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 /* 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. The READ_ONCE() are paired
258 * with the following WRITE_ONCE() in this same function.
260 if (!READ_ONCE(icmp_global.credit)) {
261 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
266 spin_lock(&icmp_global.lock);
267 delta = min_t(u32, now - icmp_global.stamp, HZ);
268 if (delta >= HZ / 50) {
269 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
271 WRITE_ONCE(icmp_global.stamp, now);
273 credit = min_t(u32, icmp_global.credit + incr,
274 READ_ONCE(sysctl_icmp_msgs_burst));
276 /* We want to use a credit of one in average, but need to randomize
277 * it for security reasons.
279 credit = max_t(int, credit - prandom_u32_max(3), 0);
282 WRITE_ONCE(icmp_global.credit, credit);
283 spin_unlock(&icmp_global.lock);
286 EXPORT_SYMBOL(icmp_global_allow);
288 static bool icmpv4_mask_allow(struct net *net, int type, int code)
290 if (type > NR_ICMP_TYPES)
293 /* Don't limit PMTU discovery. */
294 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
297 /* Limit if icmp type is enabled in ratemask. */
298 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
304 static bool icmpv4_global_allow(struct net *net, int type, int code)
306 if (icmpv4_mask_allow(net, type, code))
309 if (icmp_global_allow())
316 * Send an ICMP frame.
319 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
320 struct flowi4 *fl4, int type, int code)
322 struct dst_entry *dst = &rt->dst;
323 struct inet_peer *peer;
327 if (icmpv4_mask_allow(net, type, code))
330 /* No rate limit on loopback */
331 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
334 vif = l3mdev_master_ifindex(dst->dev);
335 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
336 rc = inet_peer_xrlim_allow(peer,
337 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
345 * Maintain the counters used in the SNMP statistics for outgoing ICMP
347 void icmp_out_count(struct net *net, unsigned char type)
349 ICMPMSGOUT_INC_STATS(net, type);
350 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
354 * Checksum each fragment, and on the first include the headers and final
357 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
360 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
363 csum = skb_copy_and_csum_bits(icmp_param->skb,
364 icmp_param->offset + offset,
367 skb->csum = csum_block_add(skb->csum, csum, odd);
368 if (icmp_pointers[icmp_param->data.icmph.type].error)
369 nf_ct_attach(skb, icmp_param->skb);
373 static void icmp_push_reply(struct icmp_bxm *icmp_param,
375 struct ipcm_cookie *ipc, struct rtable **rt)
380 sk = icmp_sk(dev_net((*rt)->dst.dev));
381 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
382 icmp_param->data_len+icmp_param->head_len,
383 icmp_param->head_len,
384 ipc, rt, MSG_DONTWAIT) < 0) {
385 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
386 ip_flush_pending_frames(sk);
387 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
388 struct icmphdr *icmph = icmp_hdr(skb);
390 struct sk_buff *skb1;
392 skb_queue_walk(&sk->sk_write_queue, skb1) {
393 csum = csum_add(csum, skb1->csum);
395 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
397 icmp_param->head_len, csum);
398 icmph->checksum = csum_fold(csum);
399 skb->ip_summed = CHECKSUM_NONE;
400 ip_push_pending_frames(sk, fl4);
405 * Driving logic for building and sending ICMP messages.
408 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
410 struct ipcm_cookie ipc;
411 struct rtable *rt = skb_rtable(skb);
412 struct net *net = dev_net(rt->dst.dev);
415 struct inet_sock *inet;
417 u32 mark = IP4_REPLY_MARK(net, skb->mark);
418 int type = icmp_param->data.icmph.type;
419 int code = icmp_param->data.icmph.code;
421 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
424 /* Needed by both icmp_global_allow and icmp_xmit_lock */
427 /* global icmp_msgs_per_sec */
428 if (!icmpv4_global_allow(net, type, code))
431 sk = icmp_xmit_lock(net);
436 icmp_param->data.icmph.checksum = 0;
439 inet->tos = ip_hdr(skb)->tos;
441 daddr = ipc.addr = ip_hdr(skb)->saddr;
442 saddr = fib_compute_spec_dst(skb);
444 if (icmp_param->replyopts.opt.opt.optlen) {
445 ipc.opt = &icmp_param->replyopts.opt;
446 if (ipc.opt->opt.srr)
447 daddr = icmp_param->replyopts.opt.opt.faddr;
449 memset(&fl4, 0, sizeof(fl4));
452 fl4.flowi4_mark = mark;
453 fl4.flowi4_uid = sock_net_uid(net, NULL);
454 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
455 fl4.flowi4_proto = IPPROTO_ICMP;
456 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
457 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
458 rt = ip_route_output_key(net, &fl4);
461 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
462 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
465 icmp_xmit_unlock(sk);
471 * The device used for looking up which routing table to use for sending an ICMP
472 * error is preferably the source whenever it is set, which should ensure the
473 * icmp error can be sent to the source host, else lookup using the routing
474 * table of the destination device, else use the main routing table (index 0).
