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 <asm/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
100/*
101 * Build xmit assembly blocks
102 */
103
104struct icmp_bxm {
105 struct sk_buff *skb;
106 int offset;
107 int data_len;
108
109 struct {
110 struct icmphdr icmph;
111 __be32 times[3];
112 } data;
113 int head_len;
114 struct ip_options_data replyopts;
115};
116
117/* An array of errno for error messages from dest unreach. */
118/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
119
120const struct icmp_err icmp_err_convert[] = {
121 {
122 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
123 .fatal = 0,
124 },
125 {
126 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
127 .fatal = 0,
128 },
129 {
130 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
131 .fatal = 1,
132 },
133 {
134 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
135 .fatal = 1,
136 },
137 {
138 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
139 .fatal = 0,
140 },
141 {
142 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
143 .fatal = 0,
144 },
145 {
146 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
147 .fatal = 1,
148 },
149 {
150 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
151 .fatal = 1,
152 },
153 {
154 .errno = ENONET, /* ICMP_HOST_ISOLATED */
155 .fatal = 1,
156 },
157 {
158 .errno = ENETUNREACH, /* ICMP_NET_ANO */
159 .fatal = 1,
160 },
161 {
162 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
163 .fatal = 1,
164 },
165 {
166 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
167 .fatal = 0,
168 },
169 {
170 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
171 .fatal = 0,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
175 .fatal = 1,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
179 .fatal = 1,
180 },
181 {
182 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
183 .fatal = 1,
184 },
185};
186EXPORT_SYMBOL(icmp_err_convert);
187
188/*
189 * ICMP control array. This specifies what to do with each ICMP.
190 */
191
192struct icmp_control {
193 void (*handler)(struct sk_buff *skb);
194 short error; /* This ICMP is classed as an error message */
195};
196
197static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
198
199/*
200 * The ICMP socket(s). This is the most convenient way to flow control
201 * our ICMP output as well as maintain a clean interface throughout
202 * all layers. All Socketless IP sends will soon be gone.
203 *
204 * On SMP we have one ICMP socket per-cpu.
205 */
206static struct sock *icmp_sk(struct net *net)
207{
208 return net->ipv4.icmp_sk[smp_processor_id()];
209}
210
211static inline struct sock *icmp_xmit_lock(struct net *net)
212{
213 struct sock *sk;
214
215 local_bh_disable();
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 local_bh_enable();
224 return NULL;
225 }
226 return sk;
227}
228
229static inline void icmp_xmit_unlock(struct sock *sk)
230{
231 spin_unlock_bh(&sk->sk_lock.slock);
232}
233
234/*
235 * Send an ICMP frame.
236 */
237
238static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 struct flowi4 *fl4, int type, int code)
240{
241 struct dst_entry *dst = &rt->dst;
242 bool rc = true;
243
244 if (type > NR_ICMP_TYPES)
245 goto out;
246
247 /* Don't limit PMTU discovery. */
248 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 goto out;
250
251 /* No rate limit on loopback */
252 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 goto out;
254
255 /* Limit if icmp type is enabled in ratemask. */
256 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(peer,
259 net->ipv4.sysctl_icmp_ratelimit);
260 if (peer)
261 inet_putpeer(peer);
262 }
263out:
264 return rc;
265}
266
267/*
268 * Maintain the counters used in the SNMP statistics for outgoing ICMP
269 */
270void icmp_out_count(struct net *net, unsigned char type)
271{
272 ICMPMSGOUT_INC_STATS(net, type);
273 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
274}
275
276/*
277 * Checksum each fragment, and on the first include the headers and final
278 * checksum.
279 */
280static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
281 struct sk_buff *skb)
282{
283 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
284 __wsum csum;
285
286 csum = skb_copy_and_csum_bits(icmp_param->skb,
287 icmp_param->offset + offset,
288 to, len, 0);
289
290 skb->csum = csum_block_add(skb->csum, csum, odd);
291 if (icmp_pointers[icmp_param->data.icmph.type].error)
292 nf_ct_attach(skb, icmp_param->skb);
293 return 0;
294}
295
296static void icmp_push_reply(struct icmp_bxm *icmp_param,
297 struct flowi4 *fl4,
298 struct ipcm_cookie *ipc, struct rtable **rt)
299{
300 struct sock *sk;
301 struct sk_buff *skb;
302
303 sk = icmp_sk(dev_net((*rt)->dst.dev));
304 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
305 icmp_param->data_len+icmp_param->head_len,
306 icmp_param->head_len,
307 ipc, rt, MSG_DONTWAIT) < 0) {
308 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
309 ip_flush_pending_frames(sk);
310 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
311 struct icmphdr *icmph = icmp_hdr(skb);
312 __wsum csum = 0;
313 struct sk_buff *skb1;
314
315 skb_queue_walk(&sk->sk_write_queue, skb1) {
316 csum = csum_add(csum, skb1->csum);
317 }
318 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
319 (char *)icmph,
320 icmp_param->head_len, csum);
321 icmph->checksum = csum_fold(csum);
322 skb->ip_summed = CHECKSUM_NONE;
323 ip_push_pending_frames(sk, fl4);
324 }
325}
326
327/*
328 * Driving logic for building and sending ICMP messages.
329 */
330
331static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
332{
333 struct ipcm_cookie ipc;
334 struct rtable *rt = skb_rtable(skb);
335 struct net *net = dev_net(rt->dst.dev);
336 struct flowi4 fl4;
337 struct sock *sk;
338 struct inet_sock *inet;
339 __be32 daddr, saddr;
340
341 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
342 return;
343
344 sk = icmp_xmit_lock(net);
345 if (sk == NULL)
346 return;
347 inet = inet_sk(sk);
348
349 icmp_param->data.icmph.checksum = 0;
350
351 inet->tos = ip_hdr(skb)->tos;
352 daddr = ipc.addr = ip_hdr(skb)->saddr;
353 saddr = fib_compute_spec_dst(skb);
354 ipc.opt = NULL;
355 ipc.tx_flags = 0;
356 ipc.ttl = 0;
357 ipc.tos = -1;
358
359 if (icmp_param->replyopts.opt.opt.optlen) {
360 ipc.opt = &icmp_param->replyopts.opt;
361 if (ipc.opt->opt.srr)
362 daddr = icmp_param->replyopts.opt.opt.faddr;
363 }
364 memset(&fl4, 0, sizeof(fl4));
365 fl4.daddr = daddr;
366 fl4.saddr = saddr;
367 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
368 fl4.flowi4_proto = IPPROTO_ICMP;
369 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
370 rt = ip_route_output_key(net, &fl4);
371 if (IS_ERR(rt))
372 goto out_unlock;
373 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
374 icmp_param->data.icmph.code))
375 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
376 ip_rt_put(rt);
377out_unlock:
378 icmp_xmit_unlock(sk);
379}
380
381static struct rtable *icmp_route_lookup(struct net *net,
382 struct flowi4 *fl4,
383 struct sk_buff *skb_in,
384 const struct iphdr *iph,
385 __be32 saddr, u8 tos,
386 int type, int code,
387 struct icmp_bxm *param)
388{
389 struct rtable *rt, *rt2;
390 struct flowi4 fl4_dec;
391 int err;
392
393 memset(fl4, 0, sizeof(*fl4));
394 fl4->daddr = (param->replyopts.opt.opt.srr ?
