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 <linux/uaccess.h>
95 #include <net/checksum.h>
97 #include <net/inet_common.h>
98 #include <net/ip_fib.h>
99 #include <net/l3mdev.h>
102 * Build xmit assembly blocks
111 struct icmphdr icmph;
115 struct ip_options_data replyopts;
118 /* An array of errno for error messages from dest unreach. */
119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
121 const struct icmp_err icmp_err_convert[] = {
123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
155 .errno = ENONET, /* ICMP_HOST_ISOLATED */
159 .errno = ENETUNREACH, /* ICMP_NET_ANO */
163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
187 EXPORT_SYMBOL(icmp_err_convert);
190 * ICMP control array. This specifies what to do with each ICMP.
193 struct icmp_control {
194 bool (*handler)(struct sk_buff *skb);
195 short error; /* This ICMP is classed as an error message */
198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
201 * The ICMP socket(s). This is the most convenient way to flow control
202 * our ICMP output as well as maintain a clean interface throughout
203 * all layers. All Socketless IP sends will soon be gone.
205 * On SMP we have one ICMP socket per-cpu.
207 static struct sock *icmp_sk(struct net *net)
209 return *this_cpu_ptr(net->ipv4.icmp_sk);
212 /* Called with BH disabled */
213 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.
228 static inline void icmp_xmit_unlock(struct sock *sk)
230 spin_unlock(&sk->sk_lock.slock);
233 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
234 int sysctl_icmp_msgs_burst __read_mostly = 50;
241 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
245 * icmp_global_allow - Are we allowed to send one more ICMP message ?
247 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
248 * Returns false if we reached the limit and can not send another packet.
249 * Note: called with BH disabled
251 bool icmp_global_allow(void)
253 u32 credit, delta, incr = 0, now = (u32)jiffies;
256 /* Check if token bucket is empty and cannot be refilled
257 * without taking the spinlock.
259 if (!icmp_global.credit) {
260 delta = min_t(u32, now - icmp_global.stamp, HZ);
265 spin_lock(&icmp_global.lock);
266 delta = min_t(u32, now - icmp_global.stamp, HZ);
267 if (delta >= HZ / 50) {
268 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
270 icmp_global.stamp = now;
272 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
277 icmp_global.credit = credit;
278 spin_unlock(&icmp_global.lock);
281 EXPORT_SYMBOL(icmp_global_allow);
283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
285 if (type > NR_ICMP_TYPES)
288 /* Don't limit PMTU discovery. */
289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
292 /* Limit if icmp type is enabled in ratemask. */
293 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
299 static bool icmpv4_global_allow(struct net *net, int type, int code)
301 if (icmpv4_mask_allow(net, type, code))
304 if (icmp_global_allow())
311 * Send an ICMP frame.
314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 struct flowi4 *fl4, int type, int code)
317 struct dst_entry *dst = &rt->dst;
318 struct inet_peer *peer;
322 if (icmpv4_mask_allow(net, type, code))
325 /* No rate limit on loopback */
326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
329 vif = l3mdev_master_ifindex(dst->dev);
330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
339 * Maintain the counters used in the SNMP statistics for outgoing ICMP
341 void icmp_out_count(struct net *net, unsigned char type)
343 ICMPMSGOUT_INC_STATS(net, type);
344 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
348 * Checksum each fragment, and on the first include the headers and final
351 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
354 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
357 csum = skb_copy_and_csum_bits(icmp_param->skb,
358 icmp_param->offset + offset,
361 skb->csum = csum_block_add(skb->csum, csum, odd);
362 if (icmp_pointers[icmp_param->data.icmph.type].error)
363 nf_ct_attach(skb, icmp_param->skb);
367 static void icmp_push_reply(struct icmp_bxm *icmp_param,
369 struct ipcm_cookie *ipc, struct rtable **rt)
374 sk = icmp_sk(dev_net((*rt)->dst.dev));
375 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
376 icmp_param->data_len+icmp_param->head_len,
377 icmp_param->head_len,
378 ipc, rt, MSG_DONTWAIT) < 0) {
379 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
380 ip_flush_pending_frames(sk);
381 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
382 struct icmphdr *icmph = icmp_hdr(skb);
384 struct sk_buff *skb1;
386 skb_queue_walk(&sk->sk_write_queue, skb1) {
387 csum = csum_add(csum, skb1->csum);
389 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
391 icmp_param->head_len, csum);
392 icmph->checksum = csum_fold(csum);
393 skb->ip_summed = CHECKSUM_NONE;
394 ip_push_pending_frames(sk, fl4);
399 * Driving logic for building and sending ICMP messages.
