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 <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 static inline struct sock *icmp_xmit_lock(struct net *net)
220 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
221 /* This can happen if the output path signals a
222 * dst_link_failure() for an outgoing ICMP packet.
230 static inline void icmp_xmit_unlock(struct sock *sk)
232 spin_unlock_bh(&sk->sk_lock.slock);
235 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
236 int sysctl_icmp_msgs_burst __read_mostly = 50;
243 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
247 * icmp_global_allow - Are we allowed to send one more ICMP message ?
249 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
250 * Returns false if we reached the limit and can not send another packet.
251 * Note: called with BH disabled
253 bool icmp_global_allow(void)
255 u32 credit, delta, incr = 0, now = (u32)jiffies;
258 /* Check if token bucket is empty and cannot be refilled
259 * without taking the spinlock.
261 if (!icmp_global.credit) {
262 delta = min_t(u32, now - icmp_global.stamp, HZ);
267 spin_lock(&icmp_global.lock);
268 delta = min_t(u32, now - icmp_global.stamp, HZ);
269 if (delta >= HZ / 50) {
270 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
272 icmp_global.stamp = now;
274 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
279 icmp_global.credit = credit;
280 spin_unlock(&icmp_global.lock);
283 EXPORT_SYMBOL(icmp_global_allow);
285 static bool icmpv4_mask_allow(struct net *net, int type, int code)
287 if (type > NR_ICMP_TYPES)
290 /* Don't limit PMTU discovery. */
291 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
294 /* Limit if icmp type is enabled in ratemask. */
295 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
301 static bool icmpv4_global_allow(struct net *net, int type, int code)
303 if (icmpv4_mask_allow(net, type, code))
306 if (icmp_global_allow())
313 * Send an ICMP frame.
316 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
317 struct flowi4 *fl4, int type, int code)
319 struct dst_entry *dst = &rt->dst;
320 struct inet_peer *peer;
324 if (icmpv4_mask_allow(net, type, code))
327 /* No rate limit on loopback */
328 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
331 vif = l3mdev_master_ifindex(dst->dev);
332 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
333 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
341 * Maintain the counters used in the SNMP statistics for outgoing ICMP
343 void icmp_out_count(struct net *net, unsigned char type)
345 ICMPMSGOUT_INC_STATS(net, type);
346 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
350 * Checksum each fragment, and on the first include the headers and final
353 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
356 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
359 csum = skb_copy_and_csum_bits(icmp_param->skb,
360 icmp_param->offset + offset,
363 skb->csum = csum_block_add(skb->csum, csum, odd);
364 if (icmp_pointers[icmp_param->data.icmph.type].error)
365 nf_ct_attach(skb, icmp_param->skb);
369 static void icmp_push_reply(struct icmp_bxm *icmp_param,
371 struct ipcm_cookie *ipc, struct rtable **rt)
376 sk = icmp_sk(dev_net((*rt)->dst.dev));
377 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
378 icmp_param->data_len+icmp_param->head_len,
379 icmp_param->head_len,
380 ipc, rt, MSG_DONTWAIT) < 0) {
381 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
382 ip_flush_pending_frames(sk);
383 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
384 struct icmphdr *icmph = icmp_hdr(skb);
386 struct sk_buff *skb1;
388 skb_queue_walk(&sk->sk_write_queue, skb1) {
389 csum = csum_add(csum, skb1->csum);
391 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
393 icmp_param->head_len, csum);
394 icmph->checksum = csum_fold(csum);
395 skb->ip_summed = CHECKSUM_NONE;
396 ip_push_pending_frames(sk, fl4);
401 * Driving logic for building and sending ICMP messages.