395 param->replyopts.opt.opt.faddr : iph->saddr);
396 fl4->saddr = saddr;
397 fl4->flowi4_tos = RT_TOS(tos);
398 fl4->flowi4_proto = IPPROTO_ICMP;
399 fl4->fl4_icmp_type = type;
400 fl4->fl4_icmp_code = code;
401 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
402 rt = __ip_route_output_key(net, fl4);
403 if (IS_ERR(rt))
404 return rt;
405
406 /* No need to clone since we're just using its address. */
407 rt2 = rt;
408
409 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
410 flowi4_to_flowi(fl4), NULL, 0);
411 if (!IS_ERR(rt)) {
412 if (rt != rt2)
413 return rt;
414 } else if (PTR_ERR(rt) == -EPERM) {
415 rt = NULL;
416 } else
417 return rt;
418
419 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
420 if (err)
421 goto relookup_failed;
422
423 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
424 rt2 = __ip_route_output_key(net, &fl4_dec);
425 if (IS_ERR(rt2))
426 err = PTR_ERR(rt2);
427 } else {
428 struct flowi4 fl4_2 = {};
429 unsigned long orefdst;
430
431 fl4_2.daddr = fl4_dec.saddr;
432 rt2 = ip_route_output_key(net, &fl4_2);
433 if (IS_ERR(rt2)) {
434 err = PTR_ERR(rt2);
435 goto relookup_failed;
436 }
437 /* Ugh! */
438 orefdst = skb_in->_skb_refdst; /* save old refdst */
439 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
440 RT_TOS(tos), rt2->dst.dev);
441
442 dst_release(&rt2->dst);
443 rt2 = skb_rtable(skb_in);
444 skb_in->_skb_refdst = orefdst; /* restore old refdst */
445 }
446
447 if (err)
448 goto relookup_failed;
449
450 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
451 flowi4_to_flowi(&fl4_dec), NULL,
452 XFRM_LOOKUP_ICMP);
453 if (!IS_ERR(rt2)) {
454 dst_release(&rt->dst);
455 memcpy(fl4, &fl4_dec, sizeof(*fl4));
456 rt = rt2;
457 } else if (PTR_ERR(rt2) == -EPERM) {
458 if (rt)
459 dst_release(&rt->dst);
460 return rt2;
461 } else {
462 err = PTR_ERR(rt2);
463 goto relookup_failed;
464 }
465 return rt;
466
467relookup_failed:
468 if (rt)
469 return rt;
470 return ERR_PTR(err);
471}
472
473/*
474 * Send an ICMP message in response to a situation
475 *
476 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
477 * MAY send more (we do).
478 * MUST NOT change this header information.
479 * MUST NOT reply to a multicast/broadcast IP address.
480 * MUST NOT reply to a multicast/broadcast MAC address.
481 * MUST reply to only the first fragment.
482 */
483
484void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
485{
486 struct iphdr *iph;
487 int room;
488 struct icmp_bxm *icmp_param;
489 struct rtable *rt = skb_rtable(skb_in);
490 struct ipcm_cookie ipc;
491 struct flowi4 fl4;
492 __be32 saddr;
493 u8 tos;
494 struct net *net;
495 struct sock *sk;
496
497 if (!rt)
498 goto out;
499 net = dev_net(rt->dst.dev);
500
501 /*
502 * Find the original header. It is expected to be valid, of course.
503 * Check this, icmp_send is called from the most obscure devices
504 * sometimes.
505 */
506 iph = ip_hdr(skb_in);
507
508 if ((u8 *)iph < skb_in->head ||
509 (skb_network_header(skb_in) + sizeof(*iph)) >
510 skb_tail_pointer(skb_in))
511 goto out;
512
513 /*
514 * No replies to physical multicast/broadcast
515 */
516 if (skb_in->pkt_type != PACKET_HOST)
517 goto out;
518
519 /*
520 * Now check at the protocol level
521 */
522 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
523 goto out;
524
525 /*
526 * Only reply to fragment 0. We byte re-order the constant
527 * mask for efficiency.
528 */
529 if (iph->frag_off & htons(IP_OFFSET))
530 goto out;
531
532 /*
533 * If we send an ICMP error to an ICMP error a mess would result..
534 */
535 if (icmp_pointers[type].error) {
536 /*
537 * We are an error, check if we are replying to an
538 * ICMP error
539 */
540 if (iph->protocol == IPPROTO_ICMP) {
541 u8 _inner_type, *itp;
542
543 itp = skb_header_pointer(skb_in,
544 skb_network_header(skb_in) +
545 (iph->ihl << 2) +
546 offsetof(struct icmphdr,
547 type) -
548 skb_in->data,
549 sizeof(_inner_type),
550 &_inner_type);
551 if (itp == NULL)
552 goto out;
553
554 /*
555 * Assume any unknown ICMP type is an error. This
556 * isn't specified by the RFC, but think about it..
557 */
558 if (*itp > NR_ICMP_TYPES ||
559 icmp_pointers[*itp].error)
560 goto out;
561 }
562 }
563
564 icmp_param = kmalloc(sizeof(*icmp_param), GFP_ATOMIC);
565 if (!icmp_param)
566 return;
567
568 sk = icmp_xmit_lock(net);
569 if (sk == NULL)
570 goto out_free;
571
572 /*
573 * Construct source address and options.
574 */
575
576 saddr = iph->daddr;
577 if (!(rt->rt_flags & RTCF_LOCAL)) {
578 struct net_device *dev = NULL;
579
580 rcu_read_lock();
581 if (rt_is_input_route(rt) &&
582 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
583 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
584
585 if (dev)
586 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
587 else
588 saddr = 0;
589 rcu_read_unlock();
590 }
591
592 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
593 IPTOS_PREC_INTERNETCONTROL) :
594 iph->tos;
595
596 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb_in))
597 goto out_unlock;
598
599
600 /*
601 * Prepare data for ICMP header.
602 */
603
604 icmp_param->data.icmph.type = type;
605 icmp_param->data.icmph.code = code;
606 icmp_param->data.icmph.un.gateway = info;
607 icmp_param->data.icmph.checksum = 0;
608 icmp_param->skb = skb_in;
609 icmp_param->offset = skb_network_offset(skb_in);
610 inet_sk(sk)->tos = tos;
611 ipc.addr = iph->saddr;
612 ipc.opt = &icmp_param->replyopts.opt;
613 ipc.tx_flags = 0;
614 ipc.ttl = 0;
615 ipc.tos = -1;
616
617 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
618 type, code, icmp_param);
619 if (IS_ERR(rt))
620 goto out_unlock;
621
622 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
623 goto ende;
624
625 /* RFC says return as much as we can without exceeding 576 bytes. */
626
627 room = dst_mtu(&rt->dst);
628 if (room > 576)
629 room = 576;
630 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.opt.optlen;
631 room -= sizeof(struct icmphdr);
632
633 icmp_param->data_len = skb_in->len - icmp_param->offset;
634 if (icmp_param->data_len > room)
635 icmp_param->data_len = room;
636 icmp_param->head_len = sizeof(struct icmphdr);
637
638 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
639ende:
640 ip_rt_put(rt);
641out_unlock:
642 icmp_xmit_unlock(sk);
643out_free:
644 kfree(icmp_param);
645out:;
646}
647EXPORT_SYMBOL(icmp_send);
648
649
650static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
651{
652 const struct iphdr *iph = (const struct iphdr *) skb->data;
653 const struct net_protocol *ipprot;
654 int protocol = iph->protocol;
655
656 /* Checkin full IP header plus 8 bytes of protocol to
657 * avoid additional coding at protocol handlers.
658 */
659 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
660 return;
661
662 raw_icmp_error(skb, protocol, info);
663
664 rcu_read_lock();
665 ipprot = rcu_dereference(inet_protos[protocol]);
666 if (ipprot && ipprot->err_handler)
667 ipprot->err_handler(skb, info);
668 rcu_read_unlock();
669}
670
671static bool icmp_tag_validation(int proto)
672{
673 bool ok;
674
675 rcu_read_lock();
676 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
677 rcu_read_unlock();
678 return ok;
679}
680
681/*
682 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, ICMP_QUENCH, and
683 * ICMP_PARAMETERPROB.
