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#include <net/addrconf.h>
96
97/*
98 * Build xmit assembly blocks
99 */
100
101struct icmp_bxm {
102 struct sk_buff *skb;
103 int offset;
104 int data_len;
105
106 struct {
107 struct icmphdr icmph;
108 __be32 times[3];
109 } data;
110 int head_len;
111 struct ip_options_data replyopts;
112};
113
114/* An array of errno for error messages from dest unreach. */
115/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
116
117const struct icmp_err icmp_err_convert[] = {
118 {
119 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
120 .fatal = 0,
121 },
122 {
123 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
124 .fatal = 0,
125 },
126 {
127 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
128 .fatal = 1,
129 },
130 {
131 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
132 .fatal = 1,
133 },
134 {
135 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
136 .fatal = 0,
137 },
138 {
139 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
140 .fatal = 0,
141 },
142 {
143 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
144 .fatal = 1,
145 },
146 {
147 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
148 .fatal = 1,
149 },
150 {
151 .errno = ENONET, /* ICMP_HOST_ISOLATED */
152 .fatal = 1,
153 },
154 {
155 .errno = ENETUNREACH, /* ICMP_NET_ANO */
156 .fatal = 1,
157 },
158 {
159 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
160 .fatal = 1,
161 },
162 {
163 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
164 .fatal = 0,
165 },
166 {
167 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
168 .fatal = 0,
169 },
170 {
171 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
172 .fatal = 1,
173 },
174 {
175 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
176 .fatal = 1,
177 },
178 {
179 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
180 .fatal = 1,
181 },
182};
183EXPORT_SYMBOL(icmp_err_convert);
184
185/*
186 * ICMP control array. This specifies what to do with each ICMP.
187 */
188
189struct icmp_control {
190 enum skb_drop_reason (*handler)(struct sk_buff *skb);
191 short error; /* This ICMP is classed as an error message */
192};
193
194static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
195
196static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
197
198/* Called with BH disabled */
199static inline struct sock *icmp_xmit_lock(struct net *net)
200{
201 struct sock *sk;
202
203 sk = this_cpu_read(ipv4_icmp_sk);
204
205 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
206 /* This can happen if the output path signals a
207 * dst_link_failure() for an outgoing ICMP packet.
208 */
209 return NULL;
210 }
211 sock_net_set(sk, net);
212 return sk;
213}
214
215static inline void icmp_xmit_unlock(struct sock *sk)
216{
217 sock_net_set(sk, &init_net);
218 spin_unlock(&sk->sk_lock.slock);
219}
220
221int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
222int sysctl_icmp_msgs_burst __read_mostly = 50;
223
224static struct {
225 spinlock_t lock;
226 u32 credit;
227 u32 stamp;
228} icmp_global = {
229 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
230};
231
232/**
233 * icmp_global_allow - Are we allowed to send one more ICMP message ?
234 *
235 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
236 * Returns false if we reached the limit and can not send another packet.
237 * Note: called with BH disabled
238 */
239bool icmp_global_allow(void)
240{
241 u32 credit, delta, incr = 0, now = (u32)jiffies;
242 bool rc = false;
243
244 /* Check if token bucket is empty and cannot be refilled
245 * without taking the spinlock. The READ_ONCE() are paired
246 * with the following WRITE_ONCE() in this same function.
247 */
248 if (!READ_ONCE(icmp_global.credit)) {
249 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
250 if (delta < HZ / 50)
251 return false;
252 }
253
254 spin_lock(&icmp_global.lock);
255 delta = min_t(u32, now - icmp_global.stamp, HZ);
256 if (delta >= HZ / 50) {
257 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
258 if (incr)
259 WRITE_ONCE(icmp_global.stamp, now);
260 }
261 credit = min_t(u32, icmp_global.credit + incr,
262 READ_ONCE(sysctl_icmp_msgs