2 * NET3: Implementation of the ICMP protocol layer.
4 * Alan Cox, <alan@redhat.com>
6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Some of the function names and the icmp unreach table for this
14 * module were derived from [icmp.c 1.0.11 06/02/93] by
15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16 * Other than that this module is a complete rewrite.
19 * Clemens Fruhwirth : introduce global icmp rate limiting
20 * with icmp type masking ability instead
21 * of broken per type icmp timeouts.
22 * Mike Shaver : RFC1122 checks.
23 * Alan Cox : Multicast ping reply as self.
24 * Alan Cox : Fix atomicity lockup in ip_build_xmit
26 * Alan Cox : Added 216,128 byte paths to the MTU
28 * Martin Mares : RFC1812 checks.
29 * Martin Mares : Can be configured to follow redirects
30 * if acting as a router _without_ a
31 * routing protocol (RFC 1812).
32 * Martin Mares : Echo requests may be configured to
33 * be ignored (RFC 1812).
34 * Martin Mares : Limitation of ICMP error message
35 * transmit rate (RFC 1812).
36 * Martin Mares : TOS and Precedence set correctly
38 * Martin Mares : Now copying as much data from the
39 * original packet as we can without
40 * exceeding 576 bytes (RFC 1812).
41 * Willy Konynenberg : Transparent proxying support.
42 * Keith Owens : RFC1191 correction for 4.2BSD based
44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
46 * Andi Kleen : Check all packet lengths properly
47 * and moved all kfree_skb() up to
49 * Andi Kleen : Move the rate limit bookkeeping
50 * into the dest entry and use a token
51 * bucket filter (thanks to ANK). Make
52 * the rates sysctl configurable.
53 * Yu Tianli : Fixed two ugly bugs in icmp_send
54 * - IP option length was accounted wrongly
55 * - ICMP header length was not accounted
57 * Tristan Greaves : Added sysctl option to ignore bogus
58 * broadcast responses from broken routers.
62 * - Should use skb_pull() instead of all the manual checking.
63 * This would also greatly simply some upper layer error handlers. --AK
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>
81 #include <net/route.h>
82 #include <net/protocol.h>
87 #include <linux/skbuff.h>
89 #include <linux/errno.h>
90 #include <linux/timer.h>
91 #include <linux/init.h>
92 #include <asm/system.h>
93 #include <asm/uaccess.h>
94 #include <net/checksum.h>
98 * Build xmit assembly blocks
107 struct icmphdr icmph;
111 struct ip_options replyopts;
112 unsigned char optbuf[40];
118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
119 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
124 struct icmp_err icmp_err_convert[] = {
126 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
130 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
134 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
138 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
142 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
146 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
150 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
154 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
158 .errno = ENONET, /* ICMP_HOST_ISOLATED */
162 .errno = ENETUNREACH, /* ICMP_NET_ANO */
166 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
170 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
174 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
178 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
182 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
186 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
191 /* Control parameters for ECHO replies. */
192 int sysctl_icmp_echo_ignore_all __read_mostly;
193 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
195 /* Control parameter - ignore bogus broadcast responses? */
196 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
199 * Configurable global rate limit.
201 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
202 * ratemask defines which icmp types are ratelimited by setting
206 * dest unreachable (3), source quench (4),
207 * time exceeded (11), parameter problem (12)
210 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
211 int sysctl_icmp_ratemask __read_mostly = 0x1818;
212 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
215 * ICMP control array. This specifies what to do with each ICMP.
218 struct icmp_control {
219 void (*handler)(struct sk_buff *skb);
220 short error; /* This ICMP is classed as an error message */
223 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
226 * The ICMP socket(s). This is the most convenient way to flow control
227 * our ICMP output as well as maintain a clean interface throughout
228 * all layers. All Socketless IP sends will soon be gone.
230 * On SMP we have one ICMP socket per-cpu.