684 */
685
686static void icmp_unreach(struct sk_buff *skb)
687{
688 const struct iphdr *iph;
689 struct icmphdr *icmph;
690 struct net *net;
691 u32 info = 0;
692
693 net = dev_net(skb_dst(skb)->dev);
694
695 /*
696 * Incomplete header ?
697 * Only checks for the IP header, there should be an
698 * additional check for longer headers in upper levels.
699 */
700
701 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
702 goto out_err;
703
704 icmph = icmp_hdr(skb);
705 iph = (const struct iphdr *)skb->data;
706
707 if (iph->ihl < 5) /* Mangled header, drop. */
708 goto out_err;
709
710 if (icmph->type == ICMP_DEST_UNREACH) {
711 switch (icmph->code & 15) {
712 case ICMP_NET_UNREACH:
713 case ICMP_HOST_UNREACH:
714 case ICMP_PROT_UNREACH:
715 case ICMP_PORT_UNREACH:
716 break;
717 case ICMP_FRAG_NEEDED:
718 /* for documentation of the ip_no_pmtu_disc
719 * values please see
720 * Documentation/networking/ip-sysctl.txt
721 */
722 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
723 default:
724 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
725 &iph->daddr);
726 break;
727 case 2:
728 goto out;
729 case 3:
730 if (!icmp_tag_validation(iph->protocol))
731 goto out;
732 /* fall through */
733 case 0:
734 info = ntohs(icmph->un.frag.mtu);
735 if (!info)
736 goto out;
737 }
738 break;
739 case ICMP_SR_FAILED:
740 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
741 &iph->daddr);
742 break;
743 default:
744 break;
745 }
746 if (icmph->code > NR_ICMP_UNREACH)
747 goto out;
748 } else if (icmph->type == ICMP_PARAMETERPROB)
749 info = ntohl(icmph->un.gateway) >> 24;
750
751 /*
752 * Throw it at our lower layers
753 *
754 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
755 * header.
756 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
757 * transport layer.
758 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
759 * transport layer.
760 */
761
762 /*
763 * Check the other end isn't violating RFC 1122. Some routers send
764 * bogus responses to broadcast frames. If you see this message
765 * first check your netmask matches at both ends, if it does then
766 * get the other vendor to fix their kit.
767 */
768
769 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
770 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
771 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
772 &ip_hdr(skb)->saddr,
773 icmph->type, icmph->code,
774 &iph->daddr, skb->dev->name);
775 goto out;
776 }
777
778 icmp_socket_deliver(skb, info);
779
780out:
781 return;
782out_err:
783 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
784 goto out;
785}
786
787
788/*
789 * Handle ICMP_REDIRECT.
790 */
791
792static void icmp_redirect(struct sk_buff *skb)
793{
794 if (skb->len < sizeof(struct iphdr)) {
795 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
796 return;
797 }
798
799 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
800 return;
801
802 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
803}
804
805/*
806 * Handle ICMP_ECHO ("ping") requests.
807 *
808 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
809 * requests.
810 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
811 * included in the reply.
812 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
813 * echo requests, MUST have default=NOT.
814 * See also WRT handling of options once they are done and working.
815 */
816
817static void icmp_echo(struct sk_buff *skb)
818{
819 struct net *net;
820
821 net = dev_net(skb_dst(skb)->dev);
822 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
823 struct icmp_bxm icmp_param;
824
825 icmp_param.data.icmph = *icmp_hdr(skb);
826 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
827 icmp_param.skb = skb;
828 icmp_param.offset = 0;
829 icmp_param.data_len = skb->len;
830 icmp_param.head_len = sizeof(struct icmphdr);
831 icmp_reply(&icmp_param, skb);
832 }
833}
834
835/*
836 * Handle ICMP Timestamp requests.
837 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
838 * SHOULD be in the kernel for minimum random latency.
839 * MUST be accurate to a few minutes.
840 * MUST be updated at least at 15Hz.
841 */
842static void icmp_timestamp(struct sk_buff *skb)
843{
844 struct timespec tv;
845 struct icmp_bxm icmp_param;
846 /*
847 * Too short.
848 */
849 if (skb->len < 4)
850 goto out_err;
851
852 /*
853 * Fill in the current time as ms since midnight UT:
854 */
855 getnstimeofday(&tv);
856 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
857 tv.tv_nsec / NSEC_PER_MSEC);
858 icmp_param.data.times[2] = icmp_param.data.times[1];
859 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
860 BUG();
861 icmp_param.data.icmph = *icmp_hdr(skb);
862 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
863 icmp_param.data.icmph.code = 0;
864 icmp_param.skb = skb;
865 icmp_param.offset = 0;
866 icmp_param.data_len = 0;
867 icmp_param.head_len = sizeof(struct icmphdr) + 12;
868 icmp_reply(&icmp_param, skb);
869out:
870 return;
871out_err:
872 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
873 goto out;
874}
875
876static void icmp_discard(struct sk_buff *skb)
877{
878}
879
880/*
881 * Deal with incoming ICMP packets.
882 */
883int icmp_rcv(struct sk_buff *skb)
884{
885 struct icmphdr *icmph;
886 struct rtable *rt = skb_rtable(skb);
887 struct net *net = dev_net(rt->dst.dev);
888
889 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
890 struct sec_path *sp = skb_sec_path(skb);
891 int nh;
892
893 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
894 XFRM_STATE_ICMP))
895 goto drop;
896
897 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
898 goto drop;
899
900 nh = skb_network_offset(skb);
901 skb_set_network_header(skb, sizeof(*icmph));
902
903 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
904 goto drop;
905
906 skb_set_network_header(skb, nh);
907 }
908
909 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
910
911 switch (skb->ip_summed) {
912 case CHECKSUM_COMPLETE:
913 if (!csum_fold(skb->csum))
914 break;
915 /* fall through */
916 case CHECKSUM_NONE:
917 skb->csum = 0;
918 if (__skb_checksum_complete(skb))
919 goto csum_error;
920 }
921
922 if (!pskb_pull(skb, sizeof(*icmph)))
923 goto error;
924
925 icmph = icmp_hdr(skb);
926
927 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
928 /*
929 * 18 is the highest 'known' ICMP type. Anything else is a mystery
930 *
931 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
932 * discarded.
933 */
934 if (icmph->type > NR_ICMP_TYPES)
935 goto error;
936
937
938 /*
939 * Parse the ICMP message
940 */
941
942 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
943 /*
944 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
945 * silently ignored (we let user decide with a sysctl).
946 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
947 * discarded if to broadcast/multicast.
948 */
949 if ((icmph->type == ICMP_ECHO ||
950 icmph->type == ICMP_TIMESTAMP) &&
951 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
952 goto error;
953 }
954 if (icmph->type != ICMP_ECHO &&
955 icmph->type != ICMP_TIMESTAMP &&
956 icmph->type != ICMP_ADDRESS &&
957 icmph->type != ICMP_ADDRESSREPLY) {
958 goto error;
959 }
960 }
961
962 icmp_pointers[icmph->type].handler(skb);
963
964drop:
965 kfree_skb(skb);
966 return 0;
967csum_error:
968 ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
969error:
970 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
971 goto drop;
972}
973
974void icmp_err(struct sk_buff *skb, u32 info)
975{
976 struct iphdr *iph = (struct iphdr *)skb->data;
977 int offset = iph->ihl<<2;
978 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
979 int type = icmp_hdr(skb)->type;
980 int code = icmp_hdr(skb)->code;
981 struct net *net = dev_net(skb->dev);
982
983 /*
984 * Use ping_err to handle all icmp errors except those
985 * triggered by ICMP_ECHOREPLY which sent from kernel.
986 */
987 if (icmph->type != ICMP_ECHOREPLY) {
988 ping_err(skb, offset, info);
989 return;
990 }
991
992 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
993 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
994 else if (type == ICMP_REDIRECT)
995 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
996}
997
998/*
999 * This table is the definition of how we handle ICMP.
