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 #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>
98 #include <net/ip_fib.h>
101 * Build xmit assembly blocks
110 struct icmphdr icmph;
114 struct ip_options_data replyopts;
117 /* An array of errno for error messages from dest unreach. */
118 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
120 const struct icmp_err icmp_err_convert[] = {
122 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
126 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
130 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
134 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
138 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
142 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
146 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
150 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
154 .errno = ENONET, /* ICMP_HOST_ISOLATED */
158 .errno = ENETUNREACH, /* ICMP_NET_ANO */
162 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
166 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
170 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
174 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
178 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
182 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
186 EXPORT_SYMBOL(icmp_err_convert);
189 * ICMP control array. This specifies what to do with each ICMP.
192 struct icmp_control {
193 void (*handler)(struct sk_buff *skb);
194 short error; /* This ICMP is classed as an error message */
197 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
200 * The ICMP socket(s). This is the most convenient way to flow control
201 * our ICMP output as well as maintain a clean interface throughout
202 * all layers. All Socketless IP sends will soon be gone.
204 * On SMP we have one ICMP socket per-cpu.
206 static struct sock *icmp_sk(struct net *net)
208 return net->ipv4.icmp_sk[smp_processor_id()];
211 static inline struct sock *icmp_xmit_lock(struct net *net)
219 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
220 /* This can happen if the output path signals a
221 * dst_link_failure() for an outgoing ICMP packet.
229 static inline void icmp_xmit_unlock(struct sock *sk)
231 spin_unlock_bh(&sk->sk_lock.slock);
235 * Send an ICMP frame.
238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 struct flowi4 *fl4, int type, int code)
241 struct dst_entry *dst = &rt->dst;
244 if (type > NR_ICMP_TYPES)
247 /* Don't limit PMTU discovery. */
248 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
251 /* No rate limit on loopback */
252 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
255 /* Limit if icmp type is enabled in ratemask. */
256 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 struct inet_peer *peer = rt_get_peer_create(rt, fl4->daddr);
258 rc = inet_peer_xrlim_allow(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;
351 saddr = fib_compute_spec_dst(skb);
354 if (icmp_param->replyopts.opt.opt.optlen) {
355 ipc.opt = &icmp_param->replyopts.opt;
356 if (ipc.opt->opt.srr)
357 daddr = icmp_param->replyopts.opt.opt.faddr;
359 memset(&fl4, 0, sizeof(fl4));
362 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
363 fl4.flowi4_proto = IPPROTO_ICMP;
364 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
365 rt = ip_route_output_key(net, &fl4);
368 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
369 icmp_param->data.icmph.code))
370 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
373 icmp_xmit_unlock(sk);
376 static struct rtable *icmp_route_lookup(struct net *net,
378 struct sk_buff *skb_in,
379 const struct iphdr *iph,
380 __be32 saddr, u8 tos,
382 struct icmp_bxm *param)
384 struct rtable *rt, *rt2;
385 struct flowi4 fl4_dec;
388 memset(fl4, 0, sizeof(*fl4));
389 fl4->daddr = (param->replyopts.opt.opt.srr ?