232 static struct sock *icmp_sk(struct net *net)
234 return net->ipv4.icmp_sk[smp_processor_id()];
237 static inline int icmp_xmit_lock(struct sock *sk)
241 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
242 /* This can happen if the output path signals a
243 * dst_link_failure() for an outgoing ICMP packet.
251 static inline void icmp_xmit_unlock(struct sock *sk)
253 spin_unlock_bh(&sk->sk_lock.slock);
257 * Send an ICMP frame.
261 * Check transmit rate limitation for given message.
262 * The rate information is held in the destination cache now.
263 * This function is generic and could be used for other purposes
264 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
266 * Note that the same dst_entry fields are modified by functions in
267 * route.c too, but these work for packet destinations while xrlim_allow
268 * works for icmp destinations. This means the rate limiting information
269 * for one "ip object" is shared - and these ICMPs are twice limited:
270 * by source and by destination.
272 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
273 * SHOULD allow setting of rate limits
275 * Shared between ICMPv4 and ICMPv6.
277 #define XRLIM_BURST_FACTOR 6
278 int xrlim_allow(struct dst_entry *dst, int timeout)
280 unsigned long now, token = dst->rate_tokens;
284 token += now - dst->rate_last;
285 dst->rate_last = now;
286 if (token > XRLIM_BURST_FACTOR * timeout)
287 token = XRLIM_BURST_FACTOR * timeout;
288 if (token >= timeout) {
292 dst->rate_tokens = token;
296 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
298 struct dst_entry *dst = &rt->u.dst;
301 if (type > NR_ICMP_TYPES)
304 /* Don't limit PMTU discovery. */
305 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
308 /* No rate limit on loopback */
309 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
312 /* Limit if icmp type is enabled in ratemask. */
313 if ((1 << type) & sysctl_icmp_ratemask)
314 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
320 * Maintain the counters used in the SNMP statistics for outgoing ICMP
322 void icmp_out_count(unsigned char type)
324 ICMPMSGOUT_INC_STATS(type);
325 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
329 * Checksum each fragment, and on the first include the headers and final
332 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
335 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
338 csum = skb_copy_and_csum_bits(icmp_param->skb,
339 icmp_param->offset + offset,
342 skb->csum = csum_block_add(skb->csum, csum, odd);
343 if (icmp_pointers[icmp_param->data.icmph.type].error)
344 nf_ct_attach(skb, icmp_param->skb);
348 static void icmp_push_reply(struct icmp_bxm *icmp_param,
349 struct ipcm_cookie *ipc, struct rtable *rt)
354 sk = icmp_sk(rt->u.dst.dev->nd_net);
355 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
356 icmp_param->data_len+icmp_param->head_len,
357 icmp_param->head_len,
358 ipc, rt, MSG_DONTWAIT) < 0)
359 ip_flush_pending_frames(sk);
360 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
361 struct icmphdr *icmph = icmp_hdr(skb);
363 struct sk_buff *skb1;
365 skb_queue_walk(&sk->sk_write_queue, skb1) {
366 csum = csum_add(csum, skb1->csum);
368 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
370 icmp_param->head_len, csum);
371 icmph->checksum = csum_fold(csum);
372 skb->ip_summed = CHECKSUM_NONE;
373 ip_push_pending_frames(sk);
378 * Driving logic for building and sending ICMP messages.
