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 #include <linux/module.h>
66 #include <linux/types.h>
67 #include <linux/jiffies.h>
68 #include <linux/kernel.h>
69 #include <linux/fcntl.h>
70 #include <linux/socket.h>
72 #include <linux/inet.h>
73 #include <linux/inetdevice.h>
74 #include <linux/netdevice.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_ipv4.h>
77 #include <linux/slab.h>
80 #include <net/route.h>
81 #include <net/protocol.h>
86 #include <linux/skbuff.h>
88 #include <linux/errno.h>
89 #include <linux/timer.h>
90 #include <linux/init.h>
91 #include <asm/system.h>
92 #include <asm/uaccess.h>
93 #include <net/checksum.h>
95 #include <net/inet_common.h>
98 * Build xmit assembly blocks
107 struct icmphdr icmph
;
111 struct ip_options_data replyopts
;
114 /* An array of errno for error messages from dest unreach. */
115 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
117 const struct icmp_err icmp_err_convert
[] = {
119 .errno
= ENETUNREACH
, /* ICMP_NET_UNREACH */
123 .errno
= EHOSTUNREACH
, /* ICMP_HOST_UNREACH */
127 .errno
= ENOPROTOOPT
/* ICMP_PROT_UNREACH */,
131 .errno
= ECONNREFUSED
, /* ICMP_PORT_UNREACH */
135 .errno
= EMSGSIZE
, /* ICMP_FRAG_NEEDED */
139 .errno
= EOPNOTSUPP
, /* ICMP_SR_FAILED */
143 .errno
= ENETUNREACH
, /* ICMP_NET_UNKNOWN */
147 .errno
= EHOSTDOWN
, /* ICMP_HOST_UNKNOWN */
151 .errno
= ENONET
, /* ICMP_HOST_ISOLATED */
155 .errno
= ENETUNREACH
, /* ICMP_NET_ANO */
159 .errno
= EHOSTUNREACH
, /* ICMP_HOST_ANO */
163 .errno
= ENETUNREACH
, /* ICMP_NET_UNR_TOS */
167 .errno
= EHOSTUNREACH
, /* ICMP_HOST_UNR_TOS */
171 .errno
= EHOSTUNREACH
, /* ICMP_PKT_FILTERED */
175 .errno
= EHOSTUNREACH
, /* ICMP_PREC_VIOLATION */
179 .errno
= EHOSTUNREACH
, /* ICMP_PREC_CUTOFF */
183 EXPORT_SYMBOL(icmp_err_convert
);
186 * ICMP control array. This specifies what to do with each ICMP.
189 struct icmp_control
{
190 void (*handler
)(struct sk_buff
*skb
);
191 short error
; /* This ICMP is classed as an error message */
194 static const struct icmp_control icmp_pointers
[NR_ICMP_TYPES
+1];
197 * The ICMP socket(s). This is the most convenient way to flow control
198 * our ICMP output as well as maintain a clean interface throughout
199 * all layers. All Socketless IP sends will soon be gone.
201 * On SMP we have one ICMP socket per-cpu.
203 static struct sock
*icmp_sk(struct net
*net
)
205 return net
->ipv4
.icmp_sk
[smp_processor_id()];
208 static inline struct sock
*icmp_xmit_lock(struct net
*net
)
216 if (unlikely(!spin_trylock(&sk
->sk_lock
.slock
))) {
217 /* This can happen if the output path signals a
218 * dst_link_failure() for an outgoing ICMP packet.
226 static inline void icmp_xmit_unlock(struct sock
*sk
)
228 spin_unlock_bh(&sk
->sk_lock
.slock
);
232 * Send an ICMP frame.