402 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
404 struct ipcm_cookie ipc;
405 struct rtable *rt = skb_rtable(skb);
406 struct net *net = dev_net(rt->dst.dev);
409 struct inet_sock *inet;
411 u32 mark = IP4_REPLY_MARK(net, skb->mark);
412 int type = icmp_param->data.icmph.type;
413 int code = icmp_param->data.icmph.code;
415 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
418 /* Needed by both icmp_global_allow and icmp_xmit_lock */
421 /* global icmp_msgs_per_sec */
422 if (!icmpv4_global_allow(net, type, code))
425 sk = icmp_xmit_lock(net);
430 icmp_param->data.icmph.checksum = 0;
433 inet->tos = ip_hdr(skb)->tos;
435 daddr = ipc.addr = ip_hdr(skb)->saddr;
436 saddr = fib_compute_spec_dst(skb);
438 if (icmp_param->replyopts.opt.opt.optlen) {
439 ipc.opt = &icmp_param->replyopts.opt;
440 if (ipc.opt->opt.srr)
441 daddr = icmp_param->replyopts.opt.opt.faddr;
443 memset(&fl4, 0, sizeof(fl4));
446 fl4.flowi4_mark = mark;
447 fl4.flowi4_uid = sock_net_uid(net, NULL);
448 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
449 fl4.flowi4_proto = IPPROTO_ICMP;
450 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
451 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
452 rt = ip_route_output_key(net, &fl4);
455 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
456 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
459 icmp_xmit_unlock(sk);
464 static struct rtable *icmp_route_lookup(struct net *net,
466 struct sk_buff *skb_in,
467 const struct iphdr *iph,
468 __be32 saddr, u8 tos, u32 mark,
470 struct icmp_bxm *param)
472 struct rtable *rt, *rt2;
473 struct flowi4 fl4_dec;
476 memset(fl4, 0, sizeof(*fl4));
477 fl4->daddr = (param->replyopts.opt.opt.srr ?
478 param->replyopts.opt.opt.faddr : iph->saddr);
480 fl4->flowi4_mark = mark;
481 fl4->flowi4_uid = sock_net_uid(net, NULL);
482 fl4->flowi4_tos = RT_TOS(tos);
483 fl4->flowi4_proto = IPPROTO_ICMP;
484 fl4->fl4_icmp_type = type;
485 fl4->fl4_icmp_code = code;
486 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
488 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
489 rt = ip_route_output_key_hash(net, fl4, skb_in);
493 /* No need to clone since we're just using its address. */
496 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
497 flowi4_to_flowi(fl4), NULL, 0);
501 } else if (PTR_ERR(rt) == -EPERM) {
506 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
508 goto relookup_failed;
510 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
511 fl4_dec.saddr) == RTN_LOCAL) {
512 rt2 = __ip_route_output_key(net, &fl4_dec);
516 struct flowi4 fl4_2 = {};
517 unsigned long orefdst;
519 fl4_2.daddr = fl4_dec.saddr;
520 rt2 = ip_route_output_key(net, &fl4_2);
523 goto relookup_failed;
526 orefdst = skb_in->_skb_refdst; /* save old refdst */
527 skb_dst_set(skb_in, NULL);
528 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
529 RT_TOS(tos), rt2->dst.dev);
531 dst_release(&rt2->dst);
532 rt2 = skb_rtable(skb_in);
533 skb_in->_skb_refdst = orefdst; /* restore old refdst */
537 goto relookup_failed;
539 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
540 flowi4_to_flowi(&fl4_dec), NULL,
543 dst_release(&rt->dst);
544 memcpy(fl4, &fl4_dec, sizeof(*fl4));
546 } else if (PTR_ERR(rt2) == -EPERM) {
548 dst_release(&rt->dst);
552 goto relookup_failed;
563 * Send an ICMP message in response to a situation
565 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
566 * MAY send more (we do).
567 * MUST NOT change this header information.
568 * MUST NOT reply to a multicast/broadcast IP address.
569 * MUST NOT reply to a multicast/broadcast MAC address.
