Loading...
1/*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
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.
10 *
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.
15 *
16 * Fixes:
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
23 * call.
24 * Alan Cox : Added 216,128 byte paths to the MTU
25 * code.
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
35 * (RFC 1812).
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
41 * path MTU bug.
42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
43 * valid (RFC 1812).
44 * Andi Kleen : Check all packet lengths properly
45 * and moved all kfree_skb() up to
46 * icmp_rcv.
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
54 * at all.
55 * Tristan Greaves : Added sysctl option to ignore bogus
56 * broadcast responses from broken routers.
57 *
58 * To Fix:
59 *
60 * - Should use skb_pull() instead of all the manual checking.
61 * This would also greatly simply some upper layer error handlers. --AK
62 *
63 */
64
65#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
66
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>
73#include <linux/in.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>
80#include <net/snmp.h>
81#include <net/ip.h>
82#include <net/route.h>
83#include <net/protocol.h>
84#include <net/icmp.h>
85#include <net/tcp.h>
86#include <net/udp.h>
87#include <net/raw.h>
88#include <net/ping.h>
89#include <linux/skbuff.h>
90#include <net/sock.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>
96#include <net/xfrm.h>
97#include <net/inet_common.h>
98#include <net/ip_fib.h>
99#include <net/l3mdev.h>
100
101/*
102 * Build xmit assembly blocks
103 */
104
105struct icmp_bxm {
106 struct sk_buff *skb;
107 int offset;
108 int data_len;
109
110 struct {
111 struct icmphdr icmph;
112 __be32 times[3];
113 } data;
114 int head_len;
115 struct ip_options_data replyopts;
116};
117
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'. */
120
121const struct icmp_err icmp_err_convert[] = {
122 {
123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
124 .fatal = 0,
125 },
126 {
127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
128 .fatal = 0,
129 },
130 {
131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
132 .fatal = 1,
133 },
134 {
135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
136 .fatal = 1,
137 },
138 {
139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
140 .fatal = 0,
141 },
142 {
143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
144 .fatal = 0,
145 },
146 {
147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
148 .fatal = 1,
149 },
150 {
151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
152 .fatal = 1,
153 },
154 {
155 .errno = ENONET, /* ICMP_HOST_ISOLATED */
156 .fatal = 1,
157 },
158 {
159 .errno = ENETUNREACH, /* ICMP_NET_ANO */
160 .fatal = 1,
161 },
162 {
163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
164 .fatal = 1,
165 },
166 {
167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
168 .fatal = 0,
169 },
170 {
171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
172 .fatal = 0,
173 },
174 {
175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
176 .fatal = 1,
177 },
178 {
179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
180 .fatal = 1,
181 },
182 {
183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
184 .fatal = 1,
185 },
186};
187EXPORT_SYMBOL(icmp_err_convert);
188
189/*
190 * ICMP control array. This specifies what to do with each ICMP.
191 */
192
193struct icmp_control {
194 bool (*handler)(struct sk_buff *skb);
195 short error; /* This ICMP is classed as an error message */
196};
197
198static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
199
200/*
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.
204 *
205 * On SMP we have one ICMP socket per-cpu.
206 */
207static struct sock *icmp_sk(struct net *net)
208{
209 return *this_cpu_ptr(net->ipv4.icmp_sk);
210}
211
212/* Called with BH disabled */
213static inline struct sock *icmp_xmit_lock(struct net *net)
214{
215 struct sock *sk;
216
217 sk = icmp_sk(net);
218
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.
222 */
223 return NULL;
224 }
225 return sk;
226}
227
228static inline void icmp_xmit_unlock(struct sock *sk)
229{
230 spin_unlock(&sk->sk_lock.slock);
231}
232
233int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
234int sysctl_icmp_msgs_burst __read_mostly = 50;
235
236static struct {
237 spinlock_t lock;
238 u32 credit;
239 u32 stamp;
240} icmp_global = {
241 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
242};
243
244/**
245 * icmp_global_allow - Are we allowed to send one more ICMP message ?
246 *
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
250 */
251bool icmp_global_allow(void)
252{
253 u32 credit, delta, incr = 0, now = (u32)jiffies;
254 bool rc = false;
255
256 /* Check if token bucket is empty and cannot be refilled
257 * without taking the spinlock.
258 */
259 if (!icmp_global.credit) {
260 delta = min_t(u32, now - icmp_global.stamp, HZ);
261 if (delta < HZ / 50)
262 return false;
263 }
264
265 spin_lock(&icmp_global.lock);
266 delta = min_t(u32, now - icmp_global.stamp, HZ);
267 if (delta >= HZ / 50) {
268 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
269 if (incr)
270 icmp_global.stamp = now;
271 }
272 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
273 if (credit) {
274 credit--;
275 rc = true;
276 }
277 icmp_global.credit = credit;
278 spin_unlock(&icmp_global.lock);
279 return rc;
280}
281EXPORT_SYMBOL(icmp_global_allow);
282
283static bool icmpv4_mask_allow(struct net *net, int type, int code)
284{
285 if (type > NR_ICMP_TYPES)
286 return true;
287
288 /* Don't limit PMTU discovery. */
289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
290 return true;
291
292 /* Limit if icmp type is enabled in ratemask. */
293 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
294 return true;
295
296 return false;
297}
298
299static bool icmpv4_global_allow(struct net *net, int type, int code)
300{
301 if (icmpv4_mask_allow(net, type, code))
302 return true;
303
304 if (icmp_global_allow())
305 return true;
306
307 return false;
308}
309
310/*
311 * Send an ICMP frame.
312 */
313
314static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 struct flowi4 *fl4, int type, int code)
316{
317 struct dst_entry *dst = &rt->dst;
318 struct inet_peer *peer;
319 bool rc = true;
320 int vif;
321
322 if (icmpv4_mask_allow(net, type, code))
323 goto out;
324
325 /* No rate limit on loopback */
326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
327 goto out;
328
329 vif = l3mdev_master_ifindex(dst->dev);
330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
332 if (peer)
333 inet_putpeer(peer);
334out:
335 return rc;
336}
337
338/*
339 * Maintain the counters used in the SNMP statistics for outgoing ICMP
340 */
341void icmp_out_count(struct net *net, unsigned char type)
342{
343 ICMPMSGOUT_INC_STATS(net, type);
344 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
345}
346
347/*
348 * Checksum each fragment, and on the first include the headers and final
349 * checksum.
350 */
351static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
352 struct sk_buff *skb)
353{
354 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
355 __wsum csum;
356
357 csum = skb_copy_and_csum_bits(icmp_param->skb,
358 icmp_param->offset + offset,
359 to, len, 0);
360
361 skb->csum = csum_block_add(skb->csum, csum, odd);
362 if (icmp_pointers[icmp_param->data.icmph.type].error)
363 nf_ct_attach(skb, icmp_param->skb);
364 return 0;
365}
366
367static void icmp_push_reply(struct icmp_bxm *icmp_param,
368 struct flowi4 *fl4,
369 struct ipcm_cookie *ipc, struct rtable **rt)
370{
371 struct sock *sk;
372 struct sk_buff *skb;
373
374 sk = icmp_sk(dev_net((*rt)->dst.dev));
375 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
376 icmp_param->data_len+icmp_param->head_len,
377 icmp_param->head_len,
378 ipc, rt, MSG_DONTWAIT) < 0) {
379 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
380 ip_flush_pending_frames(sk);
381 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
382 struct icmphdr *icmph = icmp_hdr(skb);
383 __wsum csum = 0;
384 struct sk_buff *skb1;
385
386 skb_queue_walk(&sk->sk_write_queue, skb1) {
387 csum = csum_add(csum, skb1->csum);
388 }
389 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
390 (char *)icmph,
391 icmp_param->head_len, csum);
392 icmph->checksum = csum_fold(csum);
393 skb->ip_summed = CHECKSUM_NONE;
394 ip_push_pending_frames(sk, fl4);
395 }
396}
397
398/*
399 * Driving logic for building and sending ICMP messages.