235 static inline bool icmpv4_xrlim_allow(struct net
*net
, struct rtable
*rt
,
238 struct dst_entry
*dst
= &rt
->dst
;
241 if (type
> NR_ICMP_TYPES
)
244 /* Don't limit PMTU discovery. */
245 if (type
== ICMP_DEST_UNREACH
&& code
== ICMP_FRAG_NEEDED
)
248 /* No rate limit on loopback */
249 if (dst
->dev
&& (dst
->dev
->flags
&IFF_LOOPBACK
))
252 /* Limit if icmp type is enabled in ratemask. */
253 if ((1 << type
) & net
->ipv4
.sysctl_icmp_ratemask
) {
256 rc
= inet_peer_xrlim_allow(rt
->peer
,
257 net
->ipv4
.sysctl_icmp_ratelimit
);
264 * Maintain the counters used in the SNMP statistics for outgoing ICMP
266 void icmp_out_count(struct net
*net
, unsigned char type
)
268 ICMPMSGOUT_INC_STATS(net
, type
);
269 ICMP_INC_STATS(net
, ICMP_MIB_OUTMSGS
);
273 * Checksum each fragment, and on the first include the headers and final
276 static int icmp_glue_bits(void *from
, char *to
, int offset
, int len
, int odd
,
279 struct icmp_bxm
*icmp_param
= (struct icmp_bxm
*)from
;
282 csum
= skb_copy_and_csum_bits(icmp_param
->skb
,
283 icmp_param
->offset
+ offset
,
286 skb
->csum
= csum_block_add(skb
->csum
, csum
, odd
);
287 if (icmp_pointers
[icmp_param
->data
.icmph
.type
].error
)
288 nf_ct_attach(skb
, icmp_param
->skb
);
292 static void icmp_push_reply(struct icmp_bxm
*icmp_param
,
294 struct ipcm_cookie
*ipc
, struct rtable
**rt
)
299 sk
= icmp_sk(dev_net((*rt
)->dst
.dev
));
300 if (ip_append_data(sk
, fl4
, icmp_glue_bits
, icmp_param
,
301 icmp_param
->data_len
+icmp_param
->head_len
,
302 icmp_param
->head_len
,
303 ipc
, rt
, MSG_DONTWAIT
) < 0) {
304 ICMP_INC_STATS_BH(sock_net(sk
), ICMP_MIB_OUTERRORS
);
305 ip_flush_pending_frames(sk
);
306 } else if ((skb
= skb_peek(&sk
->sk_write_queue
)) != NULL
) {
307 struct icmphdr
*icmph
= icmp_hdr(skb
);
309 struct sk_buff
*skb1
;
311 skb_queue_walk(&sk
->sk_write_queue
, skb1
) {
312 csum
= csum_add(csum
, skb1
->csum
);
314 csum
= csum_partial_copy_nocheck((void *)&icmp_param
->data
,
316 icmp_param
->head_len
, csum
);
317 icmph
->checksum
= csum_fold(csum
);
318 skb
->ip_summed
= CHECKSUM_NONE
;
319 ip_push_pending_frames(sk
, fl4
);
324 * Driving logic for building and sending ICMP messages.