1000 */
1001static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1002 [ICMP_ECHOREPLY] = {
1003 .handler = ping_rcv,
1004 },
1005 [1] = {
1006 .handler = icmp_discard,
1007 .error = 1,
1008 },
1009 [2] = {
1010 .handler = icmp_discard,
1011 .error = 1,
1012 },
1013 [ICMP_DEST_UNREACH] = {
1014 .handler = icmp_unreach,
1015 .error = 1,
1016 },
1017 [ICMP_SOURCE_QUENCH] = {
1018 .handler = icmp_unreach,
1019 .error = 1,
1020 },
1021 [ICMP_REDIRECT] = {
1022 .handler = icmp_redirect,
1023 .error = 1,
1024 },
1025 [6] = {
1026 .handler = icmp_discard,
1027 .error = 1,
1028 },
1029 [7] = {
1030 .handler = icmp_discard,
1031 .error = 1,
1032 },
1033 [ICMP_ECHO] = {
1034 .handler = icmp_echo,
1035 },
1036 [9] = {
1037 .handler = icmp_discard,
1038 .error = 1,
1039 },
1040 [10] = {
1041 .handler = icmp_discard,
1042 .error = 1,
1043 },
1044 [ICMP_TIME_EXCEEDED] = {
1045 .handler = icmp_unreach,
1046 .error = 1,
1047 },
1048 [ICMP_PARAMETERPROB] = {
1049 .handler = icmp_unreach,
1050 .error = 1,
1051 },
1052 [ICMP_TIMESTAMP] = {
1053 .handler = icmp_timestamp,
1054 },
1055 [ICMP_TIMESTAMPREPLY] = {
1056 .handler = icmp_discard,
1057 },
1058 [ICMP_INFO_REQUEST] = {
1059 .handler = icmp_discard,
1060 },
1061 [ICMP_INFO_REPLY] = {
1062 .handler = icmp_discard,
1063 },
1064 [ICMP_ADDRESS] = {
1065 .handler = icmp_discard,
1066 },
1067 [ICMP_ADDRESSREPLY] = {
1068 .handler = icmp_discard,
1069 },
1070};
1071
1072static void __net_exit icmp_sk_exit(struct net *net)
1073{
1074 int i;
1075
1076 for_each_possible_cpu(i)
1077 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1078 kfree(net->ipv4.icmp_sk);
1079 net->ipv4.icmp_sk = NULL;
1080}
1081
1082static int __net_init icmp_sk_init(struct net *net)
1083{
1084 int i, err;
1085
1086 net->ipv4.icmp_sk =
1087 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1088 if (net->ipv4.icmp_sk == NULL)
1089 return -ENOMEM;
1090
1091 for_each_possible_cpu(i) {
1092 struct sock *sk;
1093
1094 err = inet_ctl_sock_create(&sk, PF_INET,
1095 SOCK_RAW, IPPROTO_ICMP, net);
1096 if (err < 0)
1097 goto fail;
1098
1099 net->ipv4.icmp_sk[i] = sk;
1100
1101 /* Enough space for 2 64K ICMP packets, including
1102 * sk_buff/skb_shared_info struct overhead.
1103 */
1104 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1105
1106 /*
1107 * Speedup sock_wfree()
1108 */
1109 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1110 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1111 }
1112
1113 /* Control parameters for ECHO replies. */
1114 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1115 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1116
1117 /* Control parameter - ignore bogus broadcast responses? */
1118 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1119
1120 /*
1121 * Configurable global rate limit.
1122 *
1123 * ratelimit defines tokens/packet consumed for dst->rate_token
1124 * bucket ratemask defines which icmp types are ratelimited by
1125 * setting it's bit position.
1126 *
1127 * default:
1128 * dest unreachable (3), source quench (4),
1129 * time exceeded (11), parameter problem (12)
1130 */
1131
1132 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1133 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1134 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1135
1136 return 0;
1137
1138fail:
1139 for_each_possible_cpu(i)
1140 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1141 kfree(net->ipv4.icmp_sk);
1142 return err;
1143}
1144
1145static struct pernet_operations __net_initdata icmp_sk_ops = {
1146 .init = icmp_sk_init,
1147 .exit = icmp_sk_exit,
1148};
1149
1150int __init icmp_init(void)
1151{
1152 return register_pernet_subsys(&icmp_sk_ops);
1153}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET3: Implementation of the ICMP protocol layer.
4 *
5 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Some of the function names and the icmp unreach table for this
8 * module were derived from [icmp.c 1.0.11 06/02/93] by
9 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10 * Other than that this module is a complete rewrite.
11 *
12 * Fixes:
13 * Clemens Fruhwirth : introduce global icmp rate limiting
14 * with icmp type masking ability instead
15 * of broken per type icmp timeouts.
16 * Mike Shaver : RFC1122 checks.
17 * Alan Cox : Multicast ping reply as self.
18 * Alan Cox : Fix atomicity lockup in ip_build_xmit
19 * call.
20 * Alan Cox : Added 216,128 byte paths to the MTU
21 * code.
22 * Martin Mares : RFC1812 checks.
23 * Martin Mares : Can be configured to follow redirects
24 * if acting as a router _without_ a
25 * routing protocol (RFC 1812).
26 * Martin Mares : Echo requests may be configured to
27 * be ignored (RFC 1812).
28 * Martin Mares : Limitation of ICMP error message
29 * transmit rate (RFC 1812).
30 * Martin Mares : TOS and Precedence set correctly
31 * (RFC 1812).
32 * Martin Mares : Now copying as much data from the
33 * original packet as we can without
34 * exceeding 576 bytes (RFC 1812).
35 * Willy Konynenberg : Transparent proxying support.
36 * Keith Owens : RFC1191 correction for 4.2BSD based
37 * path MTU bug.
38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
39 * valid (RFC 1812).
40 * Andi Kleen : Check all packet lengths properly
41 * and moved all kfree_skb() up to
42 * icmp_rcv.
43 * Andi Kleen : Move the rate limit bookkeeping
44 * into the dest entry and use a token
45 * bucket filter (thanks to ANK). Make
46 * the rates sysctl configurable.
47 * Yu Tianli : Fixed two ugly bugs in icmp_send
48 * - IP option length was accounted wrongly
49 * - ICMP header length was not accounted
50 * at all.
51 * Tristan Greaves : Added sysctl option to ignore bogus
52 * broadcast responses from broken routers.
53 *
54 * To Fix:
55 *
56 * - Should use skb_pull() instead of all the manual checking.