400 */
401
402static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
403{
404 struct ipcm_cookie ipc;
405 struct rtable *rt = skb_rtable(skb);
406 struct net *net = dev_net(rt->dst.dev);
407 struct flowi4 fl4;
408 struct sock *sk;
409 struct inet_sock *inet;
410 __be32 daddr, saddr;
411 u32 mark = IP4_REPLY_MARK(net, skb->mark);
412 int type = icmp_param->data.icmph.type;
413 int code = icmp_param->data.icmph.code;
414
415 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
416 return;
417
418 /* Needed by both icmp_global_allow and icmp_xmit_lock */
419 local_bh_disable();
420
421 /* global icmp_msgs_per_sec */
422 if (!icmpv4_global_allow(net, type, code))
423 goto out_bh_enable;
424
425 sk = icmp_xmit_lock(net);
426 if (!sk)
427 goto out_bh_enable;
428 inet = inet_sk(sk);
429
430 icmp_param->data.icmph.checksum = 0;
431
432 inet->tos = ip_hdr(skb)->tos;
433 sk->sk_mark = mark;
434 daddr = ipc.addr = ip_hdr(skb)->saddr;
435 saddr = fib_compute_spec_dst(skb);
436 ipc.opt = NULL;
437 ipc.tx_flags = 0;
438 ipc.ttl = 0;
439 ipc.tos = -1;
440
441 if (icmp_param->replyopts.opt.opt.optlen) {
442 ipc.opt = &icmp_param->replyopts.opt;
443 if (ipc.opt->opt.srr)
444 daddr = icmp_param->replyopts.opt.opt.faddr;
445 }
446 memset(&fl4, 0, sizeof(fl4));
447 fl4.daddr = daddr;
448 fl4.saddr = saddr;
449 fl4.flowi4_mark = mark;
450 fl4.flowi4_uid = sock_net_uid(net, NULL);
451 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
452 fl4.flowi4_proto = IPPROTO_ICMP;
453 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
454 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
455 rt = ip_route_output_key(net, &fl4);
456 if (IS_ERR(rt))
457 goto out_unlock;
458 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
459 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
460 ip_rt_put(rt);
461out_unlock:
462 icmp_xmit_unlock(sk);
463out_bh_enable:
464 local_bh_enable();
465}
466
467static struct rtable *icmp_route_lookup(struct net *net,
468 struct flowi4 *fl4,
469 struct sk_buff *skb_in,
470 const struct iphdr *iph,
471 __be32 saddr, u8 tos, u32 mark,
472 int type, int code,
473 struct icmp_bxm *param)
474{
475 struct rtable *rt, *rt2;
476 struct flowi4 fl4_dec;
477 int err;
478
479 memset(fl4, 0, sizeof(*fl4));
480 fl4->daddr = (param->replyopts.opt.opt.srr ?
481 param->replyopts.opt.opt.faddr : iph->saddr);
482 fl4->saddr = saddr;
483 fl4->flowi4_mark = mark;
484 fl4->flowi4_uid = sock_net_uid(net, NULL);
485 fl4->flowi4_tos = RT_TOS(tos);
486 fl4->flowi4_proto = IPPROTO_ICMP;
487 fl4->fl4_icmp_type = type;
488 fl4->fl4_icmp_code = code;
489 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
490
491 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
492 rt = ip_route_output_key_hash(net, fl4, skb_in);
493 if (IS_ERR(rt))
494 return rt;
495
496 /* No need to clone since we're just using its address. */
497 rt2 = rt;
498
499 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
500 flowi4_to_flowi(fl4), NULL, 0);
501 if (!IS_ERR(rt)) {
502 if (rt != rt2)
503 return rt;
504 } else if (PTR_ERR(rt) == -EPERM) {
505 rt = NULL;
506 } else
507 return rt;
508
509 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
510 if (err)
511 goto relookup_failed;
512
513 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
514 fl4_dec.saddr) == RTN_LOCAL) {
515 rt2 = __ip_route_output_key(net, &fl4_dec);
516 if (IS_ERR(rt2))
517 err = PTR_ERR(rt2);
518 } else {
519 struct flowi4 fl4_2 = {};
520 unsigned long orefdst;
521
522 fl4_2.daddr = fl4_dec.saddr;
523 rt2 = ip_route_output_key(net, &fl4_2);
524 if (IS_ERR(rt2)) {
525 err = PTR_ERR(rt2);
526 goto relookup_failed;
527 }
528 /* Ugh! */
529 orefdst = skb_in->_skb_refdst; /* save old refdst */
530 skb_dst_set(skb_in, NULL);
531 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
532 RT_TOS(tos), rt2->dst.dev);
533
534 dst_release(&rt2->dst);
535 rt2 = skb_rtable(skb_in);
536 skb_in->_skb_refdst = orefdst; /* restore old refdst */
537 }
538
539 if (err)
540 goto relookup_failed;
541
542 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
543 flowi4_to_flowi(&fl4_dec), NULL,
544 XFRM_LOOKUP_ICMP);
545 if (!IS_ERR(rt2)) {
546 dst_release(&rt->dst);
547 memcpy(fl4, &fl4_dec, sizeof(*fl4));
548 rt = rt2;
549 } else if (PTR_ERR(rt2) == -EPERM) {
550 if (rt)
551 dst_release(&rt->dst);
552 return rt2;
553 } else {
554 err = PTR_ERR(rt2);
555 goto relookup_failed;
556 }
557 return rt;
558
559relookup_failed:
560 if (rt)
561 return rt;
562 return ERR_PTR(err);
563}
564
565/*
566 * Send an ICMP message in response to a situation
567 *
568 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
569 * MAY send more (we do).
570 * MUST NOT change this header information.
571 * MUST NOT reply to a multicast/broadcast IP address.
572 * MUST NOT reply to a multicast/broadcast MAC address.
573 * MUST reply to only the first fragment.
574 */
575
576void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
577{
578 struct iphdr *iph;
579 int room;
580 struct icmp_bxm icmp_param;
581 struct rtable *rt = skb_rtable(skb_in);
582 struct ipcm_cookie ipc;
583 struct flowi4 fl4;
584 __be32 saddr;
585 u8 tos;
586 u32 mark;
587 struct net *net;
588 struct sock *sk;
589
590 if (!rt)
591 goto out;
592 net = dev_net(rt->dst.dev);
593
594 /*
595 * Find the original header. It is expected to be valid, of course.
596 * Check this, icmp_send is called from the most obscure devices
597 * sometimes.
598 */
599 iph = ip_hdr(skb_in);
600
601 if ((u8 *)iph < skb_in->head ||
602 (skb_network_header(skb_in) + sizeof(*iph)) >
603 skb_tail_pointer(skb_in))
604 goto out;
605
606 /*
607 * No replies to physical multicast/broadcast
608 */
609 if (skb_in->pkt_type != PACKET_HOST)
610 goto out;
611
612 /*
613 * Now check at the protocol level
614 */
615 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
616 goto out;
617
618 /*
619 * Only reply to fragment 0. We byte re-order the constant
620 * mask for efficiency.
621 */
622 if (iph->frag_off & htons(IP_OFFSET))
623 goto out;
624
625 /*
626 * If we send an ICMP error to an ICMP error a mess would result..
627 */
628 if (icmp_pointers[type].error) {
629 /*
630 * We are an error, check if we are replying to an
631 * ICMP error
632 */
633 if (iph->protocol == IPPROTO_ICMP) {
634 u8 _inner_type, *itp;
635
636 itp = skb_header_pointer(skb_in,
637 skb_network_header(skb_in) +
638 (iph->ihl << 2) +
639 offsetof(struct icmphdr,
640 type) -
641 skb_in->data,
642 sizeof(_inner_type),
643 &_inner_type);
644 if (!itp)
645 goto out;
646
647 /*
648 * Assume any unknown ICMP type is an error. This
649 * isn't specified by the RFC, but think about it..
