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);
286 * Send an ICMP frame.
289 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
290 struct flowi4 *fl4, int type, int code)
292 struct dst_entry *dst = &rt->dst;
295 if (type > NR_ICMP_TYPES)
298 /* Don't limit PMTU discovery. */
299 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
302 /* No rate limit on loopback */
303 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
306 /* Limit if icmp type is enabled in ratemask. */
307 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
311 if (icmp_global_allow()) {
312 int vif = l3mdev_master_ifindex(dst->dev);
313 struct inet_peer *peer;
315 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
316 rc = inet_peer_xrlim_allow(peer,
317 net->ipv4.sysctl_icmp_ratelimit);
326 * Maintain the counters used in the SNMP statistics for outgoing ICMP
328 void icmp_out_count(struct net *net, unsigned char type)
330 ICMPMSGOUT_INC_STATS(net, type);
331 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
335 * Checksum each fragment, and on the first include the headers and final
338 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
341 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
344 csum = skb_copy_and_csum_bits(icmp_param->skb,
345 icmp_param->offset + offset,
348 skb->csum = csum_block_add(skb->csum, csum, odd);
349 if (icmp_pointers[icmp_param->data.icmph.type].error)
350 nf_ct_attach(skb, icmp_param->skb);
354 static void icmp_push_reply(struct icmp_bxm *icmp_param,
356 struct ipcm_cookie *ipc, struct rtable **rt)
361 sk = icmp_sk(dev_net((*rt)->dst.dev));
362 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
363 icmp_param->data_len+icmp_param->head_len,
364 icmp_param->head_len,
365 ipc, rt, MSG_DONTWAIT) < 0) {
366 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
367 ip_flush_pending_frames(sk);
368 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
369 struct icmphdr *icmph = icmp_hdr(skb);
371 struct sk_buff *skb1;
373 skb_queue_walk(&sk->sk_write_queue, skb1) {
374 csum = csum_add(csum, skb1->csum);
376 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
378 icmp_param->head_len, csum);
379 icmph->checksum = csum_fold(csum);
380 skb->ip_summed = CHECKSUM_NONE;
381 ip_push_pending_frames(sk, fl4);
386 * Driving logic for building and sending ICMP messages.
389 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
391 struct ipcm_cookie ipc;
392 struct rtable *rt = skb_rtable(skb);
393 struct net *net = dev_net(rt->dst.dev);
396 struct inet_sock *inet;
398 u32 mark = IP4_REPLY_MARK(net, skb->mark);
400 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
403 sk = icmp_xmit_lock(net);
408 icmp_param->data.icmph.checksum = 0;
410 inet->tos = ip_hdr(skb)->tos;
412 daddr = ipc.addr = ip_hdr(skb)->saddr;
413 saddr = fib_compute_spec_dst(skb);
419 if (icmp_param->replyopts.opt.opt.optlen) {
420 ipc.opt = &icmp_param->replyopts.opt;
421 if (ipc.opt->opt.srr)
422 daddr = icmp_param->replyopts.opt.opt.faddr;
424 memset(&fl4, 0, sizeof(fl4));
427 fl4.flowi4_mark = mark;
428 fl4.flowi4_uid = sock_net_uid(net, NULL);
429 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
430 fl4.flowi4_proto = IPPROTO_ICMP;
431 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
432 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
433 rt = ip_route_output_key(net, &fl4);
436 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
437 icmp_param->data.icmph.code))
438 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
441 icmp_xmit_unlock(sk);
444 #ifdef CONFIG_IP_ROUTE_MULTIPATH
446 /* Source and destination is swapped. See ip_multipath_icmp_hash */
447 static int icmp_multipath_hash_skb(const struct sk_buff *skb)
449 const struct iphdr *iph = ip_hdr(skb);
451 return fib_multipath_hash(iph->daddr, iph->saddr);
456 #define icmp_multipath_hash_skb(skb) (-1)
460 static struct rtable *icmp_route_lookup(struct net *net,
462 struct sk_buff *skb_in,
463 const struct iphdr *iph,
464 __be32 saddr, u8 tos, u32 mark,
466 struct icmp_bxm *param)
468 struct rtable *rt, *rt2;
469 struct flowi4 fl4_dec;
472 memset(fl4, 0, sizeof(*fl4));
473 fl4->daddr = (param->replyopts.opt.opt.srr ?
