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;
438 inet->tos = ip_hdr(skb)->tos;
440 daddr = ipc.addr = ip_hdr(skb)->saddr;
441 saddr = fib_compute_spec_dst(skb);
447 if (icmp_param->replyopts.opt.opt.optlen) {
448 ipc.opt = &icmp_param->replyopts.opt;
449 if (ipc.opt->opt.srr)
450 daddr = icmp_param->replyopts.opt.opt.faddr;
452 memset(&fl4, 0, sizeof(fl4));
455 fl4.flowi4_mark = mark;
456 fl4.flowi4_uid = sock_net_uid(net, NULL);
457 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
458 fl4.flowi4_proto = IPPROTO_ICMP;
459 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
460 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
461 rt = ip_route_output_key(net, &fl4);
464 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
465 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
468 icmp_xmit_unlock(sk);
474 * The device used for looking up which routing table to use for sending an ICMP
475 * error is preferably the source whenever it is set, which should ensure the
476 * icmp error can be sent to the source host, else lookup using the routing
477 * table of the destination device, else use the main routing table (index 0).
479 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
481 struct net_device *route_lookup_dev = NULL;
484 route_lookup_dev = skb->dev;
485 else if (skb_dst(skb))
486 route_lookup_dev = skb_dst(skb)->dev;
487 return route_lookup_dev;
490 static struct rtable *icmp_route_lookup(struct net *net,
492 struct sk_buff *skb_in,
493 const struct iphdr *iph,
494 __be32 saddr, u8 tos, u32 mark,
496 struct icmp_bxm *param)
498 struct net_device *route_lookup_dev;
499 struct rtable *rt, *rt2;
500 struct flowi4 fl4_dec;
503 memset(fl4, 0, sizeof(*fl4));
504 fl4->daddr = (param->replyopts.opt.opt.srr ?
505 param->replyopts.opt.opt.faddr : iph->saddr);
507 fl4->flowi4_mark = mark;
508 fl4->flowi4_uid = sock_net_uid(net, NULL);
509 fl4->flowi4_tos = RT_TOS(tos);
510 fl4->flowi4_proto = IPPROTO_ICMP;
511 fl4->fl4_icmp_type = type;
512 fl4->fl4_icmp_code = code;
513 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
514 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
516 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
517 rt = ip_route_output_key_hash(net, fl4, skb_in);
521 /* No need to clone since we're just using its address. */
524 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
525 flowi4_to_flowi(fl4), NULL, 0);
529 } else if (PTR_ERR(rt) == -EPERM) {
534 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
536 goto relookup_failed;
538 if (inet_addr_type_dev_table(net, route_lookup_dev,
539 fl4_dec.saddr) == RTN_LOCAL) {
540 rt2 = __ip_route_output_key(net, &fl4_dec);
544 struct flowi4 fl4_2 = {};
545 unsigned long orefdst;
547 fl4_2.daddr = fl4_dec.saddr;
548 rt2 = ip_route_output_key(net, &fl4_2);
551 goto relookup_failed;
554 orefdst = skb_in->_skb_refdst; /* save old refdst */
555 skb_dst_set(skb_in, NULL);
556 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
557 RT_TOS(tos), rt2->dst.dev);
559 dst_release(&rt2->dst);
560 rt2 = skb_rtable(skb_in);
561 skb_in->_skb_refdst = orefdst; /* restore old refdst */
565 goto relookup_failed;
567 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
568 flowi4_to_flowi(&fl4_dec), NULL,
571 dst_release(&rt->dst);
572 memcpy(fl4, &fl4_dec, sizeof(*fl4));
574 } else if (PTR_ERR(rt2) == -EPERM) {
576 dst_release(&rt->dst);
580 goto relookup_failed;
591 * Send an ICMP message in response to a situation
593 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
594 * MAY send more (we do).
595 * MUST NOT change this header information.
596 * MUST NOT reply to a multicast/broadcast IP address.
597 * MUST NOT reply to a multicast/broadcast MAC address.
598 * MUST reply to only the first fragment.
601 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
602 const struct ip_options *opt)
606 struct icmp_bxm icmp_param;
607 struct rtable *rt = skb_rtable(skb_in);
608 struct ipcm_cookie ipc;
618 net = dev_net(rt->dst.dev);
621 * Find the original header. It is expected to be valid, of course.