650 */
651 if (*itp > NR_ICMP_TYPES ||
652 icmp_pointers[*itp].error)
653 goto out;
654 }
655 }
656
657 /* Needed by both icmp_global_allow and icmp_xmit_lock */
658 local_bh_disable();
659
660 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
661 * incoming dev is loopback. If outgoing dev change to not be
662 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
663 */
664 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
665 !icmpv4_global_allow(net, type, code))
666 goto out_bh_enable;
667
668 sk = icmp_xmit_lock(net);
669 if (!sk)
670 goto out_bh_enable;
671
672 /*
673 * Construct source address and options.
674 */
675
676 saddr = iph->daddr;
677 if (!(rt->rt_flags & RTCF_LOCAL)) {
678 struct net_device *dev = NULL;
679
680 rcu_read_lock();
681 if (rt_is_input_route(rt) &&
682 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
683 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
684
685 if (dev)
686 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
687 else
688 saddr = 0;
689 rcu_read_unlock();
690 }
691
692 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
693 IPTOS_PREC_INTERNETCONTROL) :
694 iph->tos;
695 mark = IP4_REPLY_MARK(net, skb_in->mark);
696
697 if (ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in))
698 goto out_unlock;
699
700
701 /*
702 * Prepare data for ICMP header.
703 */
704
705 icmp_param.data.icmph.type = type;
706 icmp_param.data.icmph.code = code;
707 icmp_param.data.icmph.un.gateway = info;
708 icmp_param.data.icmph.checksum = 0;
709 icmp_param.skb = skb_in;
710 icmp_param.offset = skb_network_offset(skb_in);
711 inet_sk(sk)->tos = tos;
712 sk->sk_mark = mark;
713 ipc.addr = iph->saddr;
714 ipc.opt = &icmp_param.replyopts.opt;
715 ipc.tx_flags = 0;
716 ipc.ttl = 0;
717 ipc.tos = -1;
718
719 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
720 type, code, &icmp_param);
721 if (IS_ERR(rt))
722 goto out_unlock;
723
724 /* peer icmp_ratelimit */
725 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
726 goto ende;
727
728 /* RFC says return as much as we can without exceeding 576 bytes. */
729
730 room = dst_mtu(&rt->dst);
731 if (room > 576)
732 room = 576;
733 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
734 room -= sizeof(struct icmphdr);
735
736 icmp_param.data_len = skb_in->len - icmp_param.offset;
737 if (icmp_param.data_len > room)
738 icmp_param.data_len = room;
739 icmp_param.head_len = sizeof(struct icmphdr);
740
741 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
742ende:
743 ip_rt_put(rt);
744out_unlock:
745 icmp_xmit_unlock(sk);
746out_bh_enable:
747 local_bh_enable();
748out:;
749}
750EXPORT_SYMBOL(icmp_send);
751
752
753static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
754{
755 const struct iphdr *iph = (const struct iphdr *) skb->data;
756 const struct net_protocol *ipprot;
757 int protocol = iph->protocol;
758
759 /* Checkin full IP header plus 8 bytes of protocol to
760 * avoid additional coding at protocol handlers.
761 */
762 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
763 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
764 return;
765 }
766
767 raw_icmp_error(skb, protocol, info);
768
769 ipprot = rcu_dereference(inet_protos[protocol]);
770 if (ipprot && ipprot->err_handler)
771 ipprot->err_handler(skb, info);
772}
773
774static bool icmp_tag_validation(int proto)
775{
776 bool ok;
777
778 rcu_read_lock();
779 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
780 rcu_read_unlock();
781 return ok;
782}
783
784/*
785 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
786 * ICMP_PARAMETERPROB.
787 */
788
789static bool icmp_unreach(struct sk_buff *skb)
790{
791 const struct iphdr *iph;
792 struct icmphdr *icmph;
793 struct net *net;
794 u32 info = 0;
795
796 net = dev_net(skb_dst(skb)->dev);
797
798 /*
799 * Incomplete header ?
800 * Only checks for the IP header, there should be an
801 * additional check for longer headers in upper levels.
802 */
803
804 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
805 goto out_err;
806
807 icmph = icmp_hdr(skb);
808 iph = (const struct iphdr *)skb->data;
809
810 if (iph->ihl < 5) /* Mangled header, drop. */
811 goto out_err;
812
813 switch (icmph->type) {
814 case ICMP_DEST_UNREACH:
815 switch (icmph->code & 15) {
816 case ICMP_NET_UNREACH:
817 case ICMP_HOST_UNREACH:
818 case ICMP_PROT_UNREACH:
819 case ICMP_PORT_UNREACH:
820 break;
821 case ICMP_FRAG_NEEDED:
822 /* for documentation of the ip_no_pmtu_disc
823 * values please see
824 * Documentation/networking/ip-sysctl.txt
825 */
826 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
827 default:
828 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
829 &iph->daddr);
830 break;
831 case 2:
832 goto out;
833 case 3:
834 if (!icmp_tag_validation(iph->protocol))
835 goto out;
836 /* fall through */
837 case 0:
838 info = ntohs(icmph->un.frag.mtu);
839 }
840 break;
841 case ICMP_SR_FAILED:
842 net_dbg_ratelimited("%pI4: Source Route Failed\n",
843 &iph->daddr);
844 break;
845 default:
846 break;
847 }
848 if (icmph->code > NR_ICMP_UNREACH)
849 goto out;
850 break;
851 case ICMP_PARAMETERPROB:
852 info = ntohl(icmph->un.gateway) >> 24;
853 break;
854 case ICMP_TIME_EXCEEDED:
855 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
856 if (icmph->code == ICMP_EXC_FRAGTIME)
857 goto out;
858 break;
859 }
860
861 /*
862 * Throw it at our lower layers
863 *
864 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
865 * header.
866 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
867 * transport layer.
868 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
869 * transport layer.
870 */
871
872 /*
873 * Check the other end isn't violating RFC 1122. Some routers send
874 * bogus responses to broadcast frames. If you see this message
875 * first check your netmask matches at both ends, if it does then
876 * get the other vendor to fix their kit.
877 */
878
879 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
880 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
881 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
882 &ip_hdr(skb)->saddr,
883 icmph->type, icmph->code,
884 &iph->daddr, skb->dev->name);
885 goto out;
886 }
887
888 icmp_socket_deliver(skb, info);
889
890out:
891 return true;
892out_err:
893 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
894 return false;
895}
896
897
898/*
899 * Handle ICMP_REDIRECT.
900 */
901
902static bool icmp_redirect(struct sk_buff *skb)
903{
904 if (skb->len < sizeof(struct iphdr)) {
905 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
906 return false;
907 }
908
909 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
910 /* there aught to be a stat */
911 return false;
912 }
913
914 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
915 return true;
916}
917
918/*
919 * Handle ICMP_ECHO ("ping") requests.
920 *
921 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
922 * requests.
923 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
924 * included in the reply.
925 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
926 * echo requests, MUST have default=NOT.
927 * See also WRT handling of options once they are done and working.
928 */
929
930static bool icmp_echo(struct sk_buff *skb)
931{
932 struct net *net;
933
934 net = dev_net(skb_dst(skb)->dev);
935 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
936 struct icmp_bxm icmp_param;
937
938 icmp_param.data.icmph = *icmp_hdr(skb);
939 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
940 icmp_param.skb = skb;
941 icmp_param.offset = 0;
942 icmp_param.data_len = skb->len;
943 icmp_param.head_len = sizeof(struct icmphdr);
944 icmp_reply(&icmp_param, skb);
945 }
946 /* should there be an ICMP stat for ignored echos? */
947 return true;
948}
949
950/*
951 * Handle ICMP Timestamp requests.
952 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
953 * SHOULD be in the kernel for minimum random latency.
954 * MUST be accurate to a few minutes.
955 * MUST be updated at least at 15Hz.
956 */
957static bool icmp_timestamp(struct sk_buff *skb)
958{
959 struct icmp_bxm icmp_param;
960 /*
961 * Too short.