404 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
406 struct ipcm_cookie ipc;
407 struct rtable *rt = skb_rtable(skb);
408 struct net *net = dev_net(rt->dst.dev);
411 struct inet_sock *inet;
413 u32 mark = IP4_REPLY_MARK(net, skb->mark);
414 int type = icmp_param->data.icmph.type;
415 int code = icmp_param->data.icmph.code;
417 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
420 sk = icmp_xmit_lock(net);
425 /* global icmp_msgs_per_sec */
426 if (!icmpv4_global_allow(net, type, code))
429 icmp_param->data.icmph.checksum = 0;
431 inet->tos = ip_hdr(skb)->tos;
433 daddr = ipc.addr = ip_hdr(skb)->saddr;
434 saddr = fib_compute_spec_dst(skb);
440 if (icmp_param->replyopts.opt.opt.optlen) {
441 ipc.opt = &icmp_param->replyopts.opt;
442 if (ipc.opt->opt.srr)
443 daddr = icmp_param->replyopts.opt.opt.faddr;
445 memset(&fl4, 0, sizeof(fl4));
448 fl4.flowi4_mark = mark;
449 fl4.flowi4_uid = sock_net_uid(net, NULL);
450 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
451 fl4.flowi4_proto = IPPROTO_ICMP;
452 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
453 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
454 rt = ip_route_output_key(net, &fl4);
457 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
458 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
461 icmp_xmit_unlock(sk);
464 #ifdef CONFIG_IP_ROUTE_MULTIPATH
466 /* Source and destination is swapped. See ip_multipath_icmp_hash */
467 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
469 const struct iphdr *iph = ip_hdr(skb);
471 return fib_multipath_hash(iph->daddr, iph->saddr);
476 #define icmp_multipath_hash_skb(skb) (-1)
480 static struct rtable *icmp_route_lookup(struct net *net,
482 struct sk_buff *skb_in,
483 const struct iphdr *iph,
484 __be32 saddr, u8 tos, u32 mark,
486 struct icmp_bxm *param)
488 struct rtable *rt, *rt2;
489 struct flowi4 fl4_dec;
492 memset(fl4, 0, sizeof(*fl4));
493 fl4->daddr = (param->replyopts.opt.opt.srr ?
494 param->replyopts.opt.opt.faddr : iph->saddr);
496 fl4->flowi4_mark = mark;
497 fl4->flowi4_uid = sock_net_uid(net, NULL);
498 fl4->flowi4_tos = RT_TOS(tos);
499 fl4->flowi4_proto = IPPROTO_ICMP;
500 fl4->fl4_icmp_type = type;
501 fl4->fl4_icmp_code = code;
502 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
504 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
505 rt = __ip_route_output_key_hash(net, fl4,
506 icmp_multipath_hash_skb(skb_in));
510 /* No need to clone since we're just using its address. */
513 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
514 flowi4_to_flowi(fl4), NULL, 0);
518 } else if (PTR_ERR(rt) == -EPERM) {
523 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
525 goto relookup_failed;
527 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
528 fl4_dec.saddr) == RTN_LOCAL) {
529 rt2 = __ip_route_output_key(net, &fl4_dec);
533 struct flowi4 fl4_2 = {};
534 unsigned long orefdst;
536 fl4_2.daddr = fl4_dec.saddr;
537 rt2 = ip_route_output_key(net, &fl4_2);
540 goto relookup_failed;
543 orefdst = skb_in->_skb_refdst; /* save old refdst */
544 skb_dst_set(skb_in, NULL);
545 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
546 RT_TOS(tos), rt2->dst.dev);
548 dst_release(&rt2->dst);
549 rt2 = skb_rtable(skb_in);
550 skb_in->_skb_refdst = orefdst; /* restore old refdst */
554 goto relookup_failed;
556 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
557 flowi4_to_flowi(&fl4_dec), NULL,
560 dst_release(&rt->dst);
561 memcpy(fl4, &fl4_dec, sizeof(*fl4));
563 } else if (PTR_ERR(rt2) == -EPERM) {
565 dst_release(&rt->dst);
569 goto relookup_failed;
580 * Send an ICMP message in response to a situation
582 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
583 * MAY send more (we do).
584 * MUST NOT change this header information.
585 * MUST NOT reply to a multicast/broadcast IP address.
586 * MUST NOT reply to a multicast/broadcast MAC address.
587 * MUST reply to only the first fragment.
590 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
594 struct icmp_bxm icmp_param;
595 struct rtable *rt = skb_rtable(skb_in);
596 struct ipcm_cookie ipc;
606 net = dev_net(rt->dst.dev);
609 * Find the original header. It is expected to be valid, of course.
610 * Check this, icmp_send is called from the most obscure devices
613 iph = ip_hdr(skb_in);
615 if ((u8 *)iph < skb_in->head ||
616 (skb_network_header(skb_in) + sizeof(*iph)) >
617 skb_tail_pointer(skb_in))
621 * No replies to physical multicast/broadcast
623 if (skb_in->pkt_type != PACKET_HOST)
627 * Now check at the protocol level
629 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
633 * Only reply to fragment 0. We byte re-order the constant
634 * mask for efficiency.
636 if (iph->frag_off & htons(IP_OFFSET))
640 * If we send an ICMP error to an ICMP error a mess would result..
642 if (icmp_pointers[type].error) {
644 * We are an error, check if we are replying to an
647 if (iph->protocol == IPPROTO_ICMP) {
648 u8 _inner_type, *itp;
650 itp = skb_header_pointer(skb_in,
651 skb_network_header(skb_in) +
653 offsetof(struct icmphdr,
662 * Assume any unknown ICMP type is an error. This
663 * isn't specified by the RFC, but think about it..