57 * This would also greatly simply some upper layer error handlers. --AK
58 */
59
60#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61
62#include <linux/module.h>
63#include <linux/types.h>
64#include <linux/jiffies.h>
65#include <linux/kernel.h>
66#include <linux/fcntl.h>
67#include <linux/socket.h>
68#include <linux/in.h>
69#include <linux/inet.h>
70#include <linux/inetdevice.h>
71#include <linux/netdevice.h>
72#include <linux/string.h>
73#include <linux/netfilter_ipv4.h>
74#include <linux/slab.h>
75#include <net/snmp.h>
76#include <net/ip.h>
77#include <net/route.h>
78#include <net/protocol.h>
79#include <net/icmp.h>
80#include <net/tcp.h>
81#include <net/udp.h>
82#include <net/raw.h>
83#include <net/ping.h>
84#include <linux/skbuff.h>
85#include <net/sock.h>
86#include <linux/errno.h>
87#include <linux/timer.h>
88#include <linux/init.h>
89#include <linux/uaccess.h>
90#include <net/checksum.h>
91#include <net/xfrm.h>
92#include <net/inet_common.h>
93#include <net/ip_fib.h>
94#include <net/l3mdev.h>
95
96/*
97 * Build xmit assembly blocks
98 */
99
100struct icmp_bxm {
101 struct sk_buff *skb;
102 int offset;
103 int data_len;
104
105 struct {
106 struct icmphdr icmph;
107 __be32 times[3];
108 } data;
109 int head_len;
110 struct ip_options_data replyopts;
111};
112
113/* An array of errno for error messages from dest unreach. */
114/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115
116const struct icmp_err icmp_err_convert[] = {
117 {
118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
119 .fatal = 0,
120 },
121 {
122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
123 .fatal = 0,
124 },
125 {
126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
127 .fatal = 1,
128 },
129 {
130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
131 .fatal = 1,
132 },
133 {
134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
135 .fatal = 0,
136 },
137 {
138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
139 .fatal = 0,
140 },
141 {
142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
143 .fatal = 1,
144 },
145 {
146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
147 .fatal = 1,
148 },
149 {
150 .errno = ENONET, /* ICMP_HOST_ISOLATED */
151 .fatal = 1,
152 },
153 {
154 .errno = ENETUNREACH, /* ICMP_NET_ANO */
155 .fatal = 1,
156 },
157 {
158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
163 .fatal = 0,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
167 .fatal = 0,
168 },
169 {
170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
171 .fatal = 1,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
175 .fatal = 1,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
179 .fatal = 1,
180 },
181};
182EXPORT_SYMBOL(icmp_err_convert);
183
184/*
185 * ICMP control array. This specifies what to do with each ICMP.
186 */
187
188struct icmp_control {
189 bool (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191};
192
193static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194
195/*
196 * The ICMP socket(s). This is the most convenient way to flow control
197 * our ICMP output as well as maintain a clean interface throughout
198 * all layers. All Socketless IP sends will soon be gone.
199 *
200 * On SMP we have one ICMP socket per-cpu.
201 */
202static struct sock *icmp_sk(struct net *net)
203{
204 return this_cpu_read(*net->ipv4.icmp_sk);
205}
206
207/* Called with BH disabled */
208static inline struct sock *icmp_xmit_lock(struct net *net)
209{
210 struct sock *sk;
211
212 sk = icmp_sk(net);
213
214 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 /* This can happen if the output path signals a
216 * dst_link_failure() for an outgoing ICMP packet.
217 */
218 return NULL;
219 }
220 return sk;
221}
222
223static inline void icmp_xmit_unlock(struct sock *sk)
224{
225 spin_unlock(&sk->sk_lock.slock);
226}
227
228int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229int sysctl_icmp_msgs_burst __read_mostly = 50;
230
231static struct {
232 spinlock_t lock;
233 u32 credit;
234 u32 stamp;
235} icmp_global = {
236 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
237};
238
239/**
240 * icmp_global_allow - Are we allowed to send one more ICMP message ?
241 *
242 * Uses a token bucket to limit our ICMP messages to sysctl_icmp_msgs_per_sec.
243 * Returns false if we reached the limit and can not send another packet.
244 * Note: called with BH disabled
245 */
246bool icmp_global_allow(void)
247{
248 u32 credit, delta, incr = 0, now = (u32)jiffies;
249 bool rc = false;
250
251 /* Check if token bucket is empty and cannot be refilled
252 * without taking the spinlock. The READ_ONCE() are paired
253 * with the following WRITE_ONCE() in this same function.
254 */
255 if (!READ_ONCE(icmp_global.credit)) {
256 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
257 if (delta < HZ / 50)
258 return false;
259 }
260
261 spin_lock(&icmp_global.lock);
262 delta = min_t(u32, now - icmp_global.stamp, HZ);
263 if (delta >= HZ / 50) {
264 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
265 if (incr)
266 WRITE_ONCE(icmp_global.stamp, now);
267 }
268 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
269 if (credit) {
270 credit--;
271 rc = true;
272 }
273 WRITE_ONCE(icmp_global.credit, credit);
274 spin_unlock(&icmp_global.lock);
275 return rc;
276}
277EXPORT_SYMBOL(icmp_global_allow);
278
279static bool icmpv4_mask_allow(struct net *net, int type, int code)
280{
281 if (type > NR_ICMP_TYPES)
282 return true;
283
284 /* Don't limit PMTU discovery. */
285 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
286 return true;
287
288 /* Limit if icmp type is enabled in ratemask. */
289 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
290 return true;
291
292 return false;
293}
294
295static bool icmpv4_global_allow(struct net *net, int type, int code)
296{
297 if (icmpv4_mask_allow(net, type, code))
298 return true;
299
300 if (icmp_global_allow())
301 return true;
302
303 return false;
304}
305
306/*
307 * Send an ICMP frame.
308 */
309
310static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
311 struct flowi4 *fl4, int type, int code)
312{
313 struct dst_entry *dst = &rt->dst;
314 struct inet_peer *peer;
315 bool rc = true;
316 int vif;
317
318 if (icmpv4_mask_allow(net, type, code))
319 goto out;
320
321 /* No rate limit on loopback */
322 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
323 goto out;
324
325 vif = l3mdev_master_ifindex(dst->dev);
326 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
327 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
328 if (peer)
329 inet_putpeer(peer);
330out:
331 return rc;
332}
333
334/*
335 * Maintain the counters used in the SNMP statistics for outgoing ICMP
336 */
337void icmp_out_count(struct net *net, unsigned char type)
338{
339 ICMPMSGOUT_INC_STATS(net, type);
340 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
341}
342
343/*
344 * Checksum each fragment, and on the first include the headers and final
345 * checksum.
346 */
347static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
348 struct sk_buff *skb)
349{
350 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
351 __wsum csum;
352
353 csum = skb_copy_and_csum_bits(icmp_param->skb,
354 icmp_param->offset + offset,
355 to, len, 0);
356
357 skb->csum = csum_block_add(skb->csum, csum, odd);
358 if (icmp_pointers[icmp_param->data.icmph.type].error)
359 nf_ct_attach(skb, icmp_param->skb);
360 return 0;
361}
362
363static void icmp_push_reply(struct icmp_bxm *icmp_param,
364 struct flowi4 *fl4,
365 struct ipcm_cookie *ipc, struct rtable **rt)
366{
367 struct sock *sk;
368 struct sk_buff *skb;
369
370 sk = icmp_sk(dev_net((*rt)->dst.dev));
371 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
372 icmp_param->data_len+icmp_param->head_len,
373 icmp_param->head_len,
374 ipc, rt, MSG_DONTWAIT) < 0) {
375 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
376 ip_flush_pending_frames(sk);
377 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
378 struct icmphdr *icmph = icmp_hdr(skb);
379 __wsum csum = 0;
380 struct sk_buff *skb1;
381
382 skb_queue_walk(&sk->sk_write_queue, skb1) {
383 csum = csum_add(csum, skb1->csum);
384 }
385 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
386 (char *)icmph,
387 icmp_param->head_len, csum);
388 icmph->checksum = csum_fold(csum);
389 skb->ip_summed = CHECKSUM_NONE;
390 ip_push_pending_frames(sk, fl4);
391 }
392}
393
394/*
395 * Driving logic for building and sending ICMP messages.
