = {
1078 .handler = icmp_timestamp,
1080 [ICMP_TIMESTAMPREPLY] = {
1081 .handler = icmp_discard,
1083 [ICMP_INFO_REQUEST] = {
1084 .handler = icmp_discard,
1086 [ICMP_INFO_REPLY] = {
1087 .handler = icmp_discard,
1090 .handler = icmp_address,
1092 [ICMP_ADDRESSREPLY] = {
1093 .handler = icmp_address_reply,
1097 static void __net_exit icmp_sk_exit(struct net *net)
1101 for_each_possible_cpu(i)
1102 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1103 kfree(net->ipv4.icmp_sk);
1104 net->ipv4.icmp_sk = NULL;
1107 static int __net_init icmp_sk_init(struct net *net)
1112 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1113 if (net->ipv4.icmp_sk == NULL)
1116 for_each_possible_cpu(i) {
1119 err = inet_ctl_sock_create(&sk, PF_INET,
1120 SOCK_RAW, IPPROTO_ICMP, net);
1124 net->ipv4.icmp_sk[i] = sk;
1126 /* Enough space for 2 64K ICMP packets, including
1127 * sk_buff struct overhead.
1130 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1133 * Speedup sock_wfree()
1135 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1136 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1139 /* Control parameters for ECHO replies. */
1140 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1141 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1143 /* Control parameter - ignore bogus broadcast responses? */
1144 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1147 * Configurable global rate limit.
1149 * ratelimit defines tokens/packet consumed for dst->rate_token
1150 * bucket ratemask defines which icmp types are ratelimited by
1151 * setting it's bit position.
1154 * dest unreachable (3), source quench (4),
1155 * time exceeded (11), parameter problem (12)
1158 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1159 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1160 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1165 for_each_possible_cpu(i)
1166 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1167 kfree(net->ipv4.icmp_sk);
1171 static struct pernet_operations __net_initdata icmp_sk_ops = {
1172 .init = icmp_sk_init,
1173 .exit = icmp_sk_exit,
1176 int __init icmp_init(void)
1178 return register_pernet_subsys(&icmp_sk_ops);