570 * MUST reply to only the first fragment.
573 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
574 const struct ip_options *opt)
578 struct icmp_bxm icmp_param;
579 struct rtable *rt = skb_rtable(skb_in);
580 struct ipcm_cookie ipc;
590 net = dev_net(rt->dst.dev);
593 * Find the original header. It is expected to be valid, of course.
594 * Check this, icmp_send is called from the most obscure devices
597 iph = ip_hdr(skb_in);
599 if ((u8 *)iph < skb_in->head ||
600 (skb_network_header(skb_in) + sizeof(*iph)) >
601 skb_tail_pointer(skb_in))
605 * No replies to physical multicast/broadcast
607 if (skb_in->pkt_type != PACKET_HOST)
611 * Now check at the protocol level
613 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
617 * Only reply to fragment 0. We byte re-order the constant
618 * mask for efficiency.
620 if (iph->frag_off & htons(IP_OFFSET))
624 * If we send an ICMP error to an ICMP error a mess would result..
626 if (icmp_pointers[type].error) {
628 * We are an error, check if we are replying to an
631 if (iph->protocol == IPPROTO_ICMP) {
632 u8 _inner_type, *itp;
634 itp = skb_header_pointer(skb_in,
635 skb_network_header(skb_in) +
637 offsetof(struct icmphdr,
646 * Assume any unknown ICMP type is an error. This
647 * isn't specified by the RFC, but think about it..
649 if (*itp > NR_ICMP_TYPES ||
650 icmp_pointers[*itp].error)
655 /* Needed by both icmp_global_allow and icmp_xmit_lock */
658 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
659 * incoming dev is loopback. If outgoing dev change to not be
660 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
662 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
663 !icmpv4_global_allow(net, type, code))
666 sk = icmp_xmit_lock(net);
671 * Construct source address and options.
675 if (!(rt->rt_flags & RTCF_LOCAL)) {
676 struct net_device *dev = NULL;
679 if (rt_is_input_route(rt) &&
680 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
681 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
684 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
690 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
691 IPTOS_PREC_INTERNETCONTROL) :
693 mark = IP4_REPLY_MARK(net, skb_in->mark);
695 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
700 * Prepare data for ICMP header.
703 icmp_param.data.icmph.type = type;
704 icmp_param.data.icmph.code = code;
705 icmp_param.data.icmph.un.gateway = info;
706 icmp_param.data.icmph.checksum = 0;
707 icmp_param.skb = skb_in;
708 icmp_param.offset = skb_network_offset(skb_in);
709 inet_sk(sk)->tos = tos;
712 ipc.addr = iph->saddr;
713 ipc.opt = &icmp_param.replyopts.opt;
715 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
716 type, code, &icmp_param);
720 /* peer icmp_ratelimit */
721 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
724 /* RFC says return as much as we can without exceeding 576 bytes. */
726 room = dst_mtu(&rt->dst);
729 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
730 room -= sizeof(struct icmphdr);
732 icmp_param.data_len = skb_in->len - icmp_param.offset;
733 if (icmp_param.data_len > room)
734 icmp_param.data_len = room;
735 icmp_param.head_len = sizeof(struct icmphdr);
737 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
741 icmp_xmit_unlock(sk);
746 EXPORT_SYMBOL(__icmp_send);
749 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
751 const struct iphdr *iph = (const struct iphdr *) skb->data;
752 const struct net_protocol *ipprot;
753 int protocol = iph->protocol;
755 /* Checkin full IP header plus 8 bytes of protocol to
756 * avoid additional coding at protocol handlers.
758 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
759 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
763 raw_icmp_error(skb, protocol, info);
765 ipprot = rcu_dereference(inet_protos[protocol]);
766 if (ipprot && ipprot->err_handler)
767 ipprot->err_handler(skb, info);
770 static bool icmp_tag_validation(int proto)
775 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
781 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
782 * ICMP_PARAMETERPROB.
785 static bool icmp_unreach(struct sk_buff *skb)
787 const struct iphdr *iph;
788 struct icmphdr *icmph;
792 net = dev_net(skb_dst(skb)->dev);
795 * Incomplete header ?
796 * Only checks for the IP header, there should be an
797 * additional check for longer headers in upper levels.