476 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
478 struct net_device *route_lookup_dev = NULL;
481 route_lookup_dev = skb->dev;
482 else if (skb_dst(skb))
483 route_lookup_dev = skb_dst(skb)->dev;
484 return route_lookup_dev;
487 static struct rtable *icmp_route_lookup(struct net *net,
489 struct sk_buff *skb_in,
490 const struct iphdr *iph,
491 __be32 saddr, u8 tos, u32 mark,
493 struct icmp_bxm *param)
495 struct net_device *route_lookup_dev;
496 struct rtable *rt, *rt2;
497 struct flowi4 fl4_dec;
500 memset(fl4, 0, sizeof(*fl4));
501 fl4->daddr = (param->replyopts.opt.opt.srr ?
502 param->replyopts.opt.opt.faddr : iph->saddr);
504 fl4->flowi4_mark = mark;
505 fl4->flowi4_uid = sock_net_uid(net, NULL);
506 fl4->flowi4_tos = RT_TOS(tos);
507 fl4->flowi4_proto = IPPROTO_ICMP;
508 fl4->fl4_icmp_type = type;
509 fl4->fl4_icmp_code = code;
510 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
511 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
513 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
514 rt = ip_route_output_key_hash(net, fl4, skb_in);
518 /* No need to clone since we're just using its address. */
521 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
522 flowi4_to_flowi(fl4), NULL, 0);
526 } else if (PTR_ERR(rt) == -EPERM) {
531 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
533 goto relookup_failed;
535 if (inet_addr_type_dev_table(net, route_lookup_dev,
536 fl4_dec.saddr) == RTN_LOCAL) {
537 rt2 = __ip_route_output_key(net, &fl4_dec);
541 struct flowi4 fl4_2 = {};
542 unsigned long orefdst;
544 fl4_2.daddr = fl4_dec.saddr;
545 rt2 = ip_route_output_key(net, &fl4_2);
548 goto relookup_failed;
551 orefdst = skb_in->_skb_refdst; /* save old refdst */
552 skb_dst_set(skb_in, NULL);
553 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
554 RT_TOS(tos), rt2->dst.dev);
556 dst_release(&rt2->dst);
557 rt2 = skb_rtable(skb_in);
558 skb_in->_skb_refdst = orefdst; /* restore old refdst */
562 goto relookup_failed;
564 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
565 flowi4_to_flowi(&fl4_dec), NULL,
568 dst_release(&rt->dst);
569 memcpy(fl4, &fl4_dec, sizeof(*fl4));
571 } else if (PTR_ERR(rt2) == -EPERM) {
573 dst_release(&rt->dst);
577 goto relookup_failed;
588 * Send an ICMP message in response to a situation
590 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
591 * MAY send more (we do).
592 * MUST NOT change this header information.
593 * MUST NOT reply to a multicast/broadcast IP address.
594 * MUST NOT reply to a multicast/broadcast MAC address.
595 * MUST reply to only the first fragment.
598 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
599 const struct ip_options *opt)
603 struct icmp_bxm icmp_param;
604 struct rtable *rt = skb_rtable(skb_in);
605 struct ipcm_cookie ipc;
617 net = dev_net(rt->dst.dev);
618 else if (skb_in->dev)
619 net = dev_net(skb_in->dev);
624 * Find the original header. It is expected to be valid, of course.
625 * Check this, icmp_send is called from the most obscure devices
628 iph = ip_hdr(skb_in);
630 if ((u8 *)iph < skb_in->head ||
631 (skb_network_header(skb_in) + sizeof(*iph)) >
632 skb_tail_pointer(skb_in))
636 * No replies to physical multicast/broadcast
638 if (skb_in->pkt_type != PACKET_HOST)
642 * Now check at the protocol level
644 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
648 * Only reply to fragment 0. We byte re-order the constant
649 * mask for efficiency.
651 if (iph->frag_off & htons(IP_OFFSET))
655 * If we send an ICMP error to an ICMP error a mess would result..