396 */
397
398static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
399{
400 struct ipcm_cookie ipc;
401 struct rtable *rt = skb_rtable(skb);
402 struct net *net = dev_net(rt->dst.dev);
403 struct flowi4 fl4;
404 struct sock *sk;
405 struct inet_sock *inet;
406 __be32 daddr, saddr;
407 u32 mark = IP4_REPLY_MARK(net, skb->mark);
408 int type = icmp_param->data.icmph.type;
409 int code = icmp_param->data.icmph.code;
410
411 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
412 return;
413
414 /* Needed by both icmp_global_allow and icmp_xmit_lock */
415 local_bh_disable();
416
417 /* global icmp_msgs_per_sec */
418 if (!icmpv4_global_allow(net, type, code))
419 goto out_bh_enable;
420
421 sk = icmp_xmit_lock(net);
422 if (!sk)
423 goto out_bh_enable;
424 inet = inet_sk(sk);
425
426 icmp_param->data.icmph.checksum = 0;
427
428 ipcm_init(&ipc);
429 inet->tos = ip_hdr(skb)->tos;
430 ipc.sockc.mark = mark;
431 daddr = ipc.addr = ip_hdr(skb)->saddr;
432 saddr = fib_compute_spec_dst(skb);
433
434 if (icmp_param->replyopts.opt.opt.optlen) {
435 ipc.opt = &icmp_param->replyopts.opt;
436 if (ipc.opt->opt.srr)
437 daddr = icmp_param->replyopts.opt.opt.faddr;
438 }
439 memset(&fl4, 0, sizeof(fl4));
440 fl4.daddr = daddr;
441 fl4.saddr = saddr;
442 fl4.flowi4_mark = mark;
443 fl4.flowi4_uid = sock_net_uid(net, NULL);
444 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
445 fl4.flowi4_proto = IPPROTO_ICMP;
446 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
447 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
448 rt = ip_route_output_key(net, &fl4);
449 if (IS_ERR(rt))
450 goto out_unlock;
451 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
452 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
453 ip_rt_put(rt);
454out_unlock:
455 icmp_xmit_unlock(sk);
456out_bh_enable:
457 local_bh_enable();
458}
459
460static struct rtable *icmp_route_lookup(struct net *net,
461 struct flowi4 *fl4,
462 struct sk_buff *skb_in,
463 const struct iphdr *iph,
464 __be32 saddr, u8 tos, u32 mark,
465 int type, int code,
466 struct icmp_bxm *param)
467{
468 struct rtable *rt, *rt2;
469 struct flowi4 fl4_dec;
470 int err;
471
472 memset(fl4, 0, sizeof(*fl4));
473 fl4->daddr = (param->replyopts.opt.opt.srr ?
474 param->replyopts.opt.opt.faddr : iph->saddr);
475 fl4->saddr = saddr;
476 fl4->flowi4_mark = mark;
477 fl4->flowi4_uid = sock_net_uid(net, NULL);
478 fl4->flowi4_tos = RT_TOS(tos);
479 fl4->flowi4_proto = IPPROTO_ICMP;
480 fl4->fl4_icmp_type = type;
481 fl4->fl4_icmp_code = code;
482 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
483
484 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
485 rt = ip_route_output_key_hash(net, fl4, skb_in);
486 if (IS_ERR(rt))
487 return rt;
488
489 /* No need to clone since we're just using its address. */
490 rt2 = rt;
491
492 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
493 flowi4_to_flowi(fl4), NULL, 0);
494 if (!IS_ERR(rt)) {
495 if (rt != rt2)
496 return rt;
497 } else if (PTR_ERR(rt) == -EPERM) {
498 rt = NULL;
499 } else
500 return rt;
501
502 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
503 if (err)
504 goto relookup_failed;
505
506 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
507 fl4_dec.saddr) == RTN_LOCAL) {
508 rt2 = __ip_route_output_key(net, &fl4_dec);
509 if (IS_ERR(rt2))
510 err = PTR_ERR(rt2);
511 } else {
512 struct flowi4 fl4_2 = {};
513 unsigned long orefdst;
514
515 fl4_2.daddr = fl4_dec.saddr;
516 rt2 = ip_route_output_key(net, &fl4_2);
517 if (IS_ERR(rt2)) {
518 err = PTR_ERR(rt2);
519 goto relookup_failed;
520 }
521 /* Ugh! */
522 orefdst = skb_in->_skb_refdst; /* save old refdst */
523 skb_dst_set(skb_in, NULL);
524 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
525 RT_TOS(tos), rt2->dst.dev);
526
527 dst_release(&rt2->dst);
528 rt2 = skb_rtable(skb_in);
529 skb_in->_skb_refdst = orefdst; /* restore old refdst */
530 }
531
532 if (err)
533 goto relookup_failed;
534
535 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
536 flowi4_to_flowi(&fl4_dec), NULL,
537 XFRM_LOOKUP_ICMP);
538 if (!IS_ERR(rt2)) {
539 dst_release(&rt->dst);
540 memcpy(fl4, &fl4_dec, sizeof(*fl4));
541 rt = rt2;
542 } else if (PTR_ERR(rt2) == -EPERM) {
543 if (rt)
544 dst_release(&rt->dst);
545 return rt2;
546 } else {
547 err = PTR_ERR(rt2);
548 goto relookup_failed;
549 }
550 return rt;
551
552relookup_failed:
553 if (rt)
554 return rt;
555 return ERR_PTR(err);
556}
557
558/*
559 * Send an ICMP message in response to a situation
560 *
561 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
562 * MAY send more (we do).
563 * MUST NOT change this header information.
564 * MUST NOT reply to a multicast/broadcast IP address.
565 * MUST NOT reply to a multicast/broadcast MAC address.
566 * MUST reply to only the first fragment.
567 */
568
569void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
570 const struct ip_options *opt)
571{
572 struct iphdr *iph;
573 int room;
574 struct icmp_bxm icmp_param;
575 struct rtable *rt = skb_rtable(skb_in);
576 struct ipcm_cookie ipc;
577 struct flowi4 fl4;
578 __be32 saddr;
579 u8 tos;
580 u32 mark;
581 struct net *net;
582 struct sock *sk;
583
584 if (!rt)
585 goto out;
586
587 if (rt->dst.dev)
588 net = dev_net(rt->dst.dev);
589 else if (skb_in->dev)
590 net = dev_net(skb_in->dev);
591 else
592 goto out;
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, iph->saddr,
687 RT_SCOPE_LINK);
688 else
689 saddr = 0;
690 rcu_read_unlock();
691 }
692
693 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
694 IPTOS_PREC_INTERNETCONTROL) :
695 iph->tos;
696 mark = IP4_REPLY_MARK(net, skb_in->mark);
697
698 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
699 goto out_unlock;
700
701
702 /*
703 * Prepare data for ICMP header.
704 */
705
706 icmp_param.data.icmph.type = type;
707 icmp_param.data.icmph.code = code;
708 icmp_param.data.icmph.un.gateway = info;
709 icmp_param.data.icmph.checksum = 0;
710 icmp_param.skb = skb_in;
711 icmp_param.offset = skb_network_offset(skb_in);
712 inet_sk(sk)->tos = tos;
713 ipcm_init(&ipc);
714 ipc.addr = iph->saddr;
715 ipc.opt = &icmp_param.replyopts.opt;
716 ipc.sockc.mark = mark;
717
718 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
719 type, code, &icmp_param);
720 if (IS_ERR(rt))
721 goto out_unlock;
722
723 /* peer icmp_ratelimit */
724 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
725 goto ende;
726
727 /* RFC says return as much as we can without exceeding 576 bytes. */
728
729 room = dst_mtu(&rt->dst);
730 if (room > 576)
731 room = 576;
732 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
733 room -= sizeof(struct icmphdr);
734
735 icmp_param.data_len = skb_in->len - icmp_param.offset;
736 if (icmp_param.data_len > room)
737 icmp_param.data_len = room;
738 icmp_param.head_len = sizeof(struct icmphdr);
739
740 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
741ende:
742 ip_rt_put(rt);
743out_unlock:
744 icmp_xmit_unlock(sk);
745out_bh_enable:
746 local_bh_enable();
747out:;
748}
749EXPORT_SYMBOL(__icmp_send);
750
751#if IS_ENABLED(CONFIG_NF_NAT)
752#include <net/netfilter/nf_conntrack.h>
753void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
754{
755 struct sk_buff *cloned_skb = NULL;
756 enum ip_conntrack_info ctinfo;
757 struct nf_conn *ct;
758 __be32 orig_ip;
759
760 ct = nf_ct_get(skb_in, &ctinfo);
761 if (!ct || !(ct->status & IPS_SRC_NAT)) {
762 icmp_send(skb_in, type, code, info);
763 return;
764 }
765
766 if (skb_shared(skb_in))
767 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
768
769 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
770 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
771 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
772 skb_network_offset(skb_in) + sizeof(struct iphdr))))
773 goto out;
774
775 orig_ip = ip_hdr(skb_in)->saddr;
776 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
777 icmp_send(skb_in, type, code, info);
778 ip_hdr(skb_in)->saddr = orig_ip;
779out:
780 consume_skb(cloned_skb);
781}
782EXPORT_SYMBOL(icmp_ndo_send);
783#endif
784
785static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
786{
787 const struct iphdr *iph = (const struct iphdr *) skb->data;
788 const struct net_protocol *ipprot;
789 int protocol = iph->protocol;
790
791 /* Checkin full IP header plus 8 bytes of protocol to
792 * avoid additional coding at protocol handlers.