962 */
963 if (skb->len < 4)
964 goto out_err;
965
966 /*
967 * Fill in the current time as ms since midnight UT:
968 */
969 icmp_param.data.times[1] = inet_current_timestamp();
970 icmp_param.data.times[2] = icmp_param.data.times[1];
971
972 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
973
974 icmp_param.data.icmph = *icmp_hdr(skb);
975 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
976 icmp_param.data.icmph.code = 0;
977 icmp_param.skb = skb;
978 icmp_param.offset = 0;
979 icmp_param.data_len = 0;
980 icmp_param.head_len = sizeof(struct icmphdr) + 12;
981 icmp_reply(&icmp_param, skb);
982 return true;
983
984out_err:
985 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
986 return false;
987}
988
989static bool icmp_discard(struct sk_buff *skb)
990{
991 /* pretend it was a success */
992 return true;
993}
994
995/*
996 * Deal with incoming ICMP packets.
997 */
998int icmp_rcv(struct sk_buff *skb)
999{
1000 struct icmphdr *icmph;
1001 struct rtable *rt = skb_rtable(skb);
1002 struct net *net = dev_net(rt->dst.dev);
1003 bool success;
1004
1005 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1006 struct sec_path *sp = skb_sec_path(skb);
1007 int nh;
1008
1009 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1010 XFRM_STATE_ICMP))
1011 goto drop;
1012
1013 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1014 goto drop;
1015
1016 nh = skb_network_offset(skb);
1017 skb_set_network_header(skb, sizeof(*icmph));
1018
1019 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1020 goto drop;
1021
1022 skb_set_network_header(skb, nh);
1023 }
1024
1025 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1026
1027 if (skb_checksum_simple_validate(skb))
1028 goto csum_error;
1029
1030 if (!pskb_pull(skb, sizeof(*icmph)))
1031 goto error;
1032
1033 icmph = icmp_hdr(skb);
1034
1035 ICMPMSGIN_INC_STATS(net, icmph->type);
1036 /*
1037 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1038 *
1039 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1040 * discarded.
1041 */
1042 if (icmph->type > NR_ICMP_TYPES)
1043 goto error;
1044
1045
1046 /*
1047 * Parse the ICMP message
1048 */
1049
1050 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1051 /*
1052 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1053 * silently ignored (we let user decide with a sysctl).
1054 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1055 * discarded if to broadcast/multicast.
1056 */
1057 if ((icmph->type == ICMP_ECHO ||
1058 icmph->type == ICMP_TIMESTAMP) &&
1059 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1060 goto error;
1061 }
1062 if (icmph->type != ICMP_ECHO &&
1063 icmph->type != ICMP_TIMESTAMP &&
1064 icmph->type != ICMP_ADDRESS &&
1065 icmph->type != ICMP_ADDRESSREPLY) {
1066 goto error;
1067 }
1068 }
1069
1070 success = icmp_pointers[icmph->type].handler(skb);
1071
1072 if (success) {
1073 consume_skb(skb);
1074 return NET_RX_SUCCESS;
1075 }
1076
1077drop:
1078 kfree_skb(skb);
1079 return NET_RX_DROP;
1080csum_error:
1081 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1082error:
1083 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1084 goto drop;
1085}
1086
1087void icmp_err(struct sk_buff *skb, u32 info)
1088{
1089 struct iphdr *iph = (struct iphdr *)skb->data;
1090 int offset = iph->ihl<<2;
1091 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1092 int type = icmp_hdr(skb)->type;
1093 int code = icmp_hdr(skb)->code;
1094 struct net *net = dev_net(skb->dev);
1095
1096 /*
1097 * Use ping_err to handle all icmp errors except those
1098 * triggered by ICMP_ECHOREPLY which sent from kernel.
1099 */
1100 if (icmph->type != ICMP_ECHOREPLY) {
1101 ping_err(skb, offset, info);
1102 return;
1103 }
1104
1105 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1106 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
1107 else if (type == ICMP_REDIRECT)
1108 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
1109}
1110
1111/*
1112 * This table is the definition of how we handle ICMP.
1113 */
1114static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1115 [ICMP_ECHOREPLY] = {
1116 .handler = ping_rcv,
1117 },
1118 [1] = {
1119 .handler = icmp_discard,
1120 .error = 1,
1121 },
1122 [2] = {
1123 .handler = icmp_discard,
1124 .error = 1,
1125 },
1126 [ICMP_DEST_UNREACH] = {
1127 .handler = icmp_unreach,
1128 .error = 1,
1129 },
1130 [ICMP_SOURCE_QUENCH] = {
1131 .handler = icmp_unreach,
1132 .error = 1,
1133 },
1134 [ICMP_REDIRECT] = {
1135 .handler = icmp_redirect,
1136 .error = 1,
1137 },
1138 [6] = {
1139 .handler = icmp_discard,
1140 .error = 1,
1141 },
1142 [7] = {
1143 .handler = icmp_discard,
1144 .error = 1,
1145 },
1146 [ICMP_ECHO] = {
1147 .handler = icmp_echo,
1148 },
1149 [9] = {
1150 .handler = icmp_discard,
1151 .error = 1,
1152 },
1153 [10] = {
1154 .handler = icmp_discard,
1155 .error = 1,
1156 },
1157 [ICMP_TIME_EXCEEDED] = {
1158 .handler = icmp_unreach,
1159 .error = 1,
1160 },
1161 [ICMP_PARAMETERPROB] = {
1162 .handler = icmp_unreach,
1163 .error = 1,
1164 },
1165 [ICMP_TIMESTAMP] = {
1166 .handler = icmp_timestamp,
1167 },
1168 [ICMP_TIMESTAMPREPLY] = {
1169 .handler = icmp_discard,
1170 },
1171 [ICMP_INFO_REQUEST] = {
1172 .handler = icmp_discard,
1173 },
1174 [ICMP_INFO_REPLY] = {
1175 .handler = icmp_discard,
1176 },
1177 [ICMP_ADDRESS] = {
1178 .handler = icmp_discard,
1179 },
1180 [ICMP_ADDRESSREPLY] = {
1181 .handler = icmp_discard,
1182 },
1183};
1184
1185static void __net_exit icmp_sk_exit(struct net *net)
1186{
1187 int i;
1188
1189 for_each_possible_cpu(i)
1190 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1191 free_percpu(net->ipv4.icmp_sk);
1192 net->ipv4.icmp_sk = NULL;
1193}
1194
1195static int __net_init icmp_sk_init(struct net *net)
1196{
1197 int i, err;
1198
1199 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1200 if (!net->ipv4.icmp_sk)
1201 return -ENOMEM;
1202
1203 for_each_possible_cpu(i) {
1204 struct sock *sk;
1205
1206 err = inet_ctl_sock_create(&sk, PF_INET,
1207 SOCK_RAW, IPPROTO_ICMP, net);
1208 if (err < 0)
1209 goto fail;
1210
1211 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1212
1213 /* Enough space for 2 64K ICMP packets, including
1214 * sk_buff/skb_shared_info struct overhead.
1215 */
1216 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1217
1218 /*
1219 * Speedup sock_wfree()
1220 */
1221 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1222 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1223 }
1224
1225 /* Control parameters for ECHO replies. */
1226 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1227 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1228
1229 /* Control parameter - ignore bogus broadcast responses? */
1230 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1231
1232 /*
1233 * Configurable global rate limit.
1234 *
1235 * ratelimit defines tokens/packet consumed for dst->rate_token
1236 * bucket ratemask defines which icmp types are ratelimited by
1237 * setting it's bit position.