657 if (icmp_pointers[type].error) {
659 * We are an error, check if we are replying to an
662 if (iph->protocol == IPPROTO_ICMP) {
663 u8 _inner_type, *itp;
665 itp = skb_header_pointer(skb_in,
666 skb_network_header(skb_in) +
668 offsetof(struct icmphdr,
677 * Assume any unknown ICMP type is an error. This
678 * isn't specified by the RFC, but think about it..
680 if (*itp > NR_ICMP_TYPES ||
681 icmp_pointers[*itp].error)
686 /* Needed by both icmp_global_allow and icmp_xmit_lock */
689 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
690 * incoming dev is loopback. If outgoing dev change to not be
691 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
693 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
694 !icmpv4_global_allow(net, type, code))
697 sk = icmp_xmit_lock(net);
702 * Construct source address and options.
706 if (!(rt->rt_flags & RTCF_LOCAL)) {
707 struct net_device *dev = NULL;
710 if (rt_is_input_route(rt) &&
711 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
712 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
715 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
721 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
722 IPTOS_PREC_INTERNETCONTROL) :
724 mark = IP4_REPLY_MARK(net, skb_in->mark);
726 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
731 * Prepare data for ICMP header.
734 icmp_param.data.icmph.type = type;
735 icmp_param.data.icmph.code = code;
736 icmp_param.data.icmph.un.gateway = info;
737 icmp_param.data.icmph.checksum = 0;
738 icmp_param.skb = skb_in;
739 icmp_param.offset = skb_network_offset(skb_in);
740 inet_sk(sk)->tos = tos;
743 ipc.addr = iph->saddr;
744 ipc.opt = &icmp_param.replyopts.opt;
746 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
747 type, code, &icmp_param);
751 /* peer icmp_ratelimit */
752 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
755 /* RFC says return as much as we can without exceeding 576 bytes. */
757 room = dst_mtu(&rt->dst);
760 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
761 room -= sizeof(struct icmphdr);
762 /* Guard against tiny mtu. We need to include at least one
763 * IP network header for this message to make any sense.
765 if (room <= (int)sizeof(struct iphdr))
768 icmp_param.data_len = skb_in->len - icmp_param.offset;
769 if (icmp_param.data_len > room)
770 icmp_param.data_len = room;
771 icmp_param.head_len = sizeof(struct icmphdr);
773 /* if we don't have a source address at this point, fall back to the
774 * dummy address instead of sending out a packet with a source address
778 fl4.saddr = htonl(INADDR_DUMMY);
780 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
784 icmp_xmit_unlock(sk);
789 EXPORT_SYMBOL(__icmp_send);
791 #if IS_ENABLED(CONFIG_NF_NAT)
792 #include <net/netfilter/nf_conntrack.h>
793 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
795 struct sk_buff *cloned_skb = NULL;
796 struct ip_options opts = { 0 };
797 enum ip_conntrack_info ctinfo;
801 ct = nf_ct_get(skb_in, &ctinfo);
802 if (!ct || !(ct->status & IPS_SRC_NAT)) {
803 __icmp_send(skb_in, type, code, info, &opts);
807 if (skb_shared(skb_in))
808 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
810 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
811 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
812 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
813 skb_network_offset(skb_in) + sizeof(struct iphdr))))
816 orig_ip = ip_hdr(skb_in)->saddr;
817 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
818 __icmp_send(skb_in, type, code, info, &opts);
819 ip_hdr(skb_in)->saddr = orig_ip;
821 consume_skb(cloned_skb);
823 EXPORT_SYMBOL(icmp_ndo_send);
826 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
828 const struct iphdr *iph = (const struct iphdr *) skb->data;
829 const struct net_protocol *ipprot;
830 int protocol = iph->protocol;
832 /* Checkin full IP header plus 8 bytes of protocol to
833 * avoid additional coding at protocol handlers.
835 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
836 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
840 raw_icmp_error(skb, protocol, info);
842 ipprot = rcu_dereference(inet_protos[protocol]);
843 if (ipprot && ipprot->err_handler)
844 ipprot->err_handler(skb, info);
847 static bool icmp_tag_validation(int proto)
852 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
858 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
859 * ICMP_PARAMETERPROB.
862 static bool icmp_unreach(struct sk_buff *skb)
864 const struct iphdr *iph;
865 struct icmphdr *icmph;
869 net = dev_net(skb_dst(skb)->dev);
872 * Incomplete header ?