327 static void icmp_reply(struct icmp_bxm
*icmp_param
, struct sk_buff
*skb
)
329 struct ipcm_cookie ipc
;
330 struct rtable
*rt
= skb_rtable(skb
);
331 struct net
*net
= dev_net(rt
->dst
.dev
);
334 struct inet_sock
*inet
;
337 if (ip_options_echo(&icmp_param
->replyopts
.opt
.opt
, skb
))
340 sk
= icmp_xmit_lock(net
);
345 icmp_param
->data
.icmph
.checksum
= 0;
347 inet
->tos
= ip_hdr(skb
)->tos
;
348 daddr
= ipc
.addr
= rt
->rt_src
;
351 if (icmp_param
->replyopts
.opt
.opt
.optlen
) {
352 ipc
.opt
= &icmp_param
->replyopts
.opt
;
353 if (ipc
.opt
->opt
.srr
)
354 daddr
= icmp_param
->replyopts
.opt
.opt
.faddr
;
356 memset(&fl4
, 0, sizeof(fl4
));
358 fl4
.saddr
= rt
->rt_spec_dst
;
359 fl4
.flowi4_tos
= RT_TOS(ip_hdr(skb
)->tos
);
360 fl4
.flowi4_proto
= IPPROTO_ICMP
;
361 security_skb_classify_flow(skb
, flowi4_to_flowi(&fl4
));
362 rt
= ip_route_output_key(net
, &fl4
);
365 if (icmpv4_xrlim_allow(net
, rt
, icmp_param
->data
.icmph
.type
,
366 icmp_param
->data
.icmph
.code
))
367 icmp_push_reply(icmp_param
, &fl4
, &ipc
, &rt
);
370 icmp_xmit_unlock(sk
);
373 static struct rtable
*icmp_route_lookup(struct net
*net
,
375 struct sk_buff
*skb_in
,
376 const struct iphdr
*iph
,
377 __be32 saddr
, u8 tos
,
379 struct icmp_bxm
*param
)
381 struct rtable
*rt
, *rt2
;
384 memset(fl4
, 0, sizeof(*fl4
));
385 fl4
->daddr
= (param
->replyopts
.opt
.opt
.srr
?
386 param
->replyopts
.opt
.opt
.faddr
: iph
->saddr
);
388 fl4
->flowi4_tos
= RT_TOS(tos
);
389 fl4
->flowi4_proto
= IPPROTO_ICMP
;
390 fl4
->fl4_icmp_type
= type
;
391 fl4
->fl4_icmp_code
= code
;
392 security_skb_classify_flow(skb_in
, flowi4_to_flowi(fl4
));
393 rt
= __ip_route_output_key(net
, fl4
);
397 /* No need to clone since we're just using its address. */
400 rt
= (struct rtable
*) xfrm_lookup(net
, &rt
->dst
,
401 flowi4_to_flowi(fl4
), NULL
, 0);
405 } else if (PTR_ERR(rt
) == -EPERM
) {
410 err
= xfrm_decode_session_reverse(skb_in
, flowi4_to_flowi(fl4
), AF_INET
);
412 goto relookup_failed
;
414 if (inet_addr_type(net
, fl4
->saddr
) == RTN_LOCAL
) {
415 rt2
= __ip_route_output_key(net
, fl4
);
419 struct flowi4 fl4_2
= {};
420 unsigned long orefdst
;
422 fl4_2
.daddr
= fl4
->saddr
;
423 rt2
= ip_route_output_key(net
, &fl4_2
);
426 goto relookup_failed
;
429 orefdst
= skb_in
->_skb_refdst
; /* save old refdst */
430 err
= ip_route_input(skb_in
, fl4
->daddr
, fl4
->saddr
,
431 RT_TOS(tos
), rt2
->dst
.dev
);
433 dst_release(&rt2
->dst
);
434 rt2
= skb_rtable(skb_in
);
435 skb_in
->_skb_refdst
= orefdst
; /* restore old refdst */
439 goto relookup_failed
;
441 rt2
= (struct rtable
*) xfrm_lookup(net
, &rt2
->dst
,
442 flowi4_to_flowi(fl4
), NULL
,
445 dst_release(&rt
->dst
);
447 } else if (PTR_ERR(rt2
) == -EPERM
) {
449 dst_release(&rt
->dst
);
453 goto relookup_failed
;
464 * Send an ICMP message in response to a situation
466 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
467 * MAY send more (we do).
468 * MUST NOT change this header information.
469 * MUST NOT reply to a multicast/broadcast IP address.
470 * MUST NOT reply to a multicast/broadcast MAC address.
471 * MUST reply to only the first fragment.
474 void icmp_send(struct sk_buff
*skb_in
, int type
, int code
, __be32 info
)
478 struct icmp_bxm icmp_param
;
479 struct rtable
*rt
= skb_rtable(skb_in
);
480 struct ipcm_cookie ipc
;
489 net
= dev_net(rt
->dst
.dev
);
492 * Find the original header. It is expected to be valid, of course.