474 param->replyopts.opt.opt.faddr : iph->saddr);
476 fl4->flowi4_mark = mark;
477 fl4->flowi4_uid = sock_net_uid(net, NULL);
478 fl4->flowi4_tos = RT_TOS(tos);
479 fl4->flowi4_proto = IPPROTO_ICMP;
480 fl4->fl4_icmp_type = type;
481 fl4->fl4_icmp_code = code;
482 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
484 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
485 rt = __ip_route_output_key_hash(net, fl4,
486 icmp_multipath_hash_skb(skb_in));
490 /* No need to clone since we're just using its address. */
493 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
494 flowi4_to_flowi(fl4), NULL, 0);
498 } else if (PTR_ERR(rt) == -EPERM) {
503 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
505 goto relookup_failed;
507 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
508 fl4_dec.saddr) == RTN_LOCAL) {
509 rt2 = __ip_route_output_key(net, &fl4_dec);
513 struct flowi4 fl4_2 = {};
514 unsigned long orefdst;
516 fl4_2.daddr = fl4_dec.saddr;
517 rt2 = ip_route_output_key(net, &fl4_2);
520 goto relookup_failed;
523 orefdst = skb_in->_skb_refdst; /* save old refdst */
524 skb_dst_set(skb_in, NULL);
525 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
526 RT_TOS(tos), rt2->dst.dev);
528 dst_release(&rt2->dst);
529 rt2 = skb_rtable(skb_in);
530 skb_in->_skb_refdst = orefdst; /* restore old refdst */
534 goto relookup_failed;
536 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
537 flowi4_to_flowi(&fl4_dec), NULL,
540 dst_release(&rt->dst);
541 memcpy(fl4, &fl4_dec, sizeof(*fl4));
543 } else if (PTR_ERR(rt2) == -EPERM) {
545 dst_release(&rt->dst);
549 goto relookup_failed;
560 * Send an ICMP message in response to a situation
562 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
563 * MAY send more (we do).
564 * MUST NOT change this header information.
565 * MUST NOT reply to a multicast/broadcast IP address.
566 * MUST NOT reply to a multicast/broadcast MAC address.
567 * MUST reply to only the first fragment.
570 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
574 struct icmp_bxm *icmp_param;
575 struct rtable *rt = skb_rtable(skb_in);
576 struct ipcm_cookie ipc;
586 net = dev_net(rt->dst.dev);
589 * Find the original header. It is expected to be valid, of course.
590 * Check this, icmp_send is called from the most obscure devices
593 iph = ip_hdr(skb_in);
595 if ((u8 *)iph < skb_in->head ||
596 (skb_network_header(skb_in) + sizeof(*iph)) >
597 skb_tail_pointer(skb_in))
601 * No replies to physical multicast/broadcast
603 if (skb_in->pkt_type != PACKET_HOST)
607 * Now check at the protocol level
609 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
613 * Only reply to fragment 0. We byte re-order the constant
614 * mask for efficiency.
616 if (iph->frag_off & htons(IP_OFFSET))
620 * If we send an ICMP error to an ICMP error a mess would result..
622 if (icmp_pointers[type].error) {
624 * We are an error, check if we are replying to an
627 if (iph->protocol == IPPROTO_ICMP) {
628 u8 _inner_type, *itp;
630 itp = skb_header_pointer(skb_in,
631 skb_network_header(skb_in) +
633 offsetof(struct icmphdr,
642 * Assume any unknown ICMP type is an error. This
643 * isn't specified by the RFC, but think about it..
645 if (*itp > NR_ICMP_TYPES ||
646 icmp_pointers[*itp].error)
651 icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
655 sk = icmp_xmit_lock(net);
660 * Construct source address and options.
664 if (!(rt->rt_flags & RTCF_LOCAL)) {
665 struct net_device *dev = NULL;
668 if (rt_is_input_route(rt) &&
669 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
670 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
673 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
679 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
680 IPTOS_PREC_INTERNETCONTROL) :
682 mark = IP4_REPLY_MARK(net, skb_in->mark);
684 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
689 * Prepare data for ICMP header.
692 icmp_param->data.icmph.type = type;
693 icmp_param->data.icmph.code = code;
694 icmp_param->data.icmph.un.gateway = info;
695 icmp_param->data.icmph.checksum = 0;
696 icmp_param->skb = skb_in;
697 icmp_param->offset = skb_network_offset(skb_in);
698 inet_sk(sk)->tos = tos;
700 ipc.addr = iph->saddr;
701 ipc.opt = &icmp_param->replyopts.opt;
706 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
707 type, code, icmp_param);
711 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
714 /* RFC says return as much as we can without exceeding 576 bytes. */
716 room = dst_mtu(&rt->dst);
719 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
720 room -= sizeof(struct icmphdr);
722 icmp_param->data_len = skb_in->len - icmp_param->offset;
723 if (icmp_param->data_len > room)
724 icmp_param->data_len = room;
725 icmp_param->head_len = sizeof(struct icmphdr);
727 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
731 icmp_xmit_unlock(sk);
736 EXPORT_SYMBOL(icmp_send);
739 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
741 const struct iphdr *iph = (const struct iphdr *) skb->data;
742 const struct net_protocol *ipprot;
743 int protocol = iph->protocol;
745 /* Checkin full IP header plus 8 bytes of protocol to
746 * avoid additional coding at protocol handlers.