873 * Only checks for the IP header, there should be an
874 * additional check for longer headers in upper levels.
877 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
880 icmph = icmp_hdr(skb);
881 iph = (const struct iphdr *)skb->data;
883 if (iph->ihl < 5) /* Mangled header, drop. */
886 switch (icmph->type) {
887 case ICMP_DEST_UNREACH:
888 switch (icmph->code & 15) {
889 case ICMP_NET_UNREACH:
890 case ICMP_HOST_UNREACH:
891 case ICMP_PROT_UNREACH:
892 case ICMP_PORT_UNREACH:
894 case ICMP_FRAG_NEEDED:
895 /* for documentation of the ip_no_pmtu_disc
897 * Documentation/networking/ip-sysctl.txt
899 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
901 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
907 if (!icmp_tag_validation(iph->protocol))
911 info = ntohs(icmph->un.frag.mtu);
915 net_dbg_ratelimited("%pI4: Source Route Failed\n",
921 if (icmph->code > NR_ICMP_UNREACH)
924 case ICMP_PARAMETERPROB:
925 info = ntohl(icmph->un.gateway) >> 24;
927 case ICMP_TIME_EXCEEDED:
928 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
929 if (icmph->code == ICMP_EXC_FRAGTIME)
935 * Throw it at our lower layers
937 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
939 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
941 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
946 * Check the other end isn't violating RFC 1122. Some routers send
947 * bogus responses to broadcast frames. If you see this message
948 * first check your netmask matches at both ends, if it does then
949 * get the other vendor to fix their kit.
952 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
953 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
954 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
956 icmph->type, icmph->code,
957 &iph->daddr, skb->dev->name);
961 icmp_socket_deliver(skb, info);
966 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
972 * Handle ICMP_REDIRECT.
975 static bool icmp_redirect(struct sk_buff *skb)
977 if (skb->len < sizeof(struct iphdr)) {
978 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
982 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
983 /* there aught to be a stat */
987 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
992 * Handle ICMP_ECHO ("ping") requests.
994 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
996 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
997 * included in the reply.
998 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
999 * echo requests, MUST have default=NOT.
1000 * See also WRT handling of options once they are done and working.
1003 static bool icmp_echo(struct sk_buff *skb)
1007 net = dev_net(skb_dst(skb)->dev);
1008 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
1009 struct icmp_bxm icmp_param;
1011 icmp_param.data.icmph = *icmp_hdr(skb);
1012 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1013 icmp_param.skb = skb;
1014 icmp_param.offset = 0;
1015 icmp_param.data_len = skb->len;
1016 icmp_param.head_len = sizeof(struct icmphdr);
1017 icmp_reply(&icmp_param, skb);
1019 /* should there be an ICMP stat for ignored echos? */
1024 * Handle ICMP Timestamp requests.
1025 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1026 * SHOULD be in the kernel for minimum random latency.
1027 * MUST be accurate to a few minutes.
1028 * MUST be updated at least at 15Hz.
1030 static bool icmp_timestamp(struct sk_buff *skb)
1032 struct icmp_bxm icmp_param;
1040 * Fill in the current time as ms since midnight UT:
1042 icmp_param.data.times[1] = inet_current_timestamp();
1043 icmp_param.data.times[2] = icmp_param.data.times[1];
1045 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1047 icmp_param.data.icmph = *icmp_hdr(skb);
1048 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1049 icmp_param.data.icmph.code = 0;
1050 icmp_param.skb = skb;
1051 icmp_param.offset = 0;
1052 icmp_param.data_len = 0;
1053 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1054 icmp_reply(&icmp_param, skb);
1058 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1062 static bool icmp_discard(struct sk_buff *skb)
1064 /* pretend it was a success */
1069 * Deal with incoming ICMP packets.
1071 int icmp_rcv(struct sk_buff *skb)
1073 struct icmphdr *icmph;
1074 struct rtable *rt = skb_rtable(skb);
1075 struct net *net = dev_net(rt->dst.dev);
1078 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1079 struct sec_path *sp = skb_sec_path(skb);
1082 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1086 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1089 nh = skb_network_offset(skb);
1090 skb_set_network_header(skb, sizeof(*icmph));
1092 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1095 skb_set_network_header(skb, nh);
1098 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1100 if (skb_checksum_simple_validate(skb))
1103 if (!pskb_pull(skb, sizeof(*icmph)))
1106 icmph = icmp_hdr(skb);
1108 ICMPMSGIN_INC_STATS(net, icmph->type);
1110 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1112 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1115 if (icmph->type > NR_ICMP_TYPES)
1120 * Parse the ICMP message
1123 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1125 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1126 * silently ignored (we let user decide with a sysctl).