381 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
383 struct ipcm_cookie ipc;
384 struct rtable *rt = skb->rtable;
385 struct net *net = rt->u.dst.dev->nd_net;
386 struct sock *sk = icmp_sk(net);
387 struct inet_sock *inet = inet_sk(sk);
390 if (ip_options_echo(&icmp_param->replyopts, skb))
393 if (icmp_xmit_lock(sk))
396 icmp_param->data.icmph.checksum = 0;
398 inet->tos = ip_hdr(skb)->tos;
399 daddr = ipc.addr = rt->rt_src;
401 if (icmp_param->replyopts.optlen) {
402 ipc.opt = &icmp_param->replyopts;
404 daddr = icmp_param->replyopts.faddr;
407 struct flowi fl = { .nl_u = { .ip4_u =
409 .saddr = rt->rt_spec_dst,
410 .tos = RT_TOS(ip_hdr(skb)->tos) } },
411 .proto = IPPROTO_ICMP };
412 security_skb_classify_flow(skb, &fl);
413 if (ip_route_output_key(net, &rt, &fl))
416 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
417 icmp_param->data.icmph.code))
418 icmp_push_reply(icmp_param, &ipc, rt);
421 icmp_xmit_unlock(sk);
426 * Send an ICMP message in response to a situation
428 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
429 * MAY send more (we do).
430 * MUST NOT change this header information.
431 * MUST NOT reply to a multicast/broadcast IP address.
432 * MUST NOT reply to a multicast/broadcast MAC address.
433 * MUST reply to only the first fragment.
436 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
440 struct icmp_bxm icmp_param;
441 struct rtable *rt = skb_in->rtable;
442 struct ipcm_cookie ipc;
450 net = rt->u.dst.dev->nd_net;
454 * Find the original header. It is expected to be valid, of course.
455 * Check this, icmp_send is called from the most obscure devices
458 iph = ip_hdr(skb_in);
460 if ((u8 *)iph < skb_in->head ||
461 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
465 * No replies to physical multicast/broadcast
467 if (skb_in->pkt_type != PACKET_HOST)
471 * Now check at the protocol level
473 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
477 * Only reply to fragment 0. We byte re-order the constant
478 * mask for efficiency.
480 if (iph->frag_off & htons(IP_OFFSET))
484 * If we send an ICMP error to an ICMP error a mess would result..
486 if (icmp_pointers[type].error) {
488 * We are an error, check if we are replying to an
491 if (iph->protocol == IPPROTO_ICMP) {
492 u8 _inner_type, *itp;
494 itp = skb_header_pointer(skb_in,
495 skb_network_header(skb_in) +
497 offsetof(struct icmphdr,
506 * Assume any unknown ICMP type is an error. This
507 * isn't specified by the RFC, but think about it..
509 if (*itp > NR_ICMP_TYPES ||
510 icmp_pointers[*itp].error)
515 if (icmp_xmit_lock(sk))
519 * Construct source address and options.
523 if (!(rt->rt_flags & RTCF_LOCAL)) {
524 struct net_device *dev = NULL;
526 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
527 dev = dev_get_by_index(net, rt->fl.iif);
530 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
536 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
537 IPTOS_PREC_INTERNETCONTROL) :
540 if (ip_options_echo(&icmp_param.replyopts, skb_in))
545 * Prepare data for ICMP header.
548 icmp_param.data.icmph.type = type;
549 icmp_param.data.icmph.code = code;
550 icmp_param.data.icmph.un.gateway = info;
551 icmp_param.data.icmph.checksum = 0;
552 icmp_param.skb = skb_in;
553 icmp_param.offset = skb_network_offset(skb_in);
554 inet_sk(sk)->tos = tos;
555 ipc.addr = iph->saddr;
556 ipc.opt = &icmp_param.replyopts;
562 .daddr = icmp_param.replyopts.srr ?