390 param->replyopts.opt.opt.faddr : iph->saddr);
392 fl4->flowi4_tos = RT_TOS(tos);
393 fl4->flowi4_proto = IPPROTO_ICMP;
394 fl4->fl4_icmp_type = type;
395 fl4->fl4_icmp_code = code;
396 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
397 rt = __ip_route_output_key(net, fl4);
401 /* No need to clone since we're just using its address. */
404 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
405 flowi4_to_flowi(fl4), NULL, 0);
409 } else if (PTR_ERR(rt) == -EPERM) {
414 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
416 goto relookup_failed;
418 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
419 rt2 = __ip_route_output_key(net, &fl4_dec);
423 struct flowi4 fl4_2 = {};
424 unsigned long orefdst;
426 fl4_2.daddr = fl4_dec.saddr;
427 rt2 = ip_route_output_key(net, &fl4_2);
430 goto relookup_failed;
433 orefdst = skb_in->_skb_refdst; /* save old refdst */
434 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
435 RT_TOS(tos), rt2->dst.dev);
437 dst_release(&rt2->dst);
438 rt2 = skb_rtable(skb_in);
439 skb_in->_skb_refdst = orefdst; /* restore old refdst */
443 goto relookup_failed;
445 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
446 flowi4_to_flowi(&fl4_dec), NULL,
449 dst_release(&rt->dst);
450 memcpy(fl4, &fl4_dec, sizeof(*fl4));
452 } else if (PTR_ERR(rt2) == -EPERM) {
454 dst_release(&rt->dst);
458 goto relookup_failed;
469 * Send an ICMP message in response to a situation
471 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
472 * MAY send more (we do).
473 * MUST NOT change this header information.
474 * MUST NOT reply to a multicast/broadcast IP address.
475 * MUST NOT reply to a multicast/broadcast MAC address.
476 * MUST reply to only the first fragment.
479 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
483 struct icmp_bxm icmp_param;
484 struct rtable *rt = skb_rtable(skb_in);
485 struct ipcm_cookie ipc;
494 net = dev_net(rt->dst.dev);
497 * Find the original header. It is expected to be valid, of course.
498 * Check this, icmp_send is called from the most obscure devices
501 iph = ip_hdr(skb_in);
503 if ((u8 *)iph < skb_in->head ||
504 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
508 * No replies to physical multicast/broadcast
510 if (skb_in->pkt_type != PACKET_HOST)
514 * Now check at the protocol level
516 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
520 * Only reply to fragment 0. We byte re-order the constant
521 * mask for efficiency.
523 if (iph->frag_off & htons(IP_OFFSET))
527 * If we send an ICMP error to an ICMP error a mess would result..
529 if (icmp_pointers[type].error) {
531 * We are an error, check if we are replying to an
534 if (iph->protocol == IPPROTO_ICMP) {
535 u8 _inner_type, *itp;
537 itp = skb_header_pointer(skb_in,
538 skb_network_header(skb_in) +
540 offsetof(struct icmphdr,
549 * Assume any unknown ICMP type is an error. This
550 * isn't specified by the RFC, but think about it..
552 if (*itp > NR_ICMP_TYPES ||
553 icmp_pointers[*itp].error)
558 sk = icmp_xmit_lock(net);
563 * Construct source address and options.
567 if (!(rt->rt_flags & RTCF_LOCAL)) {
568 struct net_device *dev = NULL;
571 if (rt_is_input_route(rt) &&
572 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
573 dev = dev_get_by_index_rcu(net, rt->rt_iif);
576 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
582 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
583 IPTOS_PREC_INTERNETCONTROL) :
586 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
591 * Prepare data for ICMP header.
594 icmp_param.data.icmph.type = type;
595 icmp_param.data.icmph.code = code;
596 icmp_param.data.icmph.un.gateway = info;
597 icmp_param.data.icmph.checksum = 0;
598 icmp_param.skb = skb_in;
599 icmp_param.offset = skb_network_offset(skb_in);
600 inet_sk(sk)->tos = tos;
601 ipc.addr = iph->saddr;
602 ipc.opt = &icmp_param.replyopts.opt;
605 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
606 type, code, &icmp_param);
610 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
613 /* RFC says return as much as we can without exceeding 576 bytes. */
615 room = dst_mtu(&rt->dst);
618 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
619 room -= sizeof(struct icmphdr);
621 icmp_param.data_len = skb_in->len - icmp_param.offset;
622 if (icmp_param.data_len > room)
623 icmp_param.data_len = room;
624 icmp_param.head_len = sizeof(struct icmphdr);
626 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
630 icmp_xmit_unlock(sk);
633 EXPORT_SYMBOL(icmp_send);
637 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
640 static void icmp_unreach(struct sk_buff *skb)
642 const struct net_protocol *ipprot;
643 const struct iphdr *iph;
644 struct icmphdr *icmph;
649 net = dev_net(skb_dst(skb)->dev);
652 * Incomplete header ?