1127 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1128 * discarded if to broadcast/multicast.
1130 if ((icmph->type == ICMP_ECHO ||
1131 icmph->type == ICMP_TIMESTAMP) &&
1132 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1135 if (icmph->type != ICMP_ECHO &&
1136 icmph->type != ICMP_TIMESTAMP &&
1137 icmph->type != ICMP_ADDRESS &&
1138 icmph->type != ICMP_ADDRESSREPLY) {
1143 success = icmp_pointers[icmph->type].handler(skb);
1147 return NET_RX_SUCCESS;
1154 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1156 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1160 void icmp_err(struct sk_buff *skb, u32 info)
1162 struct iphdr *iph = (struct iphdr *)skb->data;
1163 int offset = iph->ihl<<2;
1164 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1165 int type = icmp_hdr(skb)->type;
1166 int code = icmp_hdr(skb)->code;
1167 struct net *net = dev_net(skb->dev);
1170 * Use ping_err to handle all icmp errors except those
1171 * triggered by ICMP_ECHOREPLY which sent from kernel.
1173 if (icmph->type != ICMP_ECHOREPLY) {
1174 ping_err(skb, offset, info);
1178 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1179 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1180 else if (type == ICMP_REDIRECT)
1181 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1185 * This table is the definition of how we handle ICMP.
1187 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1188 [ICMP_ECHOREPLY] = {
1189 .handler = ping_rcv,
1192 .handler = icmp_discard,
1196 .handler = icmp_discard,
1199 [ICMP_DEST_UNREACH] = {
1200 .handler = icmp_unreach,
1203 [ICMP_SOURCE_QUENCH] = {
1204 .handler = icmp_unreach,
1208 .handler = icmp_redirect,
1212 .handler = icmp_discard,
1216 .handler = icmp_discard,
1220 .handler = icmp_echo,
1223 .handler = icmp_discard,
1227 .handler = icmp_discard,
1230 [ICMP_TIME_EXCEEDED] = {
1231 .handler = icmp_unreach,
1234 [ICMP_PARAMETERPROB] = {
1235 .handler = icmp_unreach,
1238 [ICMP_TIMESTAMP] = {
1239 .handler = icmp_timestamp,
1241 [ICMP_TIMESTAMPREPLY] = {
1242 .handler = icmp_discard,
1244 [ICMP_INFO_REQUEST] = {
1245 .handler = icmp_discard,
1247 [ICMP_INFO_REPLY] = {
1248 .handler = icmp_discard,
1251 .handler = icmp_discard,
1253 [ICMP_ADDRESSREPLY] = {
1254 .handler = icmp_discard,
1258 static void __net_exit icmp_sk_exit(struct net *net)
1262 for_each_possible_cpu(i)
1263 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1264 free_percpu(net->ipv4.icmp_sk);
1265 net->ipv4.icmp_sk = NULL;
1268 static int __net_init icmp_sk_init(struct net *net)
1272 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1273 if (!net->ipv4.icmp_sk)
1276 for_each_possible_cpu(i) {
1279 err = inet_ctl_sock_create(&sk, PF_INET,
1280 SOCK_RAW, IPPROTO_ICMP, net);
1284 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1286 /* Enough space for 2 64K ICMP packets, including
1287 * sk_buff/skb_shared_info struct overhead.
1289 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1292 * Speedup sock_wfree()
1294 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1295 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1298 /* Control parameters for ECHO replies. */
1299 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1300 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1302 /* Control parameter - ignore bogus broadcast responses? */
1303 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1306 * Configurable global rate limit.
1308 * ratelimit defines tokens/packet consumed for dst->rate_token
1309 * bucket ratemask defines which icmp types are ratelimited by
1310 * setting it's bit position.
1313 * dest unreachable (3), source quench (4),
1314 * time exceeded (11), parameter problem (12)
1317 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1318 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1319 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1324 for_each_possible_cpu(i)
1325 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1326 free_percpu(net->ipv4.icmp_sk);
1330 static struct pernet_operations __net_initdata icmp_sk_ops = {
1331 .init = icmp_sk_init,
1332 .exit = icmp_sk_exit,
1335 int __init icmp_init(void)
1337 return register_pernet_subsys(&icmp_sk_ops);