1238 *
1239 * default:
1240 * dest unreachable (3), source quench (4),
1241 * time exceeded (11), parameter problem (12)
1242 */
1243
1244 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1245 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1246 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1247
1248 return 0;
1249
1250fail:
1251 for_each_possible_cpu(i)
1252 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1253 free_percpu(net->ipv4.icmp_sk);
1254 return err;
1255}
1256
1257static struct pernet_operations __net_initdata icmp_sk_ops = {
1258 .init = icmp_sk_init,
1259 .exit = icmp_sk_exit,
1260};
1261
1262int __init icmp_init(void)
1263{
1264 return register_pernet_subsys(&icmp_sk_ops);
1265}
1/*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
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.
10 *
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.
15 *
16 * Fixes:
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
23 * call.
24 * Alan Cox : Added 216,128 byte paths to the MTU
25 * code.
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
35 * (RFC 1812).
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
41 * path MTU bug.
42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
43 * valid (RFC 1812).
44 * Andi Kleen : Check all packet lengths properly
45 * and moved all kfree_skb() up to
46 * icmp_rcv.
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
54 * at all.
55 * Tristan Greaves : Added sysctl option to ignore bogus
56 * broadcast responses from broken routers.
57 *
58 * To Fix:
59 *
60 * - Should use skb_pull() instead of all the manual checking.
61 * This would also greatly simply some upper layer error handlers. --AK
62 *
63 */
64
65#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
66
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>
73#include <linux/in.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>
80#include <net/snmp.h>
81#include <net/ip.h>
82#include <net/route.h>
83#include <net/protocol.h>
84#include <net/icmp.h>
85#include <net/tcp.h>
86#include <net/udp.h>
87#include <net/raw.h>
88#include <net/ping.h>
89#include <linux/skbuff.h>
90#include <net/sock.h>
91#include <linux/errno.h>
92#include <linux/timer.h>
93#include <linux/init.h>
94#include <asm/uaccess.h>
95#include <net/checksum.h>
96#include <net/xfrm.h>
97#include <net/inet_common.h>
98
99/*
100 * Build xmit assembly blocks
101 */
102
103struct icmp_bxm {
104 struct sk_buff *skb;
105 int offset;
106 int data_len;
107
108 struct {
109 struct icmphdr icmph;
110 __be32 times[3];
111 } data;
112 int head_len;
113 struct ip_options_data replyopts;
114};
115
116/* An array of errno for error messages from dest unreach. */
117/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
118
119const struct icmp_err icmp_err_convert[] = {
120 {
121 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
122 .fatal = 0,
123 },
124 {
125 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
126 .fatal = 0,
127 },
128 {
129 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
130 .fatal = 1,
131 },
132 {
133 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
134 .fatal = 1,
135 },
136 {
137 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
138 .fatal = 0,
139 },
140 {
141 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
142 .fatal = 0,
143 },
144 {
145 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
146 .fatal = 1,
147 },
148 {
149 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
150 .fatal = 1,
151 },
152 {
153 .errno = ENONET, /* ICMP_HOST_ISOLATED */
154 .fatal = 1,
155 },
156 {
157 .errno = ENETUNREACH, /* ICMP_NET_ANO */
158 .fatal = 1,
159 },
160 {
161 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
162 .fatal = 1,
163 },
164 {
165 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
166 .fatal = 0,
167 },
168 {
169 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
170 .fatal = 0,
171 },
172 {
173 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
174 .fatal = 1,
175 },
176 {
177 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
178 .fatal = 1,
179 },
180 {
181 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
182 .fatal = 1,
183 },
184};
185EXPORT_SYMBOL(icmp_err_convert);
186
187/*
188 * ICMP control array. This specifies what to do with each ICMP.
189 */
190
191struct icmp_control {
192 void (*handler)(struct sk_buff *skb);
193 short error; /* This ICMP is classed as an error message */
194};
195
196static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
197
198/*
199 * The ICMP socket(s). This is the most convenient way to flow control
200 * our ICMP output as well as maintain a clean interface throughout
201 * all layers. All Socketless IP sends will soon be gone.
202 *
203 * On SMP we have one ICMP socket per-cpu.
204 */
205static struct sock *icmp_sk(struct net *net)
206{
207 return net->ipv4.icmp_sk[smp_processor_id()];
208}
209
210static inline struct sock *icmp_xmit_lock(struct net *net)
211{
212 struct sock *sk;
213
214 local_bh_disable();
215
216 sk = icmp_sk(net);
217
218 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 /* This can happen if the output path signals a
220 * dst_link_failure() for an outgoing ICMP packet.
221 */
222 local_bh_enable();
223 return NULL;
224 }
225 return sk;
226}
227
228static inline void icmp_xmit_unlock(struct sock *sk)
229{
230 spin_unlock_bh(&sk->sk_lock.slock);
231}
232
233/*
234 * Send an ICMP frame.
235 */
236
237static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
238 struct flowi4 *fl4, int type, int code)
239{
240 struct dst_entry *dst = &rt->dst;
241 bool rc = true;
242
243 if (type > NR_ICMP_TYPES)
244 goto out;
245
246 /* Don't limit PMTU discovery. */
247 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
248 goto out;
249
250 /* No rate limit on loopback */
251 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
252 goto out;
253
254 /* Limit if icmp type is enabled in ratemask. */
255 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
256 if (!rt->peer)
257 rt_bind_peer(rt, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(rt->peer,
259 net->ipv4.sysctl_icmp_ratelimit);
260 }
261out:
262 return rc;
263}
264
265/*
266 * Maintain the counters used in the SNMP statistics for outgoing ICMP
267 */
268void icmp_out_count(struct net *net, unsigned char type)
269{
270 ICMPMSGOUT_INC_STATS(net, type);
271 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
272}
273
274/*
275 * Checksum each fragment, and on the first include the headers and final
276 * checksum.
277 */
278static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
279 struct sk_buff *skb)
280{
281 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
282 __wsum csum;
283
284 csum = skb_copy_and_csum_bits(icmp_param->skb,
285 icmp_param->offset + offset,
286 to, len, 0);
287
288 skb->csum = csum_block_add(skb->csum, csum, odd);
289 if (icmp_pointers[icmp_param->data.icmph.type].error)
290 nf_ct_attach(skb, icmp_param->skb);
291 return 0;
292}
293
294static void icmp_push_reply(struct icmp_bxm *icmp_param,
295 struct flowi4 *fl4,
296 struct ipcm_cookie *ipc, struct rtable **rt)
297{
298 struct sock *sk;
299 struct sk_buff *skb;
300
301 sk = icmp_sk(dev_net((*rt)->dst.dev));
302 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
303 icmp_param->data_len+icmp_param->head_len,
304 icmp_param->head_len,
305 ipc, rt, MSG_DONTWAIT) < 0) {
306 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
307 ip_flush_pending_frames(sk);
308 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
309 struct icmphdr *icmph = icmp_hdr(skb);
310 __wsum csum = 0;
311 struct sk_buff *skb1;
312
313 skb_queue_walk(&sk->sk_write_queue, skb1) {
314 csum = csum_add(csum, skb1->csum);
315 }
316 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
317 (char *)icmph,
318 icmp_param->head_len, csum);
319 icmph->checksum = csum_fold(csum);
320 skb->ip_summed = CHECKSUM_NONE;
321 ip_push_pending_frames(sk, fl4);
322 }
323}
324
325/*
326 * Driving logic for building and sending ICMP messages.