653 * Only checks for the IP header, there should be an
654 * additional check for longer headers in upper levels.
657 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
660 icmph = icmp_hdr(skb);
661 iph = (const struct iphdr *)skb->data;
663 if (iph->ihl < 5) /* Mangled header, drop. */
666 if (icmph->type == ICMP_DEST_UNREACH) {
667 switch (icmph->code & 15) {
668 case ICMP_NET_UNREACH:
669 case ICMP_HOST_UNREACH:
670 case ICMP_PROT_UNREACH:
671 case ICMP_PORT_UNREACH:
673 case ICMP_FRAG_NEEDED:
674 if (ipv4_config.no_pmtu_disc) {
675 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
678 info = ntohs(icmph->un.frag.mtu);
684 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
690 if (icmph->code > NR_ICMP_UNREACH)
692 } else if (icmph->type == ICMP_PARAMETERPROB)
693 info = ntohl(icmph->un.gateway) >> 24;
696 * Throw it at our lower layers
698 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
700 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
702 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
707 * Check the other end isn't violating RFC 1122. Some routers send
708 * bogus responses to broadcast frames. If you see this message
709 * first check your netmask matches at both ends, if it does then
710 * get the other vendor to fix their kit.
713 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
714 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
715 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
717 icmph->type, icmph->code,
718 &iph->daddr, skb->dev->name);
722 /* Checkin full IP header plus 8 bytes of protocol to
723 * avoid additional coding at protocol handlers.
725 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
728 iph = (const struct iphdr *)skb->data;
729 protocol = iph->protocol;
732 * Deliver ICMP message to raw sockets. Pretty useless feature?
734 raw_icmp_error(skb, protocol, info);
737 ipprot = rcu_dereference(inet_protos[protocol]);
738 if (ipprot && ipprot->err_handler)
739 ipprot->err_handler(skb, info);
745 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
751 * Handle ICMP_REDIRECT.
754 static void icmp_redirect(struct sk_buff *skb)
756 const struct iphdr *iph;
758 if (skb->len < sizeof(struct iphdr))
762 * Get the copied header of the packet that caused the redirect
764 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
767 iph = (const struct iphdr *)skb->data;
769 switch (icmp_hdr(skb)->code & 7) {
771 case ICMP_REDIR_NETTOS:
773 * As per RFC recommendations now handle it as a host redirect.
775 case ICMP_REDIR_HOST:
776 case ICMP_REDIR_HOSTTOS:
777 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
778 icmp_hdr(skb)->un.gateway,
779 iph->saddr, skb->dev);
783 /* Ping wants to see redirects.
784 * Let's pretend they are errors of sorts... */
785 if (iph->protocol == IPPROTO_ICMP &&
787 pskb_may_pull(skb, (iph->ihl<<2)+8)) {
788 ping_err(skb, icmp_hdr(skb)->un.gateway);
794 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
799 * Handle ICMP_ECHO ("ping") requests.
801 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
803 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
804 * included in the reply.
805 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
806 * echo requests, MUST have default=NOT.
807 * See also WRT handling of options once they are done and working.
810 static void icmp_echo(struct sk_buff *skb)
814 net = dev_net(skb_dst(skb)->dev);
815 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
816 struct icmp_bxm icmp_param;
818 icmp_param.data.icmph = *icmp_hdr(skb);
819 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
820 icmp_param.skb = skb;
821 icmp_param.offset = 0;
822 icmp_param.data_len = skb->len;
823 icmp_param.head_len = sizeof(struct icmphdr);
824 icmp_reply(&icmp_param, skb);
829 * Handle ICMP Timestamp requests.