622 * Check this, icmp_send is called from the most obscure devices
625 iph = ip_hdr(skb_in);
627 if ((u8 *)iph < skb_in->head ||
628 (skb_network_header(skb_in) + sizeof(*iph)) >
629 skb_tail_pointer(skb_in))
633 * No replies to physical multicast/broadcast
635 if (skb_in->pkt_type != PACKET_HOST)
639 * Now check at the protocol level
641 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
645 * Only reply to fragment 0. We byte re-order the constant
646 * mask for efficiency.
648 if (iph->frag_off & htons(IP_OFFSET))
652 * If we send an ICMP error to an ICMP error a mess would result..
654 if (icmp_pointers[type].error) {
656 * We are an error, check if we are replying to an
659 if (iph->protocol == IPPROTO_ICMP) {
660 u8 _inner_type, *itp;
662 itp = skb_header_pointer(skb_in,
663 skb_network_header(skb_in) +
665 offsetof(struct icmphdr,
674 * Assume any unknown ICMP type is an error. This
675 * isn't specified by the RFC, but think about it..
677 if (*itp > NR_ICMP_TYPES ||
678 icmp_pointers[*itp].error)
683 /* Needed by both icmp_global_allow and icmp_xmit_lock */
686 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
687 * incoming dev is loopback. If outgoing dev change to not be
688 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
690 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
691 !icmpv4_global_allow(net, type, code))
694 sk = icmp_xmit_lock(net);
699 * Construct source address and options.
703 if (!(rt->rt_flags & RTCF_LOCAL)) {
704 struct net_device *dev = NULL;
707 if (rt_is_input_route(rt) &&
708 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
709 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
712 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
718 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
719 IPTOS_PREC_INTERNETCONTROL) :
721 mark = IP4_REPLY_MARK(net, skb_in->mark);
723 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
728 * Prepare data for ICMP header.
731 icmp_param.data.icmph.type = type;
732 icmp_param.data.icmph.code = code;
733 icmp_param.data.icmph.un.gateway = info;
734 icmp_param.data.icmph.checksum = 0;
735 icmp_param.skb = skb_in;
736 icmp_param.offset = skb_network_offset(skb_in);
737 inet_sk(sk)->tos = tos;
739 ipc.addr = iph->saddr;
740 ipc.opt = &icmp_param.replyopts.opt;
745 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
746 type, code, &icmp_param);
750 /* peer icmp_ratelimit */
751 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
754 /* RFC says return as much as we can without exceeding 576 bytes. */
756 room = dst_mtu(&rt->dst);
759 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
760 room -= sizeof(struct icmphdr);
761 /* Guard against tiny mtu. We need to include at least one
762 * IP network header for this message to make any sense.
764 if (room <= (int)sizeof(struct iphdr))
767 icmp_param.data_len = skb_in->len - icmp_param.offset;
768 if (icmp_param.data_len > room)
769 icmp_param.data_len = room;
770 icmp_param.head_len = sizeof(struct icmphdr);
772 /* if we don't have a source address at this point, fall back to the
773 * dummy address instead of sending out a packet with a source address
777 fl4.saddr = htonl(INADDR_DUMMY);
779 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
783 icmp_xmit_unlock(sk);
788 EXPORT_SYMBOL(__icmp_send);
790 #if IS_ENABLED(CONFIG_NF_NAT)
791 #include <net/netfilter/nf_conntrack.h>
792 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
794 struct sk_buff *cloned_skb = NULL;
795 struct ip_options opts = { 0 };
796 enum ip_conntrack_info ctinfo;
800 ct = nf_ct_get(skb_in, &ctinfo);
801 if (!ct || !(ct->status & IPS_SRC_NAT)) {
802 __icmp_send(skb_in, type, code, info, &opts);
806 if (skb_shared(skb_in))
807 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
809 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
810 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
811 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
812 skb_network_offset(skb_in) + sizeof(struct iphdr))))
815 orig_ip = ip_hdr(skb_in)->saddr;
816 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
817 __icmp_send(skb_in, type, code, info, &opts);
818 ip_hdr(skb_in)->saddr = orig_ip;
820 consume_skb(cloned_skb);
822 EXPORT_SYMBOL(icmp_ndo_send);
825 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
827 const struct iphdr *iph = (const struct iphdr *) skb->data;
828 const struct net_protocol *ipprot;
829 int protocol = iph->protocol;
831 /* Checkin full IP header plus 8 bytes of protocol to
832 * avoid additional coding at protocol handlers.
834 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
835 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
839 raw_icmp_error(skb, protocol, info);
841 ipprot = rcu_dereference(inet_protos[protocol]);
842 if (ipprot && ipprot->err_handler)
843 ipprot->err_handler(skb, info);
846 static bool icmp_tag_validation(int proto)
851 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
857 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
858 * ICMP_PARAMETERPROB.