793 */
794 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
795 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
796 return;
797 }
798
799 raw_icmp_error(skb, protocol, info);
800
801 ipprot = rcu_dereference(inet_protos[protocol]);
802 if (ipprot && ipprot->err_handler)
803 ipprot->err_handler(skb, info);
804}
805
806static bool icmp_tag_validation(int proto)
807{
808 bool ok;
809
810 rcu_read_lock();
811 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
812 rcu_read_unlock();
813 return ok;
814}
815
816/*
817 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
818 * ICMP_PARAMETERPROB.
819 */
820
821static bool icmp_unreach(struct sk_buff *skb)
822{
823 const struct iphdr *iph;
824 struct icmphdr *icmph;
825 struct net *net;
826 u32 info = 0;
827
828 net = dev_net(skb_dst(skb)->dev);
829
830 /*
831 * Incomplete header ?
832 * Only checks for the IP header, there should be an
833 * additional check for longer headers in upper levels.
834 */
835
836 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
837 goto out_err;
838
839 icmph = icmp_hdr(skb);
840 iph = (const struct iphdr *)skb->data;
841
842 if (iph->ihl < 5) /* Mangled header, drop. */
843 goto out_err;
844
845 switch (icmph->type) {
846 case ICMP_DEST_UNREACH:
847 switch (icmph->code & 15) {
848 case ICMP_NET_UNREACH:
849 case ICMP_HOST_UNREACH:
850 case ICMP_PROT_UNREACH:
851 case ICMP_PORT_UNREACH:
852 break;
853 case ICMP_FRAG_NEEDED:
854 /* for documentation of the ip_no_pmtu_disc
855 * values please see
856 * Documentation/networking/ip-sysctl.rst
857 */
858 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
859 default:
860 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
861 &iph->daddr);
862 break;
863 case 2:
864 goto out;
865 case 3:
866 if (!icmp_tag_validation(iph->protocol))
867 goto out;
868 fallthrough;
869 case 0:
870 info = ntohs(icmph->un.frag.mtu);
871 }
872 break;
873 case ICMP_SR_FAILED:
874 net_dbg_ratelimited("%pI4: Source Route Failed\n",
875 &iph->daddr);
876 break;
877 default:
878 break;
879 }
880 if (icmph->code > NR_ICMP_UNREACH)
881 goto out;
882 break;
883 case ICMP_PARAMETERPROB:
884 info = ntohl(icmph->un.gateway) >> 24;
885 break;
886 case ICMP_TIME_EXCEEDED:
887 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
888 if (icmph->code == ICMP_EXC_FRAGTIME)
889 goto out;
890 break;
891 }
892
893 /*
894 * Throw it at our lower layers
895 *
896 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
897 * header.
898 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
899 * transport layer.
900 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
901 * transport layer.
902 */
903
904 /*
905 * Check the other end isn't violating RFC 1122. Some routers send
906 * bogus responses to broadcast frames. If you see this message
907 * first check your netmask matches at both ends, if it does then
908 * get the other vendor to fix their kit.
909 */
910
911 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
912 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
913 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
914 &ip_hdr(skb)->saddr,
915 icmph->type, icmph->code,
916 &iph->daddr, skb->dev->name);
917 goto out;
918 }
919
920 icmp_socket_deliver(skb, info);
921
922out:
923 return true;
924out_err:
925 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
926 return false;
927}
928
929
930/*
931 * Handle ICMP_REDIRECT.
932 */
933
934static bool icmp_redirect(struct sk_buff *skb)
935{
936 if (skb->len < sizeof(struct iphdr)) {
937 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
938 return false;
939 }
940
941 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
942 /* there aught to be a stat */
943 return false;
944 }
945
946 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
947 return true;
948}
949
950/*
951 * Handle ICMP_ECHO ("ping") requests.
952 *
953 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
954 * requests.
955 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
956 * included in the reply.
957 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
958 * echo requests, MUST have default=NOT.
959 * See also WRT handling of options once they are done and working.
960 */
961
962static bool icmp_echo(struct sk_buff *skb)
963{
964 struct net *net;
965
966 net = dev_net(skb_dst(skb)->dev);
967 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
968 struct icmp_bxm icmp_param;
969
970 icmp_param.data.icmph = *icmp_hdr(skb);
971 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
972 icmp_param.skb = skb;
973 icmp_param.offset = 0;
974 icmp_param.data_len = skb->len;
975 icmp_param.head_len = sizeof(struct icmphdr);
976 icmp_reply(&icmp_param, skb);
977 }
978 /* should there be an ICMP stat for ignored echos? */
979 return true;
980}
981
982/*
983 * Handle ICMP Timestamp requests.
984 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
985 * SHOULD be in the kernel for minimum random latency.
986 * MUST be accurate to a few minutes.
987 * MUST be updated at least at 15Hz.
988 */
989static bool icmp_timestamp(struct sk_buff *skb)
990{
991 struct icmp_bxm icmp_param;
992 /*
993 * Too short.
994 */
995 if (skb->len < 4)
996 goto out_err;
997
998 /*
999 * Fill in the current time as ms since midnight UT:
1000 */
1001 icmp_param.data.times[1] = inet_current_timestamp();
1002 icmp_param.data.times[2] = icmp_param.data.times[1];
1003
1004 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1005
1006 icmp_param.data.icmph = *icmp_hdr(skb);
1007 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1008 icmp_param.data.icmph.code = 0;
1009 icmp_param.skb = skb;
1010 icmp_param.offset = 0;
1011 icmp_param.data_len = 0;
1012 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1013 icmp_reply(&icmp_param, skb);
1014 return true;
1015
1016out_err:
1017 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1018 return false;
1019}
1020
1021static bool icmp_discard(struct sk_buff *skb)
1022{
1023 /* pretend it was a success */
1024 return true;
1025}
1026
1027/*
1028 * Deal with incoming ICMP packets.
1029 */
1030int icmp_rcv(struct sk_buff *skb)
1031{
1032 struct icmphdr *icmph;
1033 struct rtable *rt = skb_rtable(skb);
1034 struct net *net = dev_net(rt->dst.dev);
1035 bool success;
1036
1037 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1038 struct sec_path *sp = skb_sec_path(skb);
1039 int nh;
1040
1041 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1042 XFRM_STATE_ICMP))
1043 goto drop;
1044
1045 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1046 goto drop;
1047
1048 nh = skb_network_offset(skb);
1049 skb_set_network_header(skb, sizeof(*icmph));
1050
1051 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1052 goto drop;
1053
1054 skb_set_network_header(skb, nh);
1055 }
1056
1057 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1058
1059 if (skb_checksum_simple_validate(skb))
1060 goto csum_error;
1061
1062 if (!pskb_pull(skb, sizeof(*icmph)))
1063 goto error;
1064
1065 icmph = icmp_hdr(skb);
1066
1067 ICMPMSGIN_INC_STATS(net, icmph->type);
1068 /*
1069 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1070 *
1071 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1072 * discarded.
