2 * NET3: Implementation of the ICMP protocol layer.
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
79 #include <linux/slab.h>
82 #include <net/route.h>
83 #include <net/protocol.h>
89 #include <linux/skbuff.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
93 #include <linux/init.h>
94 #include <asm/uaccess.h>
95 #include <net/checksum.h>
97 #include <net/inet_common.h>
100 * Build xmit assembly blocks
109 struct icmphdr icmph;
113 struct ip_options_data replyopts;
116 /* An array of errno for error messages from dest unreach. */
117 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119 const struct icmp_err icmp_err_convert[] = {
121 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
125 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
129 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
133 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
137 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
141 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
145 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
149 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
153 .errno = ENONET, /* ICMP_HOST_ISOLATED */
157 .errno = ENETUNREACH, /* ICMP_NET_ANO */
161 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
165 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
169 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
173 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
177 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
181 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
185 EXPORT_SYMBOL(icmp_err_convert);
188 * ICMP control array. This specifies what to do with each ICMP.
191 struct icmp_control {
192 void (*handler)(struct sk_buff *skb);
193 short error; /* This ICMP is classed as an error message */
196 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
199 * The ICMP socket(s). This is the most convenient way to flow control
200 * our ICMP output as well as maintain a clean interface throughout
201 * all layers. All Socketless IP sends will soon be gone.
203 * On SMP we have one ICMP socket per-cpu.
205 static struct sock *icmp_sk(struct net *net)
207 return net->ipv4.icmp_sk[smp_processor_id()];
210 static inline struct sock *icmp_xmit_lock(struct net *net)
218 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 /* This can happen if the output path signals a
220 * dst_link_failure() for an outgoing ICMP packet.
228 static inline void icmp_xmit_unlock(struct sock *sk)
230 spin_unlock_bh(&sk->sk_lock.slock);
234 * Send an ICMP frame.
237 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
238 struct flowi4 *fl4, int type, int code)
240 struct dst_entry *dst = &rt->dst;
243 if (type > NR_ICMP_TYPES)
246 /* Don't limit PMTU discovery. */
247 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
250 /* No rate limit on loopback */
251 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
254 /* Limit if icmp type is enabled in ratemask. */
255 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 rt_bind_peer(rt, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(rt->peer,
259 net->ipv4.sysctl_icmp_ratelimit);
266 * Maintain the counters used in the SNMP statistics for outgoing ICMP
268 void icmp_out_count(struct net *net, unsigned char type)
270 ICMPMSGOUT_INC_STATS(net, type);
271 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
275 * Checksum each fragment, and on the first include the headers and final
278 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
281 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
284 csum = skb_copy_and_csum_bits(icmp_param->skb,
285 icmp_param->offset + offset,
288 skb->csum = csum_block_add(skb->csum, csum, odd);
289 if (icmp_pointers[icmp_param->data.icmph.type].error)
290 nf_ct_attach(skb, icmp_param->skb);
294 static void icmp_push_reply(struct icmp_bxm *icmp_param,
296 struct ipcm_cookie *ipc, struct rtable **rt)
301 sk = icmp_sk(dev_net((*rt)->dst.dev));
302 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
303 icmp_param->data_len+icmp_param->head_len,
304 icmp_param->head_len,
305 ipc, rt, MSG_DONTWAIT) < 0) {
306 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
307 ip_flush_pending_frames(sk);
308 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
309 struct icmphdr *icmph = icmp_hdr(skb);
311 struct sk_buff *skb1;
313 skb_queue_walk(&sk->sk_write_queue, skb1) {
314 csum = csum_add(csum, skb1->csum);
316 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
318 icmp_param->head_len, csum);
319 icmph->checksum = csum_fold(csum);
320 skb->ip_summed = CHECKSUM_NONE;
321 ip_push_pending_frames(sk, fl4);
326 * Driving logic for building and sending ICMP messages.