830 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
831 * SHOULD be in the kernel for minimum random latency.
832 * MUST be accurate to a few minutes.
833 * MUST be updated at least at 15Hz.
835 static void icmp_timestamp(struct sk_buff *skb)
838 struct icmp_bxm icmp_param;
846 * Fill in the current time as ms since midnight UT:
849 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
850 tv.tv_nsec / NSEC_PER_MSEC);
851 icmp_param.data.times[2] = icmp_param.data.times[1];
852 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
854 icmp_param.data.icmph = *icmp_hdr(skb);
855 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
856 icmp_param.data.icmph.code = 0;
857 icmp_param.skb = skb;
858 icmp_param.offset = 0;
859 icmp_param.data_len = 0;
860 icmp_param.head_len = sizeof(struct icmphdr) + 12;
861 icmp_reply(&icmp_param, skb);
865 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
871 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
873 * RFC1122 (3.2.2.9). A host MUST only send replies to
874 * ADDRESS_MASK requests if it's been configured as an address mask
875 * agent. Receiving a request doesn't constitute implicit permission to
876 * act as one. Of course, implementing this correctly requires (SHOULD)
877 * a way to turn the functionality on and off. Another one for sysctl(),
880 * RFC1812 (4.3.3.9). A router MUST implement it.
881 * A router SHOULD have switch turning it on/off.
882 * This switch MUST be ON by default.
884 * Gratuitous replies, zero-source replies are not implemented,
885 * that complies with RFC. DO NOT implement them!!! All the idea
886 * of broadcast addrmask replies as specified in RFC950 is broken.
887 * The problem is that it is not uncommon to have several prefixes
888 * on one physical interface. Moreover, addrmask agent can even be
889 * not aware of existing another prefixes.
890 * If source is zero, addrmask agent cannot choose correct prefix.
891 * Gratuitous mask announcements suffer from the same problem.
892 * RFC1812 explains it, but still allows to use ADDRMASK,
893 * that is pretty silly. --ANK
895 * All these rules are so bizarre, that I removed kernel addrmask
896 * support at all. It is wrong, it is obsolete, nobody uses it in
899 * Furthermore you can do it with a usermode address agent program
903 static void icmp_address(struct sk_buff *skb)
906 net_dbg_ratelimited("a guy asks for address mask. Who is it?\n");
911 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
912 * loudly if an inconsistency is found.
913 * called with rcu_read_lock()
916 static void icmp_address_reply(struct sk_buff *skb)
918 struct rtable *rt = skb_rtable(skb);
919 struct net_device *dev = skb->dev;
920 struct in_device *in_dev;
921 struct in_ifaddr *ifa;
923 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
926 in_dev = __in_dev_get_rcu(dev);
930 if (in_dev->ifa_list &&
931 IN_DEV_LOG_MARTIANS(in_dev) &&
932 IN_DEV_FORWARD(in_dev)) {
935 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
937 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
938 if (*mp == ifa->ifa_mask &&
939 inet_ifa_match(ip_hdr(skb)->saddr, ifa))
943 net_info_ratelimited("Wrong address mask %pI4 from %s/%pI4\n",
945 dev->name, &ip_hdr(skb)->saddr);
949 static void icmp_discard(struct sk_buff *skb)
954 * Deal with incoming ICMP packets.
956 int icmp_rcv(struct sk_buff *skb)
958 struct icmphdr *icmph;
959 struct rtable *rt = skb_rtable(skb);
960 struct net *net = dev_net(rt->dst.dev);
962 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
963 struct sec_path *sp = skb_sec_path(skb);
966 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
970 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
973 nh = skb_network_offset(skb);
974 skb_set_network_header(skb, sizeof(*icmph));
976 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
979 skb_set_network_header(skb, nh);
982 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
984 switch (skb->ip_summed) {
985 case CHECKSUM_COMPLETE:
986 if (!csum_fold(skb->csum))
991 if (__skb_checksum_complete(skb))
995 if (!pskb_pull(skb, sizeof(*icmph)))
998 icmph = icmp_hdr(skb);
1000 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1002 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1004 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1007 if (icmph->type > NR_ICMP_TYPES)
1012 * Parse the ICMP message
1015 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1017 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1018 * silently ignored (we let user decide with a sysctl).