493 * Check this, icmp_send is called from the most obscure devices
496 iph
= ip_hdr(skb_in
);
498 if ((u8
*)iph
< skb_in
->head
||
499 (skb_in
->network_header
+ sizeof(*iph
)) > skb_in
->tail
)
503 * No replies to physical multicast/broadcast
505 if (skb_in
->pkt_type
!= PACKET_HOST
)
509 * Now check at the protocol level
511 if (rt
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
))
515 * Only reply to fragment 0. We byte re-order the constant
516 * mask for efficiency.
518 if (iph
->frag_off
& htons(IP_OFFSET
))
522 * If we send an ICMP error to an ICMP error a mess would result..
524 if (icmp_pointers
[type
].error
) {
526 * We are an error, check if we are replying to an
529 if (iph
->protocol
== IPPROTO_ICMP
) {
530 u8 _inner_type
, *itp
;
532 itp
= skb_header_pointer(skb_in
,
533 skb_network_header(skb_in
) +
535 offsetof(struct icmphdr
,
544 * Assume any unknown ICMP type is an error. This
545 * isn't specified by the RFC, but think about it..
547 if (*itp
> NR_ICMP_TYPES
||
548 icmp_pointers
[*itp
].error
)
553 sk
= icmp_xmit_lock(net
);
558 * Construct source address and options.
562 if (!(rt
->rt_flags
& RTCF_LOCAL
)) {
563 struct net_device
*dev
= NULL
;
566 if (rt_is_input_route(rt
) &&
567 net
->ipv4
.sysctl_icmp_errors_use_inbound_ifaddr
)
568 dev
= dev_get_by_index_rcu(net
, rt
->rt_iif
);
571 saddr
= inet_select_addr(dev
, 0, RT_SCOPE_LINK
);
577 tos
= icmp_pointers
[type
].error
? ((iph
->tos
& IPTOS_TOS_MASK
) |
578 IPTOS_PREC_INTERNETCONTROL
) :
581 if (ip_options_echo(&icmp_param
.replyopts
.opt
.opt
, skb_in
))
586 * Prepare data for ICMP header.
589 icmp_param
.data
.icmph
.type
= type
;
590 icmp_param
.data
.icmph
.code
= code
;
591 icmp_param
.data
.icmph
.un
.gateway
= info
;
592 icmp_param
.data
.icmph
.checksum
= 0;
593 icmp_param
.skb
= skb_in
;
594 icmp_param
.offset
= skb_network_offset(skb_in
);
595 inet_sk(sk
)->tos
= tos
;
596 ipc
.addr
= iph
->saddr
;
597 ipc
.opt
= &icmp_param
.replyopts
.opt
;
600 rt
= icmp_route_lookup(net
, &fl4
, skb_in
, iph
, saddr
, tos
,
601 type
, code
, &icmp_param
);
605 if (!icmpv4_xrlim_allow(net
, rt
, type
, code
))
608 /* RFC says return as much as we can without exceeding 576 bytes. */
610 room
= dst_mtu(&rt
->dst
);
613 room
-= sizeof(struct iphdr
) + icmp_param
.replyopts
.opt
.opt
.optlen
;
614 room
-= sizeof(struct icmphdr
);
616 icmp_param
.data_len
= skb_in
->len
- icmp_param
.offset
;
617 if (icmp_param
.data_len
> room
)
618 icmp_param
.data_len
= room
;
619 icmp_param
.head_len
= sizeof(struct icmphdr
);
621 icmp_push_reply(&icmp_param
, &fl4
, &ipc
, &rt
);
625 icmp_xmit_unlock(sk
);
628 EXPORT_SYMBOL(icmp_send
);
632 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
635 static void icmp_unreach(struct sk_buff
*skb
)
637 const struct iphdr
*iph
;
638 struct icmphdr
*icmph
;
640 const struct net_protocol
*ipprot
;
644 net
= dev_net(skb_dst(skb
)->dev
);
647 * Incomplete header ?