861 static bool icmp_unreach(struct sk_buff *skb)
863 const struct iphdr *iph;
864 struct icmphdr *icmph;
868 net = dev_net(skb_dst(skb)->dev);
871 * Incomplete header ?
872 * Only checks for the IP header, there should be an
873 * additional check for longer headers in upper levels.
876 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
879 icmph = icmp_hdr(skb);
880 iph = (const struct iphdr *)skb->data;
882 if (iph->ihl < 5) /* Mangled header, drop. */
885 switch (icmph->type) {
886 case ICMP_DEST_UNREACH:
887 switch (icmph->code & 15) {
888 case ICMP_NET_UNREACH:
889 case ICMP_HOST_UNREACH:
890 case ICMP_PROT_UNREACH:
891 case ICMP_PORT_UNREACH:
893 case ICMP_FRAG_NEEDED:
894 /* for documentation of the ip_no_pmtu_disc
896 * Documentation/networking/ip-sysctl.txt
898 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
900 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
906 if (!icmp_tag_validation(iph->protocol))
910 info = ntohs(icmph->un.frag.mtu);
914 net_dbg_ratelimited("%pI4: Source Route Failed\n",
920 if (icmph->code > NR_ICMP_UNREACH)
923 case ICMP_PARAMETERPROB:
924 info = ntohl(icmph->un.gateway) >> 24;
926 case ICMP_TIME_EXCEEDED:
927 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
928 if (icmph->code == ICMP_EXC_FRAGTIME)
934 * Throw it at our lower layers
936 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
938 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
940 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
945 * Check the other end isn't violating RFC 1122. Some routers send
946 * bogus responses to broadcast frames. If you see this message
947 * first check your netmask matches at both ends, if it does then
948 * get the other vendor to fix their kit.
951 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
952 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
953 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
955 icmph->type, icmph->code,
956 &iph->daddr, skb->dev->name);
960 icmp_socket_deliver(skb, info);
965 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
971 * Handle ICMP_REDIRECT.
974 static bool icmp_redirect(struct sk_buff *skb)
976 if (skb->len < sizeof(struct iphdr)) {
977 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
981 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
982 /* there aught to be a stat */
986 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
991 * Handle ICMP_ECHO ("ping") requests.
993 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
995 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
996 * included in the reply.
997 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
998 * echo requests, MUST have default=NOT.
999 * See also WRT handling of options once they are done and working.
1002 static bool icmp_echo(struct sk_buff *skb)
1006 net = dev_net(skb_dst(skb)->dev);
1007 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
1008 struct icmp_bxm icmp_param;
1010 icmp_param.data.icmph = *icmp_hdr(skb);
1011 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1012 icmp_param.skb = skb;
1013 icmp_param.offset = 0;
1014 icmp_param.data_len = skb->len;
1015 icmp_param.head_len = sizeof(struct icmphdr);
1016 icmp_reply(&icmp_param, skb);
1018 /* should there be an ICMP stat for ignored echos? */
1023 * Handle ICMP Timestamp requests.
1024 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1025 * SHOULD be in the kernel for minimum random latency.
1026 * MUST be accurate to a few minutes.
1027 * MUST be updated at least at 15Hz.
1029 static bool icmp_timestamp(struct sk_buff *skb)
1031 struct icmp_bxm icmp_param;
1039 * Fill in the current time as ms since midnight UT:
1041 icmp_param.data.times[1] = inet_current_timestamp();
1042 icmp_param.data.times[2] = icmp_param.data.times[1];
1043 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
1045 icmp_param.data.icmph = *icmp_hdr(skb);
1046 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1047 icmp_param.data.icmph.code = 0;
1048 icmp_param.skb = skb;
1049 icmp_param.offset = 0;
1050 icmp_param.data_len = 0;
1051 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1052 icmp_reply(&icmp_param, skb);
1056 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1060 static bool icmp_discard(struct sk_buff *skb)
1062 /* pretend it was a success */
1067 * Deal with incoming ICMP packets.
1069 int icmp_rcv(struct sk_buff *skb)
1071 struct icmphdr *icmph;
1072 struct rtable *rt = skb_rtable(skb);
1073 struct net *net = dev_net(rt->dst.dev);
1076 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1077 struct sec_path *sp = skb_sec_path(skb);
1080 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1084 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1087 nh = skb_network_offset(skb);
1088 skb_set_network_header(skb, sizeof(*icmph));
1090 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1093 skb_set_network_header(skb, nh);
1096 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1098 if (skb_checksum_simple_validate(skb))
1101 if (!pskb_pull(skb, sizeof(*icmph)))
1104 icmph = icmp_hdr(skb);
1106 ICMPMSGIN_INC_STATS(net, icmph->type);
1108 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1110 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1113 if (icmph->type > NR_ICMP_TYPES)
1118 * Parse the ICMP message
1121 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1123 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1124 * silently ignored (we let user decide with a sysctl).