327 */
328
329static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
330{
331 struct ipcm_cookie ipc;
332 struct rtable *rt = skb_rtable(skb);
333 struct net *net = dev_net(rt->dst.dev);
334 struct flowi4 fl4;
335 struct sock *sk;
336 struct inet_sock *inet;
337 __be32 daddr;
338
339 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
340 return;
341
342 sk = icmp_xmit_lock(net);
343 if (sk == NULL)
344 return;
345 inet = inet_sk(sk);
346
347 icmp_param->data.icmph.checksum = 0;
348
349 inet->tos = ip_hdr(skb)->tos;
350 daddr = ipc.addr = ip_hdr(skb)->saddr;
351 ipc.opt = NULL;
352 ipc.tx_flags = 0;
353 if (icmp_param->replyopts.opt.opt.optlen) {
354 ipc.opt = &icmp_param->replyopts.opt;
355 if (ipc.opt->opt.srr)
356 daddr = icmp_param->replyopts.opt.opt.faddr;
357 }
358 memset(&fl4, 0, sizeof(fl4));
359 fl4.daddr = daddr;
360 fl4.saddr = rt->rt_spec_dst;
361 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
362 fl4.flowi4_proto = IPPROTO_ICMP;
363 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
364 rt = ip_route_output_key(net, &fl4);
365 if (IS_ERR(rt))
366 goto out_unlock;
367 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
368 icmp_param->data.icmph.code))
369 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
370 ip_rt_put(rt);
371out_unlock:
372 icmp_xmit_unlock(sk);
373}
374
375static struct rtable *icmp_route_lookup(struct net *net,
376 struct flowi4 *fl4,
377 struct sk_buff *skb_in,
378 const struct iphdr *iph,
379 __be32 saddr, u8 tos,
380 int type, int code,
381 struct icmp_bxm *param)
382{
383 struct rtable *rt, *rt2;
384 struct flowi4 fl4_dec;
385 int err;
386
387 memset(fl4, 0, sizeof(*fl4));
388 fl4->daddr = (param->replyopts.opt.opt.srr ?
389 param->replyopts.opt.opt.faddr : iph->saddr);
390 fl4->saddr = saddr;
391 fl4->flowi4_tos = RT_TOS(tos);
392 fl4->flowi4_proto = IPPROTO_ICMP;
393 fl4->fl4_icmp_type = type;
394 fl4->fl4_icmp_code = code;
395 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
396 rt = __ip_route_output_key(net, fl4);
397 if (IS_ERR(rt))
398 return rt;
399
400 /* No need to clone since we're just using its address. */
401 rt2 = rt;
402
403 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
404 flowi4_to_flowi(fl4), NULL, 0);
405 if (!IS_ERR(rt)) {
406 if (rt != rt2)
407 return rt;
408 } else if (PTR_ERR(rt) == -EPERM) {
409 rt = NULL;
410 } else
411 return rt;
412
413 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
414 if (err)
415 goto relookup_failed;
416
417 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
418 rt2 = __ip_route_output_key(net, &fl4_dec);
419 if (IS_ERR(rt2))
420 err = PTR_ERR(rt2);
421 } else {
422 struct flowi4 fl4_2 = {};
423 unsigned long orefdst;
424
425 fl4_2.daddr = fl4_dec.saddr;
426 rt2 = ip_route_output_key(net, &fl4_2);
427 if (IS_ERR(rt2)) {
428 err = PTR_ERR(rt2);
429 goto relookup_failed;
430 }
431 /* Ugh! */
432 orefdst = skb_in->_skb_refdst; /* save old refdst */
433 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
434 RT_TOS(tos), rt2->dst.dev);
435
436 dst_release(&rt2->dst);
437 rt2 = skb_rtable(skb_in);
438 skb_in->_skb_refdst = orefdst; /* restore old refdst */
439 }
440
441 if (err)
442 goto relookup_failed;
443
444 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
445 flowi4_to_flowi(&fl4_dec), NULL,
446 XFRM_LOOKUP_ICMP);
447 if (!IS_ERR(rt2)) {
448 dst_release(&rt->dst);
449 memcpy(fl4, &fl4_dec, sizeof(*fl4));
450 rt = rt2;
451 } else if (PTR_ERR(rt2) == -EPERM) {
452 if (rt)
453 dst_release(&rt->dst);
454 return rt2;
455 } else {
456 err = PTR_ERR(rt2);
457 goto relookup_failed;
458 }
459 return rt;
460
461relookup_failed:
462 if (rt)
463 return rt;
464 return ERR_PTR(err);
465}
466
467/*
468 * Send an ICMP message in response to a situation
469 *
470 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
471 * MAY send more (we do).
472 * MUST NOT change this header information.
473 * MUST NOT reply to a multicast/broadcast IP address.
474 * MUST NOT reply to a multicast/broadcast MAC address.
475 * MUST reply to only the first fragment.
476 */
477
478void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
479{
480 struct iphdr *iph;
481 int room;
482 struct icmp_bxm icmp_param;
483 struct rtable *rt = skb_rtable(skb_in);
484 struct ipcm_cookie ipc;
485 struct flowi4 fl4;
486 __be32 saddr;
487 u8 tos;
488 struct net *net;
489 struct sock *sk;
490
491 if (!rt)
492 goto out;
493 net = dev_net(rt->dst.dev);
494
495 /*
496 * Find the original header. It is expected to be valid, of course.
497 * Check this, icmp_send is called from the most obscure devices
498 * sometimes.
499 */
500 iph = ip_hdr(skb_in);
501
502 if ((u8 *)iph < skb_in->head ||
503 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
504 goto out;
505
506 /*
507 * No replies to physical multicast/broadcast
508 */
509 if (skb_in->pkt_type != PACKET_HOST)
510 goto out;
511
512 /*
513 * Now check at the protocol level
514 */
515 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
516 goto out;
517
518 /*
519 * Only reply to fragment 0. We byte re-order the constant
520 * mask for efficiency.
521 */
522 if (iph->frag_off & htons(IP_OFFSET))
523 goto out;
524
525 /*
526 * If we send an ICMP error to an ICMP error a mess would result..
527 */
528 if (icmp_pointers[type].error) {
529 /*
530 * We are an error, check if we are replying to an
531 * ICMP error
532 */
533 if (iph->protocol == IPPROTO_ICMP) {
534 u8 _inner_type, *itp;
535
536 itp = skb_header_pointer(skb_in,
537 skb_network_header(skb_in) +
538 (iph->ihl << 2) +
539 offsetof(struct icmphdr,
540 type) -
541 skb_in->data,
542 sizeof(_inner_type),
543 &_inner_type);
544 if (itp == NULL)
545 goto out;
546
547 /*
548 * Assume any unknown ICMP type is an error. This
549 * isn't specified by the RFC, but think about it..
550 */
551 if (*itp > NR_ICMP_TYPES ||
552 icmp_pointers[*itp].error)
553 goto out;
554 }
555 }
556
557 sk = icmp_xmit_lock(net);
558 if (sk == NULL)
559 return;
560
561 /*
562 * Construct source address and options.
563 */
564
565 saddr = iph->daddr;
566 if (!(rt->rt_flags & RTCF_LOCAL)) {
567 struct net_device *dev = NULL;
568
569 rcu_read_lock();
570 if (rt_is_input_route(rt) &&
571 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
572 dev = dev_get_by_index_rcu(net, rt->rt_iif);
573
574 if (dev)
575 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
576 else
577 saddr = 0;
578 rcu_read_unlock();
579 }
580
581 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
582 IPTOS_PREC_INTERNETCONTROL) :
583 iph->tos;
584
585 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
586 goto out_unlock;
587
588
589 /*
590 * Prepare data for ICMP header.
591 */
592
593 icmp_param.data.icmph.type = type;
594 icmp_param.data.icmph.code = code;
595 icmp_param.data.icmph.un.gateway = info;
596 icmp_param.data.icmph.checksum = 0;
597 icmp_param.skb = skb_in;
598 icmp_param.offset = skb_network_offset(skb_in);
599 inet_sk(sk)->tos = tos;
600 ipc.addr = iph->saddr;
601 ipc.opt = &icmp_param.replyopts.opt;
602 ipc.tx_flags = 0;
603
604 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
605 type, code, &icmp_param);
606 if (IS_ERR(rt))
607 goto out_unlock;
608
609 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
610 goto ende;
611
612 /* RFC says return as much as we can without exceeding 576 bytes. */
613
614 room = dst_mtu(&rt->dst);
615 if (room > 576)
616 room = 576;
617 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
618 room -= sizeof(struct icmphdr);
619
620 icmp_param.data_len = skb_in->len - icmp_param.offset;
621 if (icmp_param.data_len > room)
622 icmp_param.data_len = room;
623 icmp_param.head_len = sizeof(struct icmphdr);
624
625 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
626ende:
627 ip_rt_put(rt);
628out_unlock:
629 icmp_xmit_unlock(sk);
630out:;
631}
632EXPORT_SYMBOL(icmp_send);
633
634
635/*
636 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
637 */
638
639static void icmp_unreach(struct sk_buff *skb)
640{
641 const struct iphdr *iph;
642 struct icmphdr *icmph;
643 int hash, protocol;
644 const struct net_protocol *ipprot;
645 u32 info = 0;
646 struct net *net;
647
648 net = dev_net(skb_dst(skb)->dev);
649
650 /*
651 * Incomplete header ?