648 * Only checks for the IP header, there should be an
649 * additional check for longer headers in upper levels.
652 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
655 icmph
= icmp_hdr(skb
);
656 iph
= (const struct iphdr
*)skb
->data
;
658 if (iph
->ihl
< 5) /* Mangled header, drop. */
661 if (icmph
->type
== ICMP_DEST_UNREACH
) {
662 switch (icmph
->code
& 15) {
663 case ICMP_NET_UNREACH
:
664 case ICMP_HOST_UNREACH
:
665 case ICMP_PROT_UNREACH
:
666 case ICMP_PORT_UNREACH
:
668 case ICMP_FRAG_NEEDED
:
669 if (ipv4_config
.no_pmtu_disc
) {
670 LIMIT_NETDEBUG(KERN_INFO
"ICMP: %pI4: fragmentation needed and DF set.\n",
673 info
= ip_rt_frag_needed(net
, iph
,
674 ntohs(icmph
->un
.frag
.mtu
),
681 LIMIT_NETDEBUG(KERN_INFO
"ICMP: %pI4: Source Route Failed.\n",
687 if (icmph
->code
> NR_ICMP_UNREACH
)
689 } else if (icmph
->type
== ICMP_PARAMETERPROB
)
690 info
= ntohl(icmph
->un
.gateway
) >> 24;
693 * Throw it at our lower layers
695 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
697 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
699 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
704 * Check the other end isn't violating RFC 1122. Some routers send
705 * bogus responses to broadcast frames. If you see this message
706 * first check your netmask matches at both ends, if it does then
707 * get the other vendor to fix their kit.
710 if (!net
->ipv4
.sysctl_icmp_ignore_bogus_error_responses
&&
711 inet_addr_type(net
, iph
->daddr
) == RTN_BROADCAST
) {
713 printk(KERN_WARNING
"%pI4 sent an invalid ICMP "
715 "error to a broadcast: %pI4 on %s\n",
717 icmph
->type
, icmph
->code
,
723 /* Checkin full IP header plus 8 bytes of protocol to
724 * avoid additional coding at protocol handlers.
726 if (!pskb_may_pull(skb
, iph
->ihl
* 4 + 8))
729 iph
= (const struct iphdr
*)skb
->data
;
730 protocol
= iph
->protocol
;
733 * Deliver ICMP message to raw sockets. Pretty useless feature?
735 raw_icmp_error(skb
, protocol
, info
);
737 hash
= protocol
& (MAX_INET_PROTOS
- 1);
739 ipprot
= rcu_dereference(inet_protos
[hash
]);
740 if (ipprot
&& ipprot
->err_handler
)
741 ipprot
->err_handler(skb
, info
);
747 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
753 * Handle ICMP_REDIRECT.
756 static void icmp_redirect(struct sk_buff
*skb
)
758 const struct iphdr
*iph
;
760 if (skb
->len
< sizeof(struct iphdr
))
764 * Get the copied header of the packet that caused the redirect
766 if (!pskb_may_pull(skb
, sizeof(struct iphdr
)))
769 iph
= (const struct iphdr
*)skb
->data
;
771 switch (icmp_hdr(skb
)->code
& 7) {
773 case ICMP_REDIR_NETTOS
:
775 * As per RFC recommendations now handle it as a host redirect.
777 case ICMP_REDIR_HOST
:
778 case ICMP_REDIR_HOSTTOS
:
779 ip_rt_redirect(ip_hdr(skb
)->saddr
, iph
->daddr
,
780 icmp_hdr(skb
)->un
.gateway
,
781 iph
->saddr
, skb
->dev
);
787 ICMP_INC_STATS_BH(dev_net(skb
->dev
), ICMP_MIB_INERRORS
);
792 * Handle ICMP_ECHO ("ping") requests.
794 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
796 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
797 * included in the reply.
798 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
799 * echo requests, MUST have default=NOT.
800 * See also WRT handling of options once they are done and working.