748 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
749 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
753 raw_icmp_error(skb, protocol, info);
755 ipprot = rcu_dereference(inet_protos[protocol]);
756 if (ipprot && ipprot->err_handler)
757 ipprot->err_handler(skb, info);
760 static bool icmp_tag_validation(int proto)
765 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
771 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
772 * ICMP_PARAMETERPROB.
775 static bool icmp_unreach(struct sk_buff *skb)
777 const struct iphdr *iph;
778 struct icmphdr *icmph;
782 net = dev_net(skb_dst(skb)->dev);
785 * Incomplete header ?
786 * Only checks for the IP header, there should be an
787 * additional check for longer headers in upper levels.
790 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
793 icmph = icmp_hdr(skb);
794 iph = (const struct iphdr *)skb->data;
796 if (iph->ihl < 5) /* Mangled header, drop. */
799 if (icmph->type == ICMP_DEST_UNREACH) {
800 switch (icmph->code & 15) {
801 case ICMP_NET_UNREACH:
802 case ICMP_HOST_UNREACH:
803 case ICMP_PROT_UNREACH:
804 case ICMP_PORT_UNREACH:
806 case ICMP_FRAG_NEEDED:
807 /* for documentation of the ip_no_pmtu_disc
809 * Documentation/networking/ip-sysctl.txt
811 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
813 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
819 if (!icmp_tag_validation(iph->protocol))
823 info = ntohs(icmph->un.frag.mtu);
827 net_dbg_ratelimited("%pI4: Source Route Failed\n",
833 if (icmph->code > NR_ICMP_UNREACH)
835 } else if (icmph->type == ICMP_PARAMETERPROB)
836 info = ntohl(icmph->un.gateway) >> 24;
839 * Throw it at our lower layers
841 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
843 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
845 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
850 * Check the other end isn't violating RFC 1122. Some routers send
851 * bogus responses to broadcast frames. If you see this message
852 * first check your netmask matches at both ends, if it does then
853 * get the other vendor to fix their kit.
856 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
857 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
858 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
860 icmph->type, icmph->code,
861 &iph->daddr, skb->dev->name);
865 icmp_socket_deliver(skb, info);
870 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
876 * Handle ICMP_REDIRECT.
879 static bool icmp_redirect(struct sk_buff *skb)
881 if (skb->len < sizeof(struct iphdr)) {
882 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
886 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
887 /* there aught to be a stat */
891 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
896 * Handle ICMP_ECHO ("ping") requests.
898 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
900 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
901 * included in the reply.
902 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
903 * echo requests, MUST have default=NOT.
904 * See also WRT handling of options once they are done and working.
907 static bool icmp_echo(struct sk_buff *skb)
911 net = dev_net(skb_dst(skb)->dev);
912 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
913 struct icmp_bxm icmp_param;
915 icmp_param.data.icmph = *icmp_hdr(skb);
916 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
917 icmp_param.skb = skb;
918 icmp_param.offset = 0;
919 icmp_param.data_len = skb->len;
920 icmp_param.head_len = sizeof(struct icmphdr);
921 icmp_reply(&icmp_param, skb);
923 /* should there be an ICMP stat for ignored echos? */
928 * Handle ICMP Timestamp requests.
929 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
930 * SHOULD be in the kernel for minimum random latency.
931 * MUST be accurate to a few minutes.
932 * MUST be updated at least at 15Hz.
934 static bool icmp_timestamp(struct sk_buff *skb)
936 struct icmp_bxm icmp_param;
944 * Fill in the current time as ms since midnight UT:
946 icmp_param.data.times[1] = inet_current_timestamp();
947 icmp_param.data.times[2] = icmp_param.data.times[1];
948 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
950 icmp_param.data.icmph = *icmp_hdr(skb);
951 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
952 icmp_param.data.icmph.code = 0;
953 icmp_param.skb = skb;
954 icmp_param.offset = 0;
955 icmp_param.data_len = 0;
956 icmp_param.head_len = sizeof(struct icmphdr) + 12;
957 icmp_reply(&icmp_param, skb);
961 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
965 static bool icmp_discard(struct sk_buff *skb)
967 /* pretend it was a success */
972 * Deal with incoming ICMP packets.