329 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
331 struct ipcm_cookie ipc;
332 struct rtable *rt = skb_rtable(skb);
333 struct net *net = dev_net(rt->dst.dev);
336 struct inet_sock *inet;
339 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
342 sk = icmp_xmit_lock(net);
347 icmp_param->data.icmph.checksum = 0;
349 inet->tos = ip_hdr(skb)->tos;
350 daddr = ipc.addr = ip_hdr(skb)->saddr;
353 if (icmp_param->replyopts.opt.opt.optlen) {
354 ipc.opt = &icmp_param->replyopts.opt;
355 if (ipc.opt->opt.srr)
356 daddr = icmp_param->replyopts.opt.opt.faddr;
358 memset(&fl4, 0, sizeof(fl4));
360 fl4.saddr = rt->rt_spec_dst;
361 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
362 fl4.flowi4_proto = IPPROTO_ICMP;
363 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
364 rt = ip_route_output_key(net, &fl4);
367 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
368 icmp_param->data.icmph.code))
369 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
372 icmp_xmit_unlock(sk);
375 static struct rtable *icmp_route_lookup(struct net *net,
377 struct sk_buff *skb_in,
378 const struct iphdr *iph,
379 __be32 saddr, u8 tos,
381 struct icmp_bxm *param)
383 struct rtable *rt, *rt2;
384 struct flowi4 fl4_dec;
387 memset(fl4, 0, sizeof(*fl4));
388 fl4->daddr = (param->replyopts.opt.opt.srr ?
389 param->replyopts.opt.opt.faddr : iph->saddr);
391 fl4->flowi4_tos = RT_TOS(tos);
392 fl4->flowi4_proto = IPPROTO_ICMP;
393 fl4->fl4_icmp_type = type;
394 fl4->fl4_icmp_code = code;
395 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
396 rt = __ip_route_output_key(net, fl4);
400 /* No need to clone since we're just using its address. */
403 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
404 flowi4_to_flowi(fl4), NULL, 0);
408 } else if (PTR_ERR(rt) == -EPERM) {
413 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
415 goto relookup_failed;
417 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
418 rt2 = __ip_route_output_key(net, &fl4_dec);
422 struct flowi4 fl4_2 = {};
423 unsigned long orefdst;
425 fl4_2.daddr = fl4_dec.saddr;
426 rt2 = ip_route_output_key(net, &fl4_2);
429 goto relookup_failed;
432 orefdst = skb_in->_skb_refdst; /* save old refdst */
433 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
434 RT_TOS(tos), rt2->dst.dev);
436 dst_release(&rt2->dst);
437 rt2 = skb_rtable(skb_in);
438 skb_in->_skb_refdst = orefdst; /* restore old refdst */
442 goto relookup_failed;
444 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
445 flowi4_to_flowi(&fl4_dec), NULL,
448 dst_release(&rt->dst);
449 memcpy(fl4, &fl4_dec, sizeof(*fl4));
451 } else if (PTR_ERR(rt2) == -EPERM) {
453 dst_release(&rt->dst);
457 goto relookup_failed;
468 * Send an ICMP message in response to a situation
470 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
471 * MAY send more (we do).
472 * MUST NOT change this header information.
473 * MUST NOT reply to a multicast/broadcast IP address.
474 * MUST NOT reply to a multicast/broadcast MAC address.
475 * MUST reply to only the first fragment.
478 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
482 struct icmp_bxm icmp_param;
483 struct rtable *rt = skb_rtable(skb_in);
484 struct ipcm_cookie ipc;
493 net = dev_net(rt->dst.dev);
496 * Find the original header. It is expected to be valid, of course.
497 * Check this, icmp_send is called from the most obscure devices
500 iph = ip_hdr(skb_in);
502 if ((u8 *)iph < skb_in->head ||
503 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
507 * No replies to physical multicast/broadcast
509 if (skb_in->pkt_type != PACKET_HOST)
513 * Now check at the protocol level
515 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
519 * Only reply to fragment 0. We byte re-order the constant
520 * mask for efficiency.
522 if (iph->frag_off & htons(IP_OFFSET))
526 * If we send an ICMP error to an ICMP error a mess would result..
528 if (icmp_pointers[type].error) {
530 * We are an error, check if we are replying to an
533 if (iph->protocol == IPPROTO_ICMP) {
534 u8 _inner_type, *itp;
536 itp = skb_header_pointer(skb_in,
537 skb_network_header(skb_in) +
539 offsetof(struct icmphdr,
548 * Assume any unknown ICMP type is an error. This
549 * isn't specified by the RFC, but think about it..
551 if (*itp > NR_ICMP_TYPES ||
552 icmp_pointers[*itp].error)
557 sk = icmp_xmit_lock(net);
562 * Construct source address and options.