803 static void icmp_echo(struct sk_buff
*skb
)
807 net
= dev_net(skb_dst(skb
)->dev
);
808 if (!net
->ipv4
.sysctl_icmp_echo_ignore_all
) {
809 struct icmp_bxm icmp_param
;
811 icmp_param
.data
.icmph
= *icmp_hdr(skb
);
812 icmp_param
.data
.icmph
.type
= ICMP_ECHOREPLY
;
813 icmp_param
.skb
= skb
;
814 icmp_param
.offset
= 0;
815 icmp_param
.data_len
= skb
->len
;
816 icmp_param
.head_len
= sizeof(struct icmphdr
);
817 icmp_reply(&icmp_param
, skb
);
822 * Handle ICMP Timestamp requests.
823 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
824 * SHOULD be in the kernel for minimum random latency.
825 * MUST be accurate to a few minutes.
826 * MUST be updated at least at 15Hz.
828 static void icmp_timestamp(struct sk_buff
*skb
)
831 struct icmp_bxm icmp_param
;
839 * Fill in the current time as ms since midnight UT:
842 icmp_param
.data
.times
[1] = htonl((tv
.tv_sec
% 86400) * MSEC_PER_SEC
+
843 tv
.tv_nsec
/ NSEC_PER_MSEC
);
844 icmp_param
.data
.times
[2] = icmp_param
.data
.times
[1];
845 if (skb_copy_bits(skb
, 0, &icmp_param
.data
.times
[0], 4))
847 icmp_param
.data
.icmph
= *icmp_hdr(skb
);
848 icmp_param
.data
.icmph
.type
= ICMP_TIMESTAMPREPLY
;
849 icmp_param
.data
.icmph
.code
= 0;
850 icmp_param
.skb
= skb
;
851 icmp_param
.offset
= 0;
852 icmp_param
.data_len
= 0;
853 icmp_param
.head_len
= sizeof(struct icmphdr
) + 12;
854 icmp_reply(&icmp_param
, skb
);
858 ICMP_INC_STATS_BH(dev_net(skb_dst(skb
)->dev
), ICMP_MIB_INERRORS
);
864 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
866 * RFC1122 (3.2.2.9). A host MUST only send replies to
867 * ADDRESS_MASK requests if it's been configured as an address mask
868 * agent. Receiving a request doesn't constitute implicit permission to
869 * act as one. Of course, implementing this correctly requires (SHOULD)
870 * a way to turn the functionality on and off. Another one for sysctl(),
873 * RFC1812 (4.3.3.9). A router MUST implement it.
874 * A router SHOULD have switch turning it on/off.
875 * This switch MUST be ON by default.
877 * Gratuitous replies, zero-source replies are not implemented,
878 * that complies with RFC. DO NOT implement them!!! All the idea
879 * of broadcast addrmask replies as specified in RFC950 is broken.
880 * The problem is that it is not uncommon to have several prefixes
881 * on one physical interface. Moreover, addrmask agent can even be
882 * not aware of existing another prefixes.
883 * If source is zero, addrmask agent cannot choose correct prefix.
884 * Gratuitous mask announcements suffer from the same problem.
885 * RFC1812 explains it, but still allows to use ADDRMASK,
886 * that is pretty silly. --ANK
888 * All these rules are so bizarre, that I removed kernel addrmask
889 * support at all. It is wrong, it is obsolete, nobody uses it in
892 * Furthermore you can do it with a usermode address agent program
896 static void icmp_address(struct sk_buff
*skb
)
900 printk(KERN_DEBUG
"a guy asks for address mask. Who is it?\n");
905 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
906 * loudly if an inconsistency is found.