665 if (*itp > NR_ICMP_TYPES ||
666 icmp_pointers[*itp].error)
671 sk = icmp_xmit_lock(net);
675 /* Check global sysctl_icmp_msgs_per_sec ratelimit */
676 if (!icmpv4_global_allow(net, type, code))
680 * Construct source address and options.
684 if (!(rt->rt_flags & RTCF_LOCAL)) {
685 struct net_device *dev = NULL;
688 if (rt_is_input_route(rt) &&
689 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
690 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
693 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
699 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
700 IPTOS_PREC_INTERNETCONTROL) :
702 mark = IP4_REPLY_MARK(net, skb_in->mark);
704 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
709 * Prepare data for ICMP header.
712 icmp_param.data.icmph.type = type;
713 icmp_param.data.icmph.code = code;
714 icmp_param.data.icmph.un.gateway = info;
715 icmp_param.data.icmph.checksum = 0;
716 icmp_param.skb = skb_in;
717 icmp_param.offset = skb_network_offset(skb_in);
718 inet_sk(sk)->tos = tos;
720 ipc.addr = iph->saddr;
721 ipc.opt = &icmp_param.replyopts.opt;
726 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
727 type, code, &icmp_param);
731 /* peer icmp_ratelimit */
732 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
735 /* RFC says return as much as we can without exceeding 576 bytes. */
737 room = dst_mtu(&rt->dst);
740 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
741 room -= sizeof(struct icmphdr);
743 icmp_param.data_len = skb_in->len - icmp_param.offset;
744 if (icmp_param.data_len > room)
745 icmp_param.data_len = room;
746 icmp_param.head_len = sizeof(struct icmphdr);
748 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
752 icmp_xmit_unlock(sk);
755 EXPORT_SYMBOL(icmp_send);
758 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
760 const struct iphdr *iph = (const struct iphdr *) skb->data;
761 const struct net_protocol *ipprot;
762 int protocol = iph->protocol;
764 /* Checkin full IP header plus 8 bytes of protocol to
765 * avoid additional coding at protocol handlers.
767 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
768 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
772 raw_icmp_error(skb, protocol, info);
774 ipprot = rcu_dereference(inet_protos[protocol]);
775 if (ipprot && ipprot->err_handler)
776 ipprot->err_handler(skb, info);
779 static bool icmp_tag_validation(int proto)
784 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
790 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
791 * ICMP_PARAMETERPROB.
794 static bool icmp_unreach(struct sk_buff *skb)
796 const struct iphdr *iph;
797 struct icmphdr *icmph;
801 net = dev_net(skb_dst(skb)->dev);
804 * Incomplete header ?
805 * Only checks for the IP header, there should be an
806 * additional check for longer headers in upper levels.
809 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
812 icmph = icmp_hdr(skb);
813 iph = (const struct iphdr *)skb->data;
815 if (iph->ihl < 5) /* Mangled header, drop. */
818 if (icmph->type == ICMP_DEST_UNREACH) {
819 switch (icmph->code & 15) {
820 case ICMP_NET_UNREACH:
821 case ICMP_HOST_UNREACH:
822 case ICMP_PROT_UNREACH:
823 case ICMP_PORT_UNREACH:
825 case ICMP_FRAG_NEEDED:
826 /* for documentation of the ip_no_pmtu_disc
828 * Documentation/networking/ip-sysctl.txt
830 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
832 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
838 if (!icmp_tag_validation(iph->protocol))
842 info = ntohs(icmph->un.frag.mtu);
846 net_dbg_ratelimited("%pI4: Source Route Failed\n",
852 if (icmph->code > NR_ICMP_UNREACH)
854 } else if (icmph->type == ICMP_PARAMETERPROB)
855 info = ntohl(icmph->un.gateway) >> 24;
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];
967 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
969 icmp_param.data.icmph = *icmp_hdr(skb);
970 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
971 icmp_param.data.icmph.code = 0;
972 icmp_param.skb = skb;
973 icmp_param.offset = 0;
974 icmp_param.data_len = 0;
975 icmp_param.head_len = sizeof(struct icmphdr) + 12;
976 icmp_reply(&icmp_param, skb);
980 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
984 static bool icmp_discard(struct sk_buff *skb)
986 /* pretend it was a success */
991 * Deal with incoming ICMP packets.