974 int icmp_rcv(struct sk_buff *skb)
976 struct icmphdr *icmph;
977 struct rtable *rt = skb_rtable(skb);
978 struct net *net = dev_net(rt->dst.dev);
981 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
982 struct sec_path *sp = skb_sec_path(skb);
985 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
989 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
992 nh = skb_network_offset(skb);
993 skb_set_network_header(skb, sizeof(*icmph));
995 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
998 skb_set_network_header(skb, nh);
1001 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1003 if (skb_checksum_simple_validate(skb))
1006 if (!pskb_pull(skb, sizeof(*icmph)))
1009 icmph = icmp_hdr(skb);
1011 ICMPMSGIN_INC_STATS(net, icmph->type);
1013 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1015 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1018 if (icmph->type > NR_ICMP_TYPES)
1023 * Parse the ICMP message
1026 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1028 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1029 * silently ignored (we let user decide with a sysctl).
1030 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1031 * discarded if to broadcast/multicast.
1033 if ((icmph->type == ICMP_ECHO ||
1034 icmph->type == ICMP_TIMESTAMP) &&
1035 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1038 if (icmph->type != ICMP_ECHO &&
1039 icmph->type != ICMP_TIMESTAMP &&
1040 icmph->type != ICMP_ADDRESS &&
1041 icmph->type != ICMP_ADDRESSREPLY) {
1046 success = icmp_pointers[icmph->type].handler(skb);
1050 return NET_RX_SUCCESS;
1057 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1059 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1063 void icmp_err(struct sk_buff *skb, u32 info)
1065 struct iphdr *iph = (struct iphdr *)skb->data;
1066 int offset = iph->ihl<<2;
1067 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1068 int type = icmp_hdr(skb)->type;
1069 int code = icmp_hdr(skb)->code;
1070 struct net *net = dev_net(skb->dev);
1073 * Use ping_err to handle all icmp errors except those
1074 * triggered by ICMP_ECHOREPLY which sent from kernel.
1076 if (icmph->type != ICMP_ECHOREPLY) {
1077 ping_err(skb, offset, info);
1081 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1082 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1083 else if (type == ICMP_REDIRECT)
1084 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1088 * This table is the definition of how we handle ICMP.
1090 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1091 [ICMP_ECHOREPLY] = {
1092 .handler = ping_rcv,
1095 .handler = icmp_discard,
1099 .handler = icmp_discard,
1102 [ICMP_DEST_UNREACH] = {
1103 .handler = icmp_unreach,
1106 [ICMP_SOURCE_QUENCH] = {
1107 .handler = icmp_unreach,
1111 .handler = icmp_redirect,
1115 .handler = icmp_discard,
1119 .handler = icmp_discard,
1123 .handler = icmp_echo,
1126 .handler = icmp_discard,
1130 .handler = icmp_discard,
1133 [ICMP_TIME_EXCEEDED] = {
1134 .handler = icmp_unreach,
1137 [ICMP_PARAMETERPROB] = {
1138 .handler = icmp_unreach,
1141 [ICMP_TIMESTAMP] = {
1142 .handler = icmp_timestamp,
1144 [ICMP_TIMESTAMPREPLY] = {
1145 .handler = icmp_discard,
1147 [ICMP_INFO_REQUEST] = {
1148 .handler = icmp_discard,
1150 [ICMP_INFO_REPLY] = {
1151 .handler = icmp_discard,
1154 .handler = icmp_discard,
1156 [ICMP_ADDRESSREPLY] = {
1157 .handler = icmp_discard,
1161 static void __net_exit icmp_sk_exit(struct net *net)
1165 for_each_possible_cpu(i)
1166 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1167 free_percpu(net->ipv4.icmp_sk);
1168 net->ipv4.icmp_sk = NULL;
1171 static int __net_init icmp_sk_init(struct net *net)
1175 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1176 if (!net->ipv4.icmp_sk)
1179 for_each_possible_cpu(i) {
1182 err = inet_ctl_sock_create(&sk, PF_INET,
1183 SOCK_RAW, IPPROTO_ICMP, net);
1187 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1189 /* Enough space for 2 64K ICMP packets, including
1190 * sk_buff/skb_shared_info struct overhead.
1192 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1195 * Speedup sock_wfree()
1197 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1198 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1201 /* Control parameters for ECHO replies. */
1202 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1203 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1205 /* Control parameter - ignore bogus broadcast responses? */
1206 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1209 * Configurable global rate limit.
1211 * ratelimit defines tokens/packet consumed for dst->rate_token
1212 * bucket ratemask defines which icmp types are ratelimited by
1213 * setting it's bit position.
1216 * dest unreachable (3), source quench (4),
1217 * time exceeded (11), parameter problem (12)
1220 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1221 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1222 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1227 for_each_possible_cpu(i)
1228 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1229 free_percpu(net->ipv4.icmp_sk);
1233 static struct pernet_operations __net_initdata icmp_sk_ops = {
1234 .init = icmp_sk_init,
1235 .exit = icmp_sk_exit,
1238 int __init icmp_init(void)
1240 return register_pernet_subsys(&icmp_sk_ops);