652 * Only checks for the IP header, there should be an
653 * additional check for longer headers in upper levels.
654 */
655
656 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
657 goto out_err;
658
659 icmph = icmp_hdr(skb);
660 iph = (const struct iphdr *)skb->data;
661
662 if (iph->ihl < 5) /* Mangled header, drop. */
663 goto out_err;
664
665 if (icmph->type == ICMP_DEST_UNREACH) {
666 switch (icmph->code & 15) {
667 case ICMP_NET_UNREACH:
668 case ICMP_HOST_UNREACH:
669 case ICMP_PROT_UNREACH:
670 case ICMP_PORT_UNREACH:
671 break;
672 case ICMP_FRAG_NEEDED:
673 if (ipv4_config.no_pmtu_disc) {
674 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
675 &iph->daddr);
676 } else {
677 info = ip_rt_frag_needed(net, iph,
678 ntohs(icmph->un.frag.mtu),
679 skb->dev);
680 if (!info)
681 goto out;
682 }
683 break;
684 case ICMP_SR_FAILED:
685 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
686 &iph->daddr);
687 break;
688 default:
689 break;
690 }
691 if (icmph->code > NR_ICMP_UNREACH)
692 goto out;
693 } else if (icmph->type == ICMP_PARAMETERPROB)
694 info = ntohl(icmph->un.gateway) >> 24;
695
696 /*
697 * Throw it at our lower layers
698 *
699 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
700 * header.
701 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
702 * transport layer.
703 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
704 * transport layer.
705 */
706
707 /*
708 * Check the other end isn't violating RFC 1122. Some routers send
709 * bogus responses to broadcast frames. If you see this message
710 * first check your netmask matches at both ends, if it does then
711 * get the other vendor to fix their kit.
712 */
713
714 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
715 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
716 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
717 &ip_hdr(skb)->saddr,
718 icmph->type, icmph->code,
719 &iph->daddr, skb->dev->name);
720 goto out;
721 }
722
723 /* Checkin full IP header plus 8 bytes of protocol to
724 * avoid additional coding at protocol handlers.
725 */
726 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
727 goto out;
728
729 iph = (const struct iphdr *)skb->data;
730 protocol = iph->protocol;
731
732 /*
733 * Deliver ICMP message to raw sockets. Pretty useless feature?
734 */
735 raw_icmp_error(skb, protocol, info);
736
737 hash = protocol & (MAX_INET_PROTOS - 1);
738 rcu_read_lock();
739 ipprot = rcu_dereference(inet_protos[hash]);
740 if (ipprot && ipprot->err_handler)
741 ipprot->err_handler(skb, info);
742 rcu_read_unlock();
743
744out:
745 return;
746out_err:
747 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
748 goto out;
749}
750
751
752/*
753 * Handle ICMP_REDIRECT.
754 */
755
756static void icmp_redirect(struct sk_buff *skb)
757{
758 const struct iphdr *iph;
759
760 if (skb->len < sizeof(struct iphdr))
761 goto out_err;
762
763 /*
764 * Get the copied header of the packet that caused the redirect
765 */
766 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
767 goto out;
768
769 iph = (const struct iphdr *)skb->data;
770
771 switch (icmp_hdr(skb)->code & 7) {
772 case ICMP_REDIR_NET:
773 case ICMP_REDIR_NETTOS:
774 /*
775 * As per RFC recommendations now handle it as a host redirect.
776 */
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);
782 break;
783 }
784
785 /* Ping wants to see redirects.
786 * Let's pretend they are errors of sorts... */
787 if (iph->protocol == IPPROTO_ICMP &&
788 iph->ihl >= 5 &&
789 pskb_may_pull(skb, (iph->ihl<<2)+8)) {
790 ping_err(skb, icmp_hdr(skb)->un.gateway);
791 }
792
793out:
794 return;
795out_err:
796 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
797 goto out;
798}
799
800/*
801 * Handle ICMP_ECHO ("ping") requests.
802 *
803 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
804 * requests.
805 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
806 * included in the reply.
807 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
808 * echo requests, MUST have default=NOT.
809 * See also WRT handling of options once they are done and working.
810 */
811
812static void icmp_echo(struct sk_buff *skb)
813{
814 struct net *net;
815
816 net = dev_net(skb_dst(skb)->dev);
817 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
818 struct icmp_bxm icmp_param;
819
820 icmp_param.data.icmph = *icmp_hdr(skb);
821 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
822 icmp_param.skb = skb;
823 icmp_param.offset = 0;
824 icmp_param.data_len = skb->len;
825 icmp_param.head_len = sizeof(struct icmphdr);
826 icmp_reply(&icmp_param, skb);
827 }
828}
829
830/*
831 * Handle ICMP Timestamp requests.
832 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
833 * SHOULD be in the kernel for minimum random latency.
834 * MUST be accurate to a few minutes.
835 * MUST be updated at least at 15Hz.
836 */
837static void icmp_timestamp(struct sk_buff *skb)
838{
839 struct timespec tv;
840 struct icmp_bxm icmp_param;
841 /*
842 * Too short.
843 */
844 if (skb->len < 4)
845 goto out_err;
846
847 /*
848 * Fill in the current time as ms since midnight UT:
849 */
850 getnstimeofday(&tv);
851 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
852 tv.tv_nsec / NSEC_PER_MSEC);
853 icmp_param.data.times[2] = icmp_param.data.times[1];
854 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
855 BUG();
856 icmp_param.data.icmph = *icmp_hdr(skb);
857 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
858 icmp_param.data.icmph.code = 0;
859 icmp_param.skb = skb;
860 icmp_param.offset = 0;
861 icmp_param.data_len = 0;
862 icmp_param.head_len = sizeof(struct icmphdr) + 12;
863 icmp_reply(&icmp_param, skb);
864out:
865 return;
866out_err:
867 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
868 goto out;
869}
870
871
872/*
873 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
874 *
875 * RFC1122 (3.2.2.9). A host MUST only send replies to
876 * ADDRESS_MASK requests if it's been configured as an address mask
877 * agent. Receiving a request doesn't constitute implicit permission to
878 * act as one. Of course, implementing this correctly requires (SHOULD)
879 * a way to turn the functionality on and off. Another one for sysctl(),
880 * I guess. -- MS
881 *
882 * RFC1812 (4.3.3.9). A router MUST implement it.
883 * A router SHOULD have switch turning it on/off.
884 * This switch MUST be ON by default.
885 *
886 * Gratuitous replies, zero-source replies are not implemented,
887 * that complies with RFC. DO NOT implement them!!! All the idea
888 * of broadcast addrmask replies as specified in RFC950 is broken.
889 * The problem is that it is not uncommon to have several prefixes
890 * on one physical interface. Moreover, addrmask agent can even be
891 * not aware of existing another prefixes.
892 * If source is zero, addrmask agent cannot choose correct prefix.
893 * Gratuitous mask announcements suffer from the same problem.
894 * RFC1812 explains it, but still allows to use ADDRMASK,
895 * that is pretty silly. --ANK
896 *
897 * All these rules are so bizarre, that I removed kernel addrmask
898 * support at all. It is wrong, it is obsolete, nobody uses it in
899 * any case. --ANK
900 *
901 * Furthermore you can do it with a usermode address agent program
902 * anyway...