566 if (!(rt->rt_flags & RTCF_LOCAL)) {
567 struct net_device *dev = NULL;
570 if (rt_is_input_route(rt) &&
571 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
572 dev = dev_get_by_index_rcu(net, rt->rt_iif);
575 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
581 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
582 IPTOS_PREC_INTERNETCONTROL) :
585 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
590 * Prepare data for ICMP header.
593 icmp_param.data.icmph.type = type;
594 icmp_param.data.icmph.code = code;
595 icmp_param.data.icmph.un.gateway = info;
596 icmp_param.data.icmph.checksum = 0;
597 icmp_param.skb = skb_in;
598 icmp_param.offset = skb_network_offset(skb_in);
599 inet_sk(sk)->tos = tos;
600 ipc.addr = iph->saddr;
601 ipc.opt = &icmp_param.replyopts.opt;
604 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
605 type, code, &icmp_param);
609 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
612 /* RFC says return as much as we can without exceeding 576 bytes. */
614 room = dst_mtu(&rt->dst);
617 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
618 room -= sizeof(struct icmphdr);
620 icmp_param.data_len = skb_in->len - icmp_param.offset;
621 if (icmp_param.data_len > room)
622 icmp_param.data_len = room;
623 icmp_param.head_len = sizeof(struct icmphdr);
625 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
629 icmp_xmit_unlock(sk);
632 EXPORT_SYMBOL(icmp_send);
636 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
639 static void icmp_unreach(struct sk_buff *skb)
641 const struct iphdr *iph;
642 struct icmphdr *icmph;
644 const struct net_protocol *ipprot;
648 net = dev_net(skb_dst(skb)->dev);
651 * Incomplete header ?
652 * Only checks for the IP header, there should be an
653 * additional check for longer headers in upper levels.
656 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
659 icmph = icmp_hdr(skb);
660 iph = (const struct iphdr *)skb->data;
662 if (iph->ihl < 5) /* Mangled header, drop. */
665 if (icmph->type == ICMP_DEST_UNREACH) {
666 switch (icmph->code & 15) {
667 case ICMP_NET_UNREACH:
668 case ICMP_HOST_UNREACH:
669 case ICMP_PROT_UNREACH:
670 case ICMP_PORT_UNREACH:
672 case ICMP_FRAG_NEEDED:
673 if (ipv4_config.no_pmtu_disc) {
674 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
677 info = ip_rt_frag_needed(net, iph,
678 ntohs(icmph->un.frag.mtu),
685 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
691 if (icmph->code > NR_ICMP_UNREACH)
693 } else if (icmph->type == ICMP_PARAMETERPROB)
694 info = ntohl(icmph->un.gateway) >> 24;
697 * Throw it at our lower layers
699 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
701 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
703 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
708 * Check the other end isn't violating RFC 1122. Some routers send
709 * bogus responses to broadcast frames. If you see this message
710 * first check your netmask matches at both ends, if it does then
711 * get the other vendor to fix their kit.
714 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
715 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
716 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
718 icmph->type, icmph->code,
719 &iph->daddr, skb->dev->name);
723 /* Checkin full IP header plus 8 bytes of protocol to
724 * avoid additional coding at protocol handlers.
726 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
729 iph = (const struct iphdr *)skb->data;
730 protocol = iph->protocol;
733 * Deliver ICMP message to raw sockets. Pretty useless feature?
735 raw_icmp_error(skb, protocol, info);
737 hash = protocol & (MAX_INET_PROTOS - 1);
739 ipprot = rcu_dereference(inet_protos[hash]);
740 if (ipprot && ipprot->err_handler)
741 ipprot->err_handler(skb, info);
747 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
753 * Handle ICMP_REDIRECT.
756 static void icmp_redirect(struct sk_buff *skb)
758 const struct iphdr *iph;
760 if (skb->len < sizeof(struct iphdr))
764 * Get the copied header of the packet that caused the redirect
766 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
769 iph = (const struct iphdr *)skb->data;
771 switch (icmp_hdr(skb)->code & 7) {
773 case ICMP_REDIR_NETTOS:
775 * As per RFC recommendations now handle it as a host redirect.
777 case ICMP_REDIR_HOST:
778 case ICMP_REDIR_HOSTTOS:
779 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
780 icmp_hdr(skb)->un.gateway,
781 iph->saddr, skb->dev);
785 /* Ping wants to see redirects.