800 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
803 icmph = icmp_hdr(skb);
804 iph = (const struct iphdr *)skb->data;
806 if (iph->ihl < 5) /* Mangled header, drop. */
809 switch (icmph->type) {
810 case ICMP_DEST_UNREACH:
811 switch (icmph->code & 15) {
812 case ICMP_NET_UNREACH:
813 case ICMP_HOST_UNREACH:
814 case ICMP_PROT_UNREACH:
815 case ICMP_PORT_UNREACH:
817 case ICMP_FRAG_NEEDED:
818 /* for documentation of the ip_no_pmtu_disc
820 * Documentation/networking/ip-sysctl.txt
822 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
824 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
830 if (!icmp_tag_validation(iph->protocol))
834 info = ntohs(icmph->un.frag.mtu);
838 net_dbg_ratelimited("%pI4: Source Route Failed\n",
844 if (icmph->code > NR_ICMP_UNREACH)
847 case ICMP_PARAMETERPROB:
848 info = ntohl(icmph->un.gateway) >> 24;
850 case ICMP_TIME_EXCEEDED:
851 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
852 if (icmph->code == ICMP_EXC_FRAGTIME)
858 * Throw it at our lower layers
860 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
862 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
864 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
869 * Check the other end isn't violating RFC 1122. Some routers send
870 * bogus responses to broadcast frames. If you see this message
871 * first check your netmask matches at both ends, if it does then
872 * get the other vendor to fix their kit.
875 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
876 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
877 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
879 icmph->type, icmph->code,
880 &iph->daddr, skb->dev->name);
884 icmp_socket_deliver(skb, info);
889 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
895 * Handle ICMP_REDIRECT.
898 static bool icmp_redirect(struct sk_buff *skb)
900 if (skb->len < sizeof(struct iphdr)) {
901 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
905 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
906 /* there aught to be a stat */
910 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
915 * Handle ICMP_ECHO ("ping") requests.
917 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
919 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
920 * included in the reply.
921 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
922 * echo requests, MUST have default=NOT.
923 * See also WRT handling of options once they are done and working.
926 static bool icmp_echo(struct sk_buff *skb)
930 net = dev_net(skb_dst(skb)->dev);
931 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
932 struct icmp_bxm icmp_param;
934 icmp_param.data.icmph = *icmp_hdr(skb);
935 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
936 icmp_param.skb = skb;
937 icmp_param.offset = 0;
938 icmp_param.data_len = skb->len;
939 icmp_param.head_len = sizeof(struct icmphdr);
940 icmp_reply(&icmp_param, skb);
942 /* should there be an ICMP stat for ignored echos? */
947 * Handle ICMP Timestamp requests.
948 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
949 * SHOULD be in the kernel for minimum random latency.
950 * MUST be accurate to a few minutes.
951 * MUST be updated at least at 15Hz.
953 static bool icmp_timestamp(struct sk_buff *skb)
955 struct icmp_bxm icmp_param;
963 * Fill in the current time as ms since midnight UT:
965 icmp_param.data.times[1] = inet_current_timestamp();
966 icmp_param.data.times[2] = icmp_param.data.times[1];
968 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
970 icmp_param.data.icmph = *icmp_hdr(skb);
971 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
972 icmp_param.data.icmph.code = 0;
973 icmp_param.skb = skb;
974 icmp_param.offset = 0;
975 icmp_param.data_len = 0;
976 icmp_param.head_len = sizeof(struct icmphdr) + 12;
977 icmp_reply(&icmp_param, skb);
981 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
985 static bool icmp_discard(struct sk_buff *skb)
987 /* pretend it was a success */
992 * Deal with incoming ICMP packets.
994 int icmp_rcv(struct sk_buff *skb)
996 struct icmphdr *icmph;
997 struct rtable *rt = skb_rtable(skb);
998 struct net *net = dev_net(rt->dst.dev);
1001 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1002 struct sec_path *sp = skb_sec_path(skb);
1005 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1009 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1012 nh = skb_network_offset(skb);
1013 skb_set_network_header(skb, sizeof(*icmph));
1015 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1018 skb_set_network_header(skb, nh);
1021 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1023 if (skb_checksum_simple_validate(skb))
1026 if (!pskb_pull(skb, sizeof(*icmph)))
1029 icmph = icmp_hdr(skb);
1031 ICMPMSGIN_INC_STATS(net, icmph->type);
1033 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1035 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1038 if (icmph->type > NR_ICMP_TYPES)
1043 * Parse the ICMP message
1046 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1048 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1049 * silently ignored (we let user decide with a sysctl).