903 */
904
905static void icmp_address(struct sk_buff *skb)
906{
907#if 0
908 net_dbg_ratelimited("a guy asks for address mask. Who is it?\n");
909#endif
910}
911
912/*
913 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
914 * loudly if an inconsistency is found.
915 * called with rcu_read_lock()
916 */
917
918static void icmp_address_reply(struct sk_buff *skb)
919{
920 struct rtable *rt = skb_rtable(skb);
921 struct net_device *dev = skb->dev;
922 struct in_device *in_dev;
923 struct in_ifaddr *ifa;
924
925 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
926 return;
927
928 in_dev = __in_dev_get_rcu(dev);
929 if (!in_dev)
930 return;
931
932 if (in_dev->ifa_list &&
933 IN_DEV_LOG_MARTIANS(in_dev) &&
934 IN_DEV_FORWARD(in_dev)) {
935 __be32 _mask, *mp;
936
937 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
938 BUG_ON(mp == NULL);
939 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
940 if (*mp == ifa->ifa_mask &&
941 inet_ifa_match(ip_hdr(skb)->saddr, ifa))
942 break;
943 }
944 if (!ifa)
945 net_info_ratelimited("Wrong address mask %pI4 from %s/%pI4\n",
946 mp,
947 dev->name, &ip_hdr(skb)->saddr);
948 }
949}
950
951static void icmp_discard(struct sk_buff *skb)
952{
953}
954
955/*
956 * Deal with incoming ICMP packets.
957 */
958int icmp_rcv(struct sk_buff *skb)
959{
960 struct icmphdr *icmph;
961 struct rtable *rt = skb_rtable(skb);
962 struct net *net = dev_net(rt->dst.dev);
963
964 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
965 struct sec_path *sp = skb_sec_path(skb);
966 int nh;
967
968 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
969 XFRM_STATE_ICMP))
970 goto drop;
971
972 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
973 goto drop;
974
975 nh = skb_network_offset(skb);
976 skb_set_network_header(skb, sizeof(*icmph));
977
978 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
979 goto drop;
980
981 skb_set_network_header(skb, nh);
982 }
983
984 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
985
986 switch (skb->ip_summed) {
987 case CHECKSUM_COMPLETE:
988 if (!csum_fold(skb->csum))
989 break;
990 /* fall through */
991 case CHECKSUM_NONE:
992 skb->csum = 0;
993 if (__skb_checksum_complete(skb))
994 goto error;
995 }
996
997 if (!pskb_pull(skb, sizeof(*icmph)))
998 goto error;
999
1000 icmph = icmp_hdr(skb);
1001
1002 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1003 /*
1004 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1005 *
1006 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1007 * discarded.
1008 */
1009 if (icmph->type > NR_ICMP_TYPES)
1010 goto error;
1011
1012
1013 /*
1014 * Parse the ICMP message
1015 */
1016
1017 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1018 /*
1019 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1020 * silently ignored (we let user decide with a sysctl).
1021 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1022 * discarded if to broadcast/multicast.
1023 */
1024 if ((icmph->type == ICMP_ECHO ||
1025 icmph->type == ICMP_TIMESTAMP) &&
1026 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1027 goto error;
1028 }
1029 if (icmph->type != ICMP_ECHO &&
1030 icmph->type != ICMP_TIMESTAMP &&
1031 icmph->type != ICMP_ADDRESS &&
1032 icmph->type != ICMP_ADDRESSREPLY) {
1033 goto error;
1034 }
1035 }
1036
1037 icmp_pointers[icmph->type].handler(skb);
1038
1039drop:
1040 kfree_skb(skb);
1041 return 0;
1042error:
1043 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1044 goto drop;
1045}
1046
1047/*
1048 * This table is the definition of how we handle ICMP.
1049 */
1050static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1051 [ICMP_ECHOREPLY] = {
1052 .handler = ping_rcv,
1053 },
1054 [1] = {
1055 .handler = icmp_discard,
1056 .error = 1,
1057 },
1058 [2] = {
1059 .handler = icmp_discard,
1060 .error = 1,
1061 },
1062 [ICMP_DEST_UNREACH] = {
1063 .handler = icmp_unreach,
1064 .error = 1,
1065 },
1066 [ICMP_SOURCE_QUENCH] = {
1067 .handler = icmp_unreach,
1068 .error = 1,
1069 },
1070 [ICMP_REDIRECT] = {
1071 .handler = icmp_redirect,
1072 .error = 1,
1073 },
1074 [6] = {
1075 .handler = icmp_discard,
1076 .error = 1,
1077 },
1078 [7] = {
1079 .handler = icmp_discard,
1080 .error = 1,
1081 },
1082 [ICMP_ECHO] = {
1083 .handler = icmp_echo,
1084 },
1085 [9] = {
1086 .handler = icmp_discard,
1087 .error = 1,
1088 },
1089 [10] = {
1090 .handler = icmp_discard,
1091 .error = 1,
1092 },
1093 [ICMP_TIME_EXCEEDED] = {
1094 .handler = icmp_unreach,
1095 .error = 1,
1096 },
1097 [ICMP_PARAMETERPROB] = {
1098 .handler = icmp_unreach,
1099 .error = 1,
1100 },
1101 [ICMP_TIMESTAMP] = {
1102 .handler = icmp_timestamp,
1103 },
1104 [ICMP_TIMESTAMPREPLY] = {
1105 .handler = icmp_discard,
1106 },
1107 [ICMP_INFO_REQUEST] = {
1108 .handler = icmp_discard,
1109 },
1110 [ICMP_INFO_REPLY] = {
1111 .handler = icmp_discard,
1112 },
1113 [ICMP_ADDRESS] = {
1114 .handler = icmp_address,
1115 },
1116 [ICMP_ADDRESSREPLY] = {
1117 .handler = icmp_address_reply,
1118 },
1119};
1120
1121static void __net_exit icmp_sk_exit(struct net *net)
1122{
1123 int i;
1124
1125 for_each_possible_cpu(i)
1126 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1127 kfree(net->ipv4.icmp_sk);
1128 net->ipv4.icmp_sk = NULL;
1129}
1130
1131static int __net_init icmp_sk_init(struct net *net)
1132{
1133 int i, err;
1134
1135 net->ipv4.icmp_sk =
1136 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1137 if (net->ipv4.icmp_sk == NULL)
1138 return -ENOMEM;
1139
1140 for_each_possible_cpu(i) {
1141 struct sock *sk;
1142
1143 err = inet_ctl_sock_create(&sk, PF_INET,
1144 SOCK_RAW, IPPROTO_ICMP, net);
1145 if (err < 0)
1146 goto fail;
1147
1148 net->ipv4.icmp_sk[i] = sk;
1149
1150 /* Enough space for 2 64K ICMP packets, including
1151 * sk_buff/skb_shared_info struct overhead.
1152 */
1153 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1154
1155 /*
1156 * Speedup sock_wfree()
1157 */
1158 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1159 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1160 }
1161
1162 /* Control parameters for ECHO replies. */
1163 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1164 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1165
1166 /* Control parameter - ignore bogus broadcast responses? */
1167 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1168
1169 /*
1170 * Configurable global rate limit.
1171 *
1172 * ratelimit defines tokens/packet consumed for dst->rate_token
1173 * bucket ratemask defines which icmp types are ratelimited by
1174 * setting it's bit position.
1175 *
1176 * default:
1177 * dest unreachable (3), source quench (4),
1178 * time exceeded (11), parameter problem (12)
1179 */
1180
1181 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1182 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1183 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1184
1185 return 0;
1186
1187fail:
1188 for_each_possible_cpu(i)
1189 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1190 kfree(net->ipv4.icmp_sk);
1191 return err;
1192}
1193
1194static struct pernet_operations __net_initdata icmp_sk_ops = {
1195 .init = icmp_sk_init,
1196 .exit = icmp_sk_exit,
1197};
1198
1199int __init icmp_init(void)
1200{
1201 return register_pernet_subsys(&icmp_sk_ops);
1202}