786 * Let's pretend they are errors of sorts... */
787 if (iph->protocol == IPPROTO_ICMP &&
789 pskb_may_pull(skb, (iph->ihl<<2)+8)) {
790 ping_err(skb, icmp_hdr(skb)->un.gateway);
796 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
801 * Handle ICMP_ECHO ("ping") requests.
803 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
805 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
806 * included in the reply.
807 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
808 * echo requests, MUST have default=NOT.
809 * See also WRT handling of options once they are done and working.
812 static void icmp_echo(struct sk_buff *skb)
816 net = dev_net(skb_dst(skb)->dev);
817 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
818 struct icmp_bxm icmp_param;
820 icmp_param.data.icmph = *icmp_hdr(skb);
821 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
822 icmp_param.skb = skb;
823 icmp_param.offset = 0;
824 icmp_param.data_len = skb->len;
825 icmp_param.head_len = sizeof(struct icmphdr);
826 icmp_reply(&icmp_param, skb);
831 * Handle ICMP Timestamp requests.
832 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
833 * SHOULD be in the kernel for minimum random latency.
834 * MUST be accurate to a few minutes.
835 * MUST be updated at least at 15Hz.
837 static void icmp_timestamp(struct sk_buff *skb)
840 struct icmp_bxm icmp_param;
848 * Fill in the current time as ms since midnight UT:
851 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
852 tv.tv_nsec / NSEC_PER_MSEC);
853 icmp_param.data.times[2] = icmp_param.data.times[1];
854 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
856 icmp_param.data.icmph = *icmp_hdr(skb);
857 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
858 icmp_param.data.icmph.code = 0;
859 icmp_param.skb = skb;
860 icmp_param.offset = 0;
861 icmp_param.data_len = 0;
862 icmp_param.head_len = sizeof(struct icmphdr) + 12;
863 icmp_reply(&icmp_param, skb);
867 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
873 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
875 * RFC1122 (3.2.2.9). A host MUST only send replies to
876 * ADDRESS_MASK requests if it's been configured as an address mask
877 * agent. Receiving a request doesn't constitute implicit permission to
878 * act as one. Of course, implementing this correctly requires (SHOULD)
879 * a way to turn the functionality on and off. Another one for sysctl(),
882 * RFC1812 (4.3.3.9). A router MUST implement it.
883 * A router SHOULD have switch turning it on/off.
884 * This switch MUST be ON by default.
886 * Gratuitous replies, zero-source replies are not implemented,
887 * that complies with RFC. DO NOT implement them!!! All the idea
888 * of broadcast addrmask replies as specified in RFC950 is broken.
889 * The problem is that it is not uncommon to have several prefixes
890 * on one physical interface. Moreover, addrmask agent can even be
891 * not aware of existing another prefixes.
892 * If source is zero, addrmask agent cannot choose correct prefix.
893 * Gratuitous mask announcements suffer from the same problem.
894 * RFC1812 explains it, but still allows to use ADDRMASK,
895 * that is pretty silly. --ANK
897 * All these rules are so bizarre, that I removed kernel addrmask
898 * support at all. It is wrong, it is obsolete, nobody uses it in
901 * Furthermore you can do it with a usermode address agent program
905 static void icmp_address(struct sk_buff *skb)
908 net_dbg_ratelimited("a guy asks for address mask. Who is it?\n");
913 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
914 * loudly if an inconsistency is found.
915 * called with rcu_read_lock()
918 static void icmp_address_reply(struct sk_buff *skb)
920 struct rtable *rt = skb_rtable(skb);
921 struct net_device *dev = skb->dev;
922 struct in_device *in_dev;
923 struct in_ifaddr *ifa;
925 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
928 in_dev = __in_dev_get_rcu(dev);
932 if (in_dev->ifa_list &&
933 IN_DEV_LOG_MARTIANS(in_dev) &&
934 IN_DEV_FORWARD(in_dev)) {
937 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
939 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
940 if (*mp == ifa->ifa_mask &&
941 inet_ifa_match(ip_hdr(skb)->saddr, ifa))
945 net_info_ratelimited("Wrong address mask %pI4 from %s/%pI4\n",
947 dev->name, &ip_hdr(skb)->saddr);
951 static void icmp_discard(struct sk_buff *skb)
956 * Deal with incoming ICMP packets.