1125 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1126 * discarded if to broadcast/multicast.
1128 if ((icmph->type == ICMP_ECHO ||
1129 icmph->type == ICMP_TIMESTAMP) &&
1130 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1133 if (icmph->type != ICMP_ECHO &&
1134 icmph->type != ICMP_TIMESTAMP &&
1135 icmph->type != ICMP_ADDRESS &&
1136 icmph->type != ICMP_ADDRESSREPLY) {
1141 success = icmp_pointers[icmph->type].handler(skb);
1145 return NET_RX_SUCCESS;
1152 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1154 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1158 void icmp_err(struct sk_buff *skb, u32 info)
1160 struct iphdr *iph = (struct iphdr *)skb->data;
1161 int offset = iph->ihl<<2;
1162 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1163 int type = icmp_hdr(skb)->type;
1164 int code = icmp_hdr(skb)->code;
1165 struct net *net = dev_net(skb->dev);
1168 * Use ping_err to handle all icmp errors except those
1169 * triggered by ICMP_ECHOREPLY which sent from kernel.
1171 if (icmph->type != ICMP_ECHOREPLY) {
1172 ping_err(skb, offset, info);
1176 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1177 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1178 else if (type == ICMP_REDIRECT)
1179 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1183 * This table is the definition of how we handle ICMP.
1185 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1186 [ICMP_ECHOREPLY] = {
1187 .handler = ping_rcv,
1190 .handler = icmp_discard,
1194 .handler = icmp_discard,
1197 [ICMP_DEST_UNREACH] = {
1198 .handler = icmp_unreach,
1201 [ICMP_SOURCE_QUENCH] = {
1202 .handler = icmp_unreach,
1206 .handler = icmp_redirect,
1210 .handler = icmp_discard,
1214 .handler = icmp_discard,
1218 .handler = icmp_echo,
1221 .handler = icmp_discard,
1225 .handler = icmp_discard,
1228 [ICMP_TIME_EXCEEDED] = {
1229 .handler = icmp_unreach,
1232 [ICMP_PARAMETERPROB] = {
1233 .handler = icmp_unreach,
1236 [ICMP_TIMESTAMP] = {
1237 .handler = icmp_timestamp,
1239 [ICMP_TIMESTAMPREPLY] = {
1240 .handler = icmp_discard,
1242 [ICMP_INFO_REQUEST] = {
1243 .handler = icmp_discard,
1245 [ICMP_INFO_REPLY] = {
1246 .handler = icmp_discard,
1249 .handler = icmp_discard,
1251 [ICMP_ADDRESSREPLY] = {
1252 .handler = icmp_discard,
1256 static void __net_exit icmp_sk_exit(struct net *net)
1260 for_each_possible_cpu(i)
1261 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1262 free_percpu(net->ipv4.icmp_sk);
1263 net->ipv4.icmp_sk = NULL;
1266 static int __net_init icmp_sk_init(struct net *net)
1270 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1271 if (!net->ipv4.icmp_sk)
1274 for_each_possible_cpu(i) {
1277 err = inet_ctl_sock_create(&sk, PF_INET,
1278 SOCK_RAW, IPPROTO_ICMP, net);
1282 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1284 /* Enough space for 2 64K ICMP packets, including
1285 * sk_buff/skb_shared_info struct overhead.
1287 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1290 * Speedup sock_wfree()
1292 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1293 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1296 /* Control parameters for ECHO replies. */
1297 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1298 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1300 /* Control parameter - ignore bogus broadcast responses? */
1301 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1304 * Configurable global rate limit.
1306 * ratelimit defines tokens/packet consumed for dst->rate_token
1307 * bucket ratemask defines which icmp types are ratelimited by
1308 * setting it's bit position.
1311 * dest unreachable (3), source quench (4),
1312 * time exceeded (11), parameter problem (12)
1315 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1316 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1317 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1322 for_each_possible_cpu(i)
1323 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1324 free_percpu(net->ipv4.icmp_sk);
1328 static struct pernet_operations __net_initdata icmp_sk_ops = {
1329 .init = icmp_sk_init,
1330 .exit = icmp_sk_exit,
1333 int __init icmp_init(void)
1335 return register_pernet_subsys(&icmp_sk_ops);