1019 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1020 * discarded if to broadcast/multicast.
1022 if ((icmph->type == ICMP_ECHO ||
1023 icmph->type == ICMP_TIMESTAMP) &&
1024 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1027 if (icmph->type != ICMP_ECHO &&
1028 icmph->type != ICMP_TIMESTAMP &&
1029 icmph->type != ICMP_ADDRESS &&
1030 icmph->type != ICMP_ADDRESSREPLY) {
1035 icmp_pointers[icmph->type].handler(skb);
1041 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1046 * This table is the definition of how we handle ICMP.
1048 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1049 [ICMP_ECHOREPLY] = {
1050 .handler = ping_rcv,
1053 .handler = icmp_discard,
1057 .handler = icmp_discard,
1060 [ICMP_DEST_UNREACH] = {
1061 .handler = icmp_unreach,
1064 [ICMP_SOURCE_QUENCH] = {
1065 .handler = icmp_unreach,
1069 .handler = icmp_redirect,
1073 .handler = icmp_discard,
1077 .handler = icmp_discard,
1081 .handler = icmp_echo,
1084 .handler = icmp_discard,
1088 .handler = icmp_discard,
1091 [ICMP_TIME_EXCEEDED] = {
1092 .handler = icmp_unreach,
1095 [ICMP_PARAMETERPROB] = {
1096 .handler = icmp_unreach,
1099 [ICMP_TIMESTAMP] = {
1100 .handler = icmp_timestamp,
1102 [ICMP_TIMESTAMPREPLY] = {
1103 .handler = icmp_discard,
1105 [ICMP_INFO_REQUEST] = {
1106 .handler = icmp_discard,
1108 [ICMP_INFO_REPLY] = {
1109 .handler = icmp_discard,
1112 .handler = icmp_address,
1114 [ICMP_ADDRESSREPLY] = {
1115 .handler = icmp_address_reply,
1119 static void __net_exit icmp_sk_exit(struct net *net)
1123 for_each_possible_cpu(i)
1124 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1125 kfree(net->ipv4.icmp_sk);
1126 net->ipv4.icmp_sk = NULL;
1129 static int __net_init icmp_sk_init(struct net *net)
1134 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1135 if (net->ipv4.icmp_sk == NULL)
1138 for_each_possible_cpu(i) {
1141 err = inet_ctl_sock_create(&sk, PF_INET,
1142 SOCK_RAW, IPPROTO_ICMP, net);
1146 net->ipv4.icmp_sk[i] = sk;
1148 /* Enough space for 2 64K ICMP packets, including
1149 * sk_buff/skb_shared_info struct overhead.
1151 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1154 * Speedup sock_wfree()
1156 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1157 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1160 /* Control parameters for ECHO replies. */
1161 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1162 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1164 /* Control parameter - ignore bogus broadcast responses? */
1165 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1168 * Configurable global rate limit.
1170 * ratelimit defines tokens/packet consumed for dst->rate_token
1171 * bucket ratemask defines which icmp types are ratelimited by
1172 * setting it's bit position.
1175 * dest unreachable (3), source quench (4),
1176 * time exceeded (11), parameter problem (12)
1179 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1180 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1181 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1186 for_each_possible_cpu(i)
1187 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1188 kfree(net->ipv4.icmp_sk);
1192 static struct pernet_operations __net_initdata icmp_sk_ops = {
1193 .init = icmp_sk_init,
1194 .exit = icmp_sk_exit,
1197 int __init icmp_init(void)
1199 return register_pernet_subsys(&icmp_sk_ops);