563 icmp_param.replyopts.faddr :
569 .proto = IPPROTO_ICMP,
580 security_skb_classify_flow(skb_in, &fl);
581 if (__ip_route_output_key(net, &rt, &fl))
584 /* No need to clone since we're just using its address. */
587 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
600 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
603 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
604 err = __ip_route_output_key(net, &rt2, &fl);
606 struct flowi fl2 = {};
607 struct dst_entry *odst;
609 fl2.fl4_dst = fl.fl4_src;
610 if (ip_route_output_key(net, &rt2, &fl2))
615 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
616 RT_TOS(tos), rt2->u.dst.dev);
618 dst_release(&rt2->u.dst);
619 rt2 = skb_in->rtable;
626 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
628 if (err == -ENOENT) {
634 dst_release(&rt->u.dst);
642 if (!icmpv4_xrlim_allow(rt, type, code))
645 /* RFC says return as much as we can without exceeding 576 bytes. */
647 room = dst_mtu(&rt->u.dst);
650 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
651 room -= sizeof(struct icmphdr);
653 icmp_param.data_len = skb_in->len - icmp_param.offset;
654 if (icmp_param.data_len > room)
655 icmp_param.data_len = room;
656 icmp_param.head_len = sizeof(struct icmphdr);
658 icmp_push_reply(&icmp_param, &ipc, rt);
662 icmp_xmit_unlock(sk);
668 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
671 static void icmp_unreach(struct sk_buff *skb)
674 struct icmphdr *icmph;
676 struct net_protocol *ipprot;
680 net = skb->dst->dev->nd_net;
683 * Incomplete header ?
684 * Only checks for the IP header, there should be an
685 * additional check for longer headers in upper levels.
688 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
691 icmph = icmp_hdr(skb);
692 iph = (struct iphdr *)skb->data;
694 if (iph->ihl < 5) /* Mangled header, drop. */
697 if (icmph->type == ICMP_DEST_UNREACH) {
698 switch (icmph->code & 15) {
699 case ICMP_NET_UNREACH:
700 case ICMP_HOST_UNREACH:
701 case ICMP_PROT_UNREACH:
702 case ICMP_PORT_UNREACH:
704 case ICMP_FRAG_NEEDED:
705 if (ipv4_config.no_pmtu_disc) {
706 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
707 "fragmentation needed "
709 NIPQUAD(iph->daddr));
711 info = ip_rt_frag_needed(net, iph,
712 ntohs(icmph->un.frag.mtu));
718 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
720 NIPQUAD(iph->daddr));
725 if (icmph->code > NR_ICMP_UNREACH)
727 } else if (icmph->type == ICMP_PARAMETERPROB)
728 info = ntohl(icmph->un.gateway) >> 24;
731 * Throw it at our lower layers
733 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
735 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
737 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
742 * Check the other end isnt violating RFC 1122. Some routers send
743 * bogus responses to broadcast frames. If you see this message
744 * first check your netmask matches at both ends, if it does then
745 * get the other vendor to fix their kit.
748 if (!sysctl_icmp_ignore_bogus_error_responses &&
749 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
751 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
753 "error to a broadcast: %u.%u.%u.%u on %s\n",
754 NIPQUAD(ip_hdr(skb)->saddr),
755 icmph->type, icmph->code,
761 /* Checkin full IP header plus 8 bytes of protocol to
762 * avoid additional coding at protocol handlers.
764 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
767 iph = (struct iphdr *)skb->data;
768 protocol = iph->protocol;
771 * Deliver ICMP message to raw sockets. Pretty useless feature?
773 raw_icmp_error(skb, protocol, info);
775 hash = protocol & (MAX_INET_PROTOS - 1);
777 ipprot = rcu_dereference(inet_protos[hash]);
778 if (ipprot && ipprot->err_handler)
779 ipprot->err_handler(skb, info);
785 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
791 * Handle ICMP_REDIRECT.
794 static void icmp_redirect(struct sk_buff *skb)
798 if (skb->len < sizeof(struct iphdr))
802 * Get the copied header of the packet that caused the redirect
804 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
807 iph = (struct iphdr *)skb->data;
809 switch (icmp_hdr(skb)->code & 7) {
811 case ICMP_REDIR_NETTOS:
813 * As per RFC recommendations now handle it as a host redirect.
815 case ICMP_REDIR_HOST:
816 case ICMP_REDIR_HOSTTOS:
817 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
818 icmp_hdr(skb)->un.gateway,
819 iph->saddr, skb->dev);
825 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
830 * Handle ICMP_ECHO ("ping") requests.