907 * called with rcu_read_lock()
910 static void icmp_address_reply(struct sk_buff
*skb
)
912 struct rtable
*rt
= skb_rtable(skb
);
913 struct net_device
*dev
= skb
->dev
;
914 struct in_device
*in_dev
;
915 struct in_ifaddr
*ifa
;
917 if (skb
->len
< 4 || !(rt
->rt_flags
&RTCF_DIRECTSRC
))
920 in_dev
= __in_dev_get_rcu(dev
);
924 if (in_dev
->ifa_list
&&
925 IN_DEV_LOG_MARTIANS(in_dev
) &&
926 IN_DEV_FORWARD(in_dev
)) {
929 mp
= skb_header_pointer(skb
, 0, sizeof(_mask
), &_mask
);
931 for (ifa
= in_dev
->ifa_list
; ifa
; ifa
= ifa
->ifa_next
) {
932 if (*mp
== ifa
->ifa_mask
&&
933 inet_ifa_match(rt
->rt_src
, ifa
))
936 if (!ifa
&& net_ratelimit()) {
937 printk(KERN_INFO
"Wrong address mask %pI4 from %s/%pI4\n",
938 mp
, dev
->name
, &rt
->rt_src
);
943 static void icmp_discard(struct sk_buff
*skb
)
948 * Deal with incoming ICMP packets.
950 int icmp_rcv(struct sk_buff
*skb
)
952 struct icmphdr
*icmph
;
953 struct rtable
*rt
= skb_rtable(skb
);
954 struct net
*net
= dev_net(rt
->dst
.dev
);
956 if (!xfrm4_policy_check(NULL
, XFRM_POLICY_IN
, skb
)) {
957 struct sec_path
*sp
= skb_sec_path(skb
);
960 if (!(sp
&& sp
->xvec
[sp
->len
- 1]->props
.flags
&
964 if (!pskb_may_pull(skb
, sizeof(*icmph
) + sizeof(struct iphdr
)))
967 nh
= skb_network_offset(skb
);
968 skb_set_network_header(skb
, sizeof(*icmph
));
970 if (!xfrm4_policy_check_reverse(NULL
, XFRM_POLICY_IN
, skb
))
973 skb_set_network_header(skb
, nh
);
976 ICMP_INC_STATS_BH(net
, ICMP_MIB_INMSGS
);
978 switch (skb
->ip_summed
) {
979 case CHECKSUM_COMPLETE
:
980 if (!csum_fold(skb
->csum
))
985 if (__skb_checksum_complete(skb
))
989 if (!pskb_pull(skb
, sizeof(*icmph
)))
992 icmph
= icmp_hdr(skb
);
994 ICMPMSGIN_INC_STATS_BH(net
, icmph
->type
);
996 * 18 is the highest 'known' ICMP type. Anything else is a mystery
998 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1001 if (icmph
->type
> NR_ICMP_TYPES
)
1006 * Parse the ICMP message
1009 if (rt
->rt_flags
& (RTCF_BROADCAST
| RTCF_MULTICAST
)) {
1011 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1012 * silently ignored (we let user decide with a sysctl).
1013 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1014 * discarded if to broadcast/multicast.
1016 if ((icmph
->type
== ICMP_ECHO
||
1017 icmph
->type
== ICMP_TIMESTAMP
) &&
1018 net
->ipv4
.sysctl_icmp_echo_ignore_broadcasts
) {
1021 if (icmph
->type
!= ICMP_ECHO
&&
1022 icmph
->type
!= ICMP_TIMESTAMP
&&
1023 icmph
->type
!= ICMP_ADDRESS
&&
1024 icmph
->type
!= ICMP_ADDRESSREPLY
) {
1029 icmp_pointers
[icmph
->type
].handler(skb
);
1035 ICMP_INC_STATS_BH(net
, ICMP_MIB_INERRORS
);
1040 * This table is the definition of how we handle ICMP.