993 int icmp_rcv(struct sk_buff *skb)
995 struct icmphdr *icmph;
996 struct rtable *rt = skb_rtable(skb);
997 struct net *net = dev_net(rt->dst.dev);
1000 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1001 struct sec_path *sp = skb_sec_path(skb);
1004 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1008 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1011 nh = skb_network_offset(skb);
1012 skb_set_network_header(skb, sizeof(*icmph));
1014 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1017 skb_set_network_header(skb, nh);
1020 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1022 if (skb_checksum_simple_validate(skb))
1025 if (!pskb_pull(skb, sizeof(*icmph)))
1028 icmph = icmp_hdr(skb);
1030 ICMPMSGIN_INC_STATS(net, icmph->type);
1032 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1034 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1037 if (icmph->type > NR_ICMP_TYPES)
1042 * Parse the ICMP message
1045 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1047 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1048 * silently ignored (we let user decide with a sysctl).
1049 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1050 * discarded if to broadcast/multicast.
1052 if ((icmph->type == ICMP_ECHO ||
1053 icmph->type == ICMP_TIMESTAMP) &&
1054 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1057 if (icmph->type != ICMP_ECHO &&
1058 icmph->type != ICMP_TIMESTAMP &&
1059 icmph->type != ICMP_ADDRESS &&
1060 icmph->type != ICMP_ADDRESSREPLY) {
1065 success = icmp_pointers[icmph->type].handler(skb);
1069 return NET_RX_SUCCESS;
1076 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1078 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1082 void icmp_err(struct sk_buff *skb, u32 info)
1084 struct iphdr *iph = (struct iphdr *)skb->data;
1085 int offset = iph->ihl<<2;
1086 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1087 int type = icmp_hdr(skb)->type;
1088 int code = icmp_hdr(skb)->code;
1089 struct net *net = dev_net(skb->dev);
1092 * Use ping_err to handle all icmp errors except those
1093 * triggered by ICMP_ECHOREPLY which sent from kernel.
1095 if (icmph->type != ICMP_ECHOREPLY) {
1096 ping_err(skb, offset, info);
1100 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1101 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1102 else if (type == ICMP_REDIRECT)
1103 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1107 * This table is the definition of how we handle ICMP.
1109 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1110 [ICMP_ECHOREPLY] = {
1111 .handler = ping_rcv,
1114 .handler = icmp_discard,
1118 .handler = icmp_discard,
1121 [ICMP_DEST_UNREACH] = {
1122 .handler = icmp_unreach,
1125 [ICMP_SOURCE_QUENCH] = {
1126 .handler = icmp_unreach,
1130 .handler = icmp_redirect,
1134 .handler = icmp_discard,
1138 .handler = icmp_discard,
1142 .handler = icmp_echo,
1145 .handler = icmp_discard,
1149 .handler = icmp_discard,
1152 [ICMP_TIME_EXCEEDED] = {
1153 .handler = icmp_unreach,
1156 [ICMP_PARAMETERPROB] = {
1157 .handler = icmp_unreach,
1160 [ICMP_TIMESTAMP] = {
1161 .handler = icmp_timestamp,
1163 [ICMP_TIMESTAMPREPLY] = {
1164 .handler = icmp_discard,
1166 [ICMP_INFO_REQUEST] = {
1167 .handler = icmp_discard,
1169 [ICMP_INFO_REPLY] = {
1170 .handler = icmp_discard,
1173 .handler = icmp_discard,
1175 [ICMP_ADDRESSREPLY] = {
1176 .handler = icmp_discard,
1180 static void __net_exit icmp_sk_exit(struct net *net)
1184 for_each_possible_cpu(i)
1185 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1186 free_percpu(net->ipv4.icmp_sk);
1187 net->ipv4.icmp_sk = NULL;
1190 static int __net_init icmp_sk_init(struct net *net)
1194 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1195 if (!net->ipv4.icmp_sk)
1198 for_each_possible_cpu(i) {
1201 err = inet_ctl_sock_create(&sk, PF_INET,
1202 SOCK_RAW, IPPROTO_ICMP, net);
1206 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1208 /* Enough space for 2 64K ICMP packets, including
1209 * sk_buff/skb_shared_info struct overhead.
1211 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1214 * Speedup sock_wfree()
1216 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1217 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1220 /* Control parameters for ECHO replies. */
1221 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1222 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1224 /* Control parameter - ignore bogus broadcast responses? */
1225 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1228 * Configurable global rate limit.
1230 * ratelimit defines tokens/packet consumed for dst->rate_token
1231 * bucket ratemask defines which icmp types are ratelimited by
1232 * setting it's bit position.
1235 * dest unreachable (3), source quench (4),
1236 * time exceeded (11), parameter problem (12)
1239 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1240 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1241 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1246 for_each_possible_cpu(i)
1247 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1248 free_percpu(net->ipv4.icmp_sk);
1252 static struct pernet_operations __net_initdata icmp_sk_ops = {
1253 .init = icmp_sk_init,
1254 .exit = icmp_sk_exit,
1257 int __init icmp_init(void)
1259 return register_pernet_subsys(&icmp_sk_ops);