832 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
834 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
835 * included in the reply.
836 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
837 * echo requests, MUST have default=NOT.
838 * See also WRT handling of options once they are done and working.
841 static void icmp_echo(struct sk_buff *skb)
843 if (!sysctl_icmp_echo_ignore_all) {
844 struct icmp_bxm icmp_param;
846 icmp_param.data.icmph = *icmp_hdr(skb);
847 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
848 icmp_param.skb = skb;
849 icmp_param.offset = 0;
850 icmp_param.data_len = skb->len;
851 icmp_param.head_len = sizeof(struct icmphdr);
852 icmp_reply(&icmp_param, skb);
857 * Handle ICMP Timestamp requests.
858 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
859 * SHOULD be in the kernel for minimum random latency.
860 * MUST be accurate to a few minutes.
861 * MUST be updated at least at 15Hz.
863 static void icmp_timestamp(struct sk_buff *skb)
866 struct icmp_bxm icmp_param;
874 * Fill in the current time as ms since midnight UT:
876 do_gettimeofday(&tv);
877 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
879 icmp_param.data.times[2] = icmp_param.data.times[1];
880 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
882 icmp_param.data.icmph = *icmp_hdr(skb);
883 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
884 icmp_param.data.icmph.code = 0;
885 icmp_param.skb = skb;
886 icmp_param.offset = 0;
887 icmp_param.data_len = 0;
888 icmp_param.head_len = sizeof(struct icmphdr) + 12;
889 icmp_reply(&icmp_param, skb);
893 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
899 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
901 * RFC1122 (3.2.2.9). A host MUST only send replies to
902 * ADDRESS_MASK requests if it's been configured as an address mask
903 * agent. Receiving a request doesn't constitute implicit permission to
904 * act as one. Of course, implementing this correctly requires (SHOULD)
905 * a way to turn the functionality on and off. Another one for sysctl(),
908 * RFC1812 (4.3.3.9). A router MUST implement it.
909 * A router SHOULD have switch turning it on/off.
910 * This switch MUST be ON by default.
912 * Gratuitous replies, zero-source replies are not implemented,
913 * that complies with RFC. DO NOT implement them!!! All the idea
914 * of broadcast addrmask replies as specified in RFC950 is broken.
915 * The problem is that it is not uncommon to have several prefixes
916 * on one physical interface. Moreover, addrmask agent can even be
917 * not aware of existing another prefixes.
918 * If source is zero, addrmask agent cannot choose correct prefix.
919 * Gratuitous mask announcements suffer from the same problem.
920 * RFC1812 explains it, but still allows to use ADDRMASK,
921 * that is pretty silly. --ANK
923 * All these rules are so bizarre, that I removed kernel addrmask
924 * support at all. It is wrong, it is obsolete, nobody uses it in
927 * Furthermore you can do it with a usermode address agent program
931 static void icmp_address(struct sk_buff *skb)
935 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
940 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
941 * loudly if an inconsistency is found.
944 static void icmp_address_reply(struct sk_buff *skb)
946 struct rtable *rt = skb->rtable;
947 struct net_device *dev = skb->dev;
948 struct in_device *in_dev;
949 struct in_ifaddr *ifa;
951 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
954 in_dev = in_dev_get(dev);
958 if (in_dev->ifa_list &&
959 IN_DEV_LOG_MARTIANS(in_dev) &&
960 IN_DEV_FORWARD(in_dev)) {
963 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
965 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
966 if (*mp == ifa->ifa_mask &&
967 inet_ifa_match(rt->rt_src, ifa))
970 if (!ifa && net_ratelimit()) {
971 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
973 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
981 static void icmp_discard(struct sk_buff *skb)
986 * Deal with incoming ICMP packets.