1073 */
1074 if (icmph->type > NR_ICMP_TYPES)
1075 goto error;
1076
1077
1078 /*
1079 * Parse the ICMP message
1080 */
1081
1082 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1083 /*
1084 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1085 * silently ignored (we let user decide with a sysctl).
1086 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1087 * discarded if to broadcast/multicast.
1088 */
1089 if ((icmph->type == ICMP_ECHO ||
1090 icmph->type == ICMP_TIMESTAMP) &&
1091 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1092 goto error;
1093 }
1094 if (icmph->type != ICMP_ECHO &&
1095 icmph->type != ICMP_TIMESTAMP &&
1096 icmph->type != ICMP_ADDRESS &&
1097 icmph->type != ICMP_ADDRESSREPLY) {
1098 goto error;
1099 }
1100 }
1101
1102 success = icmp_pointers[icmph->type].handler(skb);
1103
1104 if (success) {
1105 consume_skb(skb);
1106 return NET_RX_SUCCESS;
1107 }
1108
1109drop:
1110 kfree_skb(skb);
1111 return NET_RX_DROP;
1112csum_error:
1113 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1114error:
1115 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1116 goto drop;
1117}
1118
1119static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1120{
1121 struct icmp_extobj_hdr *objh, _objh;
1122 struct icmp_ext_hdr *exth, _exth;
1123 u16 olen;
1124
1125 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1126 if (!exth)
1127 return false;
1128 if (exth->version != 2)
1129 return true;
1130
1131 if (exth->checksum &&
1132 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1133 return false;
1134
1135 off += sizeof(_exth);
1136 while (off < skb->len) {
1137 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1138 if (!objh)
1139 return false;
1140
1141 olen = ntohs(objh->length);
1142 if (olen < sizeof(_objh))
1143 return false;
1144
1145 off += olen;
1146 if (off > skb->len)
1147 return false;
1148 }
1149
1150 return true;
1151}
1152
1153void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1154 struct sock_ee_data_rfc4884 *out,
1155 int thlen, int off)
1156{
1157 int hlen;
1158
1159 /* original datagram headers: end of icmph to payload (skb->data) */
1160 hlen = -skb_transport_offset(skb) - thlen;
1161
1162 /* per rfc 4884: minimal datagram length of 128 bytes */
1163 if (off < 128 || off < hlen)
1164 return;
1165
1166 /* kernel has stripped headers: return payload offset in bytes */
1167 off -= hlen;
1168 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1169 return;
1170
1171 out->len = off;
1172
1173 if (!ip_icmp_error_rfc4884_validate(skb, off))
1174 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1175}
1176EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1177
1178int icmp_err(struct sk_buff *skb, u32 info)
1179{
1180 struct iphdr *iph = (struct iphdr *)skb->data;
1181 int offset = iph->ihl<<2;
1182 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1183 int type = icmp_hdr(skb)->type;
1184 int code = icmp_hdr(skb)->code;
1185 struct net *net = dev_net(skb->dev);
1186
1187 /*
1188 * Use ping_err to handle all icmp errors except those
1189 * triggered by ICMP_ECHOREPLY which sent from kernel.
1190 */
1191 if (icmph->type != ICMP_ECHOREPLY) {
1192 ping_err(skb, offset, info);
1193 return 0;
1194 }
1195
1196 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1197 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1198 else if (type == ICMP_REDIRECT)
1199 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1200
1201 return 0;
1202}
1203
1204/*
1205 * This table is the definition of how we handle ICMP.
1206 */
1207static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1208 [ICMP_ECHOREPLY] = {
1209 .handler = ping_rcv,
1210 },
1211 [1] = {
1212 .handler = icmp_discard,
1213 .error = 1,
1214 },
1215 [2] = {
1216 .handler = icmp_discard,
1217 .error = 1,
1218 },
1219 [ICMP_DEST_UNREACH] = {
1220 .handler = icmp_unreach,
1221 .error = 1,
1222 },
1223 [ICMP_SOURCE_QUENCH] = {
1224 .handler = icmp_unreach,
1225 .error = 1,
1226 },
1227 [ICMP_REDIRECT] = {
1228 .handler = icmp_redirect,
1229 .error = 1,
1230 },
1231 [6] = {
1232 .handler = icmp_discard,
1233 .error = 1,
1234 },
1235 [7] = {
1236 .handler = icmp_discard,
1237 .error = 1,
1238 },
1239 [ICMP_ECHO] = {
1240 .handler = icmp_echo,
1241 },
1242 [9] = {
1243 .handler = icmp_discard,
1244 .error = 1,
1245 },
1246 [10] = {
1247 .handler = icmp_discard,
1248 .error = 1,
1249 },
1250 [ICMP_TIME_EXCEEDED] = {
1251 .handler = icmp_unreach,
1252 .error = 1,
1253 },
1254 [ICMP_PARAMETERPROB] = {
1255 .handler = icmp_unreach,
1256 .error = 1,
1257 },
1258 [ICMP_TIMESTAMP] = {
1259 .handler = icmp_timestamp,
1260 },
1261 [ICMP_TIMESTAMPREPLY] = {
1262 .handler = icmp_discard,
1263 },
1264 [ICMP_INFO_REQUEST] = {
1265 .handler = icmp_discard,
1266 },
1267 [ICMP_INFO_REPLY] = {
1268 .handler = icmp_discard,
1269 },
1270 [ICMP_ADDRESS] = {
1271 .handler = icmp_discard,
1272 },
1273 [ICMP_ADDRESSREPLY] = {
1274 .handler = icmp_discard,
1275 },
1276};
1277
1278static void __net_exit icmp_sk_exit(struct net *net)
1279{
1280 int i;
1281
1282 for_each_possible_cpu(i)
1283 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1284 free_percpu(net->ipv4.icmp_sk);
1285 net->ipv4.icmp_sk = NULL;
1286}
1287
1288static int __net_init icmp_sk_init(struct net *net)
1289{
1290 int i, err;
1291
1292 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1293 if (!net->ipv4.icmp_sk)
1294 return -ENOMEM;
1295
1296 for_each_possible_cpu(i) {
1297 struct sock *sk;
1298
1299 err = inet_ctl_sock_create(&sk, PF_INET,
1300 SOCK_RAW, IPPROTO_ICMP, net);
1301 if (err < 0)
1302 goto fail;
1303
1304 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1305
1306 /* Enough space for 2 64K ICMP packets, including
1307 * sk_buff/skb_shared_info struct overhead.
1308 */
1309 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1310
1311 /*
1312 * Speedup sock_wfree()
1313 */
1314 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1315 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1316 }
1317
1318 /* Control parameters for ECHO replies. */
1319 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1320 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1321
1322 /* Control parameter - ignore bogus broadcast responses? */
1323 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1324
1325 /*
1326 * Configurable global rate limit.
1327 *
1328 * ratelimit defines tokens/packet consumed for dst->rate_token
1329 * bucket ratemask defines which icmp types are ratelimited by
1330 * setting it's bit position.
1331 *
1332 * default:
1333 * dest unreachable (3), source quench (4),
1334 * time exceeded (11), parameter problem (12)
1335 */
1336
1337 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1338 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1339 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1340
1341 return 0;
1342
1343fail:
1344 icmp_sk_exit(net);
1345 return err;
1346}
1347
1348static struct pernet_operations __net_initdata icmp_sk_ops = {
1349 .init = icmp_sk_init,
1350 .exit = icmp_sk_exit,
1351};
1352
1353int __init icmp_init(void)
1354{
1355 return register_pernet_subsys(&icmp_sk_ops);
1356}