1050 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1051 * discarded if to broadcast/multicast.
1053 if ((icmph->type == ICMP_ECHO ||
1054 icmph->type == ICMP_TIMESTAMP) &&
1055 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1058 if (icmph->type != ICMP_ECHO &&
1059 icmph->type != ICMP_TIMESTAMP &&
1060 icmph->type != ICMP_ADDRESS &&
1061 icmph->type != ICMP_ADDRESSREPLY) {
1066 success = icmp_pointers[icmph->type].handler(skb);
1070 return NET_RX_SUCCESS;
1077 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1079 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1083 int icmp_err(struct sk_buff *skb, u32 info)
1085 struct iphdr *iph = (struct iphdr *)skb->data;
1086 int offset = iph->ihl<<2;
1087 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1088 int type = icmp_hdr(skb)->type;
1089 int code = icmp_hdr(skb)->code;
1090 struct net *net = dev_net(skb->dev);
1093 * Use ping_err to handle all icmp errors except those
1094 * triggered by ICMP_ECHOREPLY which sent from kernel.
1096 if (icmph->type != ICMP_ECHOREPLY) {
1097 ping_err(skb, offset, info);
1101 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1102 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1103 else if (type == ICMP_REDIRECT)
1104 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1110 * This table is the definition of how we handle ICMP.
1112 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1113 [ICMP_ECHOREPLY] = {
1114 .handler = ping_rcv,
1117 .handler = icmp_discard,
1121 .handler = icmp_discard,
1124 [ICMP_DEST_UNREACH] = {
1125 .handler = icmp_unreach,
1128 [ICMP_SOURCE_QUENCH] = {
1129 .handler = icmp_unreach,
1133 .handler = icmp_redirect,
1137 .handler = icmp_discard,
1141 .handler = icmp_discard,
1145 .handler = icmp_echo,
1148 .handler = icmp_discard,
1152 .handler = icmp_discard,
1155 [ICMP_TIME_EXCEEDED] = {
1156 .handler = icmp_unreach,
1159 [ICMP_PARAMETERPROB] = {
1160 .handler = icmp_unreach,
1163 [ICMP_TIMESTAMP] = {
1164 .handler = icmp_timestamp,
1166 [ICMP_TIMESTAMPREPLY] = {
1167 .handler = icmp_discard,
1169 [ICMP_INFO_REQUEST] = {
1170 .handler = icmp_discard,
1172 [ICMP_INFO_REPLY] = {
1173 .handler = icmp_discard,
1176 .handler = icmp_discard,
1178 [ICMP_ADDRESSREPLY] = {
1179 .handler = icmp_discard,
1183 static void __net_exit icmp_sk_exit(struct net *net)
1187 for_each_possible_cpu(i)
1188 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1189 free_percpu(net->ipv4.icmp_sk);
1190 net->ipv4.icmp_sk = NULL;
1193 static int __net_init icmp_sk_init(struct net *net)
1197 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1198 if (!net->ipv4.icmp_sk)
1201 for_each_possible_cpu(i) {
1204 err = inet_ctl_sock_create(&sk, PF_INET,
1205 SOCK_RAW, IPPROTO_ICMP, net);
1209 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1211 /* Enough space for 2 64K ICMP packets, including
1212 * sk_buff/skb_shared_info struct overhead.
1214 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1217 * Speedup sock_wfree()
1219 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1220 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1223 /* Control parameters for ECHO replies. */
1224 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1225 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1227 /* Control parameter - ignore bogus broadcast responses? */
1228 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1231 * Configurable global rate limit.
1233 * ratelimit defines tokens/packet consumed for dst->rate_token
1234 * bucket ratemask defines which icmp types are ratelimited by
1235 * setting it's bit position.
1238 * dest unreachable (3), source quench (4),
1239 * time exceeded (11), parameter problem (12)
1242 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1243 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1244 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1249 for_each_possible_cpu(i)
1250 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1251 free_percpu(net->ipv4.icmp_sk);
1255 static struct pernet_operations __net_initdata icmp_sk_ops = {
1256 .init = icmp_sk_init,
1257 .exit = icmp_sk_exit,
1260 int __init icmp_init(void)
1262 return register_pernet_subsys(&icmp_sk_ops);