988 int icmp_rcv(struct sk_buff *skb)
990 struct icmphdr *icmph;
991 struct rtable *rt = skb->rtable;
993 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
996 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
1000 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1003 nh = skb_network_offset(skb);
1004 skb_set_network_header(skb, sizeof(*icmph));
1006 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1009 skb_set_network_header(skb, nh);
1012 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1014 switch (skb->ip_summed) {
1015 case CHECKSUM_COMPLETE:
1016 if (!csum_fold(skb->csum))
1021 if (__skb_checksum_complete(skb))
1025 if (!pskb_pull(skb, sizeof(*icmph)))
1028 icmph = icmp_hdr(skb);
1030 ICMPMSGIN_INC_STATS_BH(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 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 icmp_pointers[icmph->type].handler(skb);
1071 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1076 * This table is the definition of how we handle ICMP.
1078 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1079 [ICMP_ECHOREPLY] = {
1080 .handler = icmp_discard,
1083 .handler = icmp_discard,
1087 .handler = icmp_discard,
1090 [ICMP_DEST_UNREACH] = {
1091 .handler = icmp_unreach,
1094 [ICMP_SOURCE_QUENCH] = {
1095 .handler = icmp_unreach,
1099 .handler = icmp_redirect,
1103 .handler = icmp_discard,
1107 .handler = icmp_discard,
1111 .handler = icmp_echo,
1114 .handler = icmp_discard,
1118 .handler = icmp_discard,
1121 [ICMP_TIME_EXCEEDED] = {
1122 .handler = icmp_unreach,
1125 [ICMP_PARAMETERPROB] = {
1126 .handler = icmp_unreach,
1129 [ICMP_TIMESTAMP] = {
1130 .handler = icmp_timestamp,
1132 [ICMP_TIMESTAMPREPLY] = {
1133 .handler = icmp_discard,
1135 [ICMP_INFO_REQUEST] = {
1136 .handler = icmp_discard,
1138 [ICMP_INFO_REPLY] = {
1139 .handler = icmp_discard,
1142 .handler = icmp_address,
1144 [ICMP_ADDRESSREPLY] = {
1145 .handler = icmp_address_reply,
1149 static void __net_exit icmp_sk_exit(struct net *net)
1153 for_each_possible_cpu(i)
1154 sk_release_kernel(net->ipv4.icmp_sk[i]);
1155 kfree(net->ipv4.icmp_sk);
1156 net->ipv4.icmp_sk = NULL;
1159 int __net_init icmp_sk_init(struct net *net)
1164 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1165 if (net->ipv4.icmp_sk == NULL)
1168 for_each_possible_cpu(i) {
1170 struct socket *sock;
1171 struct inet_sock *inet;
1173 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, &sock);
1177 net->ipv4.icmp_sk[i] = sk = sock->sk;
1178 sk_change_net(sk, net);
1180 sk->sk_allocation = GFP_ATOMIC;
1182 /* Enough space for 2 64K ICMP packets, including
1183 * sk_buff struct overhead.
1186 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1190 inet->pmtudisc = IP_PMTUDISC_DONT;
1192 /* Unhash it so that IP input processing does not even
1193 * see it, we do not wish this socket to see incoming
1196 sk->sk_prot->unhash(sk);
1201 for_each_possible_cpu(i)
1202 sk_release_kernel(net->ipv4.icmp_sk[i]);
1203 kfree(net->ipv4.icmp_sk);
1207 static struct pernet_operations __net_initdata icmp_sk_ops = {
1208 .init = icmp_sk_init,
1209 .exit = icmp_sk_exit,
1212 int __init icmp_init(void)
1214 return register_pernet_device(&icmp_sk_ops);
1217 EXPORT_SYMBOL(icmp_err_convert);
1218 EXPORT_SYMBOL(icmp_send);
1219 EXPORT_SYMBOL(icmp_statistics);
1220 EXPORT_SYMBOL(xrlim_allow);