958 int icmp_rcv(struct sk_buff *skb)
960 struct icmphdr *icmph;
961 struct rtable *rt = skb_rtable(skb);
962 struct net *net = dev_net(rt->dst.dev);
964 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
965 struct sec_path *sp = skb_sec_path(skb);
968 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
972 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
975 nh = skb_network_offset(skb);
976 skb_set_network_header(skb, sizeof(*icmph));
978 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
981 skb_set_network_header(skb, nh);
984 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
986 switch (skb->ip_summed) {
987 case CHECKSUM_COMPLETE:
988 if (!csum_fold(skb->csum))
993 if (__skb_checksum_complete(skb))
997 if (!pskb_pull(skb, sizeof(*icmph)))
1000 icmph = icmp_hdr(skb);
1002 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1004 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1006 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1009 if (icmph->type > NR_ICMP_TYPES)
1014 * Parse the ICMP message
1017 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1019 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1020 * silently ignored (we let user decide with a sysctl).
1021 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1022 * discarded if to broadcast/multicast.
1024 if ((icmph->type == ICMP_ECHO ||
1025 icmph->type == ICMP_TIMESTAMP) &&
1026 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1029 if (icmph->type != ICMP_ECHO &&
1030 icmph->type != ICMP_TIMESTAMP &&
1031 icmph->type != ICMP_ADDRESS &&
1032 icmph->type != ICMP_ADDRESSREPLY) {
1037 icmp_pointers[icmph->type].handler(skb);
1043 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1048 * This table is the definition of how we handle ICMP.
1050 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1051 [ICMP_ECHOREPLY] = {
1052 .handler = ping_rcv,
1055 .handler = icmp_discard,
1059 .handler = icmp_discard,
1062 [ICMP_DEST_UNREACH] = {
1063 .handler = icmp_unreach,
1066 [ICMP_SOURCE_QUENCH] = {
1067 .handler = icmp_unreach,
1071 .handler = icmp_redirect,
1075 .handler = icmp_discard,
1079 .handler = icmp_discard,
1083 .handler = icmp_echo,
1086 .handler = icmp_discard,
1090 .handler = icmp_discard,
1093 [ICMP_TIME_EXCEEDED] = {
1094 .handler = icmp_unreach,
1097 [ICMP_PARAMETERPROB] = {
1098 .handler = icmp_unreach,
1101 [ICMP_TIMESTAMP] = {
1102 .handler = icmp_timestamp,
1104 [ICMP_TIMESTAMPREPLY] = {
1105 .handler = icmp_discard,
1107 [ICMP_INFO_REQUEST] = {
1108 .handler = icmp_discard,
1110 [ICMP_INFO_REPLY] = {
1111 .handler = icmp_discard,
1114 .handler = icmp_address,
1116 [ICMP_ADDRESSREPLY] = {
1117 .handler = icmp_address_reply,
1121 static void __net_exit icmp_sk_exit(struct net *net)
1125 for_each_possible_cpu(i)
1126 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1127 kfree(net->ipv4.icmp_sk);
1128 net->ipv4.icmp_sk = NULL;
1131 static int __net_init icmp_sk_init(struct net *net)
1136 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1137 if (net->ipv4.icmp_sk == NULL)
1140 for_each_possible_cpu(i) {
1143 err = inet_ctl_sock_create(&sk, PF_INET,
1144 SOCK_RAW, IPPROTO_ICMP, net);
1148 net->ipv4.icmp_sk[i] = sk;
1150 /* Enough space for 2 64K ICMP packets, including
1151 * sk_buff/skb_shared_info struct overhead.
1153 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1156 * Speedup sock_wfree()
1158 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1159 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1162 /* Control parameters for ECHO replies. */
1163 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1164 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1166 /* Control parameter - ignore bogus broadcast responses? */
1167 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1170 * Configurable global rate limit.
1172 * ratelimit defines tokens/packet consumed for dst->rate_token
1173 * bucket ratemask defines which icmp types are ratelimited by
1174 * setting it's bit position.
1177 * dest unreachable (3), source quench (4),
1178 * time exceeded (11), parameter problem (12)
1181 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1182 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1183 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1188 for_each_possible_cpu(i)
1189 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1190 kfree(net->ipv4.icmp_sk);
1194 static struct pernet_operations __net_initdata icmp_sk_ops = {
1195 .init = icmp_sk_init,
1196 .exit = icmp_sk_exit,
1199 int __init icmp_init(void)
1201 return register_pernet_subsys(&icmp_sk_ops);