1042 static const struct icmp_control icmp_pointers
[NR_ICMP_TYPES
+ 1] = {
1043 [ICMP_ECHOREPLY
] = {
1044 .handler
= icmp_discard
,
1047 .handler
= icmp_discard
,
1051 .handler
= icmp_discard
,
1054 [ICMP_DEST_UNREACH
] = {
1055 .handler
= icmp_unreach
,
1058 [ICMP_SOURCE_QUENCH
] = {
1059 .handler
= icmp_unreach
,
1063 .handler
= icmp_redirect
,
1067 .handler
= icmp_discard
,
1071 .handler
= icmp_discard
,
1075 .handler
= icmp_echo
,
1078 .handler
= icmp_discard
,
1082 .handler
= icmp_discard
,
1085 [ICMP_TIME_EXCEEDED
] = {
1086 .handler
= icmp_unreach
,
1089 [ICMP_PARAMETERPROB
] = {
1090 .handler
= icmp_unreach
,
1093 [ICMP_TIMESTAMP
] = {
1094 .handler
= icmp_timestamp
,
1096 [ICMP_TIMESTAMPREPLY
] = {
1097 .handler
= icmp_discard
,
1099 [ICMP_INFO_REQUEST
] = {
1100 .handler
= icmp_discard
,
1102 [ICMP_INFO_REPLY
] = {
1103 .handler
= icmp_discard
,
1106 .handler
= icmp_address
,
1108 [ICMP_ADDRESSREPLY
] = {
1109 .handler
= icmp_address_reply
,
1113 static void __net_exit
icmp_sk_exit(struct net
*net
)
1117 for_each_possible_cpu(i
)
1118 inet_ctl_sock_destroy(net
->ipv4
.icmp_sk
[i
]);
1119 kfree(net
->ipv4
.icmp_sk
);
1120 net
->ipv4
.icmp_sk
= NULL
;
1123 static int __net_init
icmp_sk_init(struct net
*net
)
1128 kzalloc(nr_cpu_ids
* sizeof(struct sock
*), GFP_KERNEL
);
1129 if (net
->ipv4
.icmp_sk
== NULL
)
1132 for_each_possible_cpu(i
) {
1135 err
= inet_ctl_sock_create(&sk
, PF_INET
,
1136 SOCK_RAW
, IPPROTO_ICMP
, net
);
1140 net
->ipv4
.icmp_sk
[i
] = sk
;
1142 /* Enough space for 2 64K ICMP packets, including
1143 * sk_buff struct overhead.
1146 (2 * ((64 * 1024) + sizeof(struct sk_buff
)));
1149 * Speedup sock_wfree()
1151 sock_set_flag(sk
, SOCK_USE_WRITE_QUEUE
);
1152 inet_sk(sk
)->pmtudisc
= IP_PMTUDISC_DONT
;
1155 /* Control parameters for ECHO replies. */
1156 net
->ipv4
.sysctl_icmp_echo_ignore_all
= 0;
1157 net
->ipv4
.sysctl_icmp_echo_ignore_broadcasts
= 1;
1159 /* Control parameter - ignore bogus broadcast responses? */
1160 net
->ipv4
.sysctl_icmp_ignore_bogus_error_responses
= 1;
1163 * Configurable global rate limit.
1165 * ratelimit defines tokens/packet consumed for dst->rate_token
1166 * bucket ratemask defines which icmp types are ratelimited by
1167 * setting it's bit position.
1170 * dest unreachable (3), source quench (4),
1171 * time exceeded (11), parameter problem (12)
1174 net
->ipv4
.sysctl_icmp_ratelimit
= 1 * HZ
;
1175 net
->ipv4
.sysctl_icmp_ratemask
= 0x1818;
1176 net
->ipv4
.sysctl_icmp_errors_use_inbound_ifaddr
= 0;
1181 for_each_possible_cpu(i
)
1182 inet_ctl_sock_destroy(net
->ipv4
.icmp_sk
[i
]);
1183 kfree(net
->ipv4
.icmp_sk
);
1187 static struct pernet_operations __net_initdata icmp_sk_ops
= {
1188 .init
= icmp_sk_init
,
1189 .exit
= icmp_sk_exit
,
1192 int __init
icmp_init(void)
1194 return register_pernet_subsys(&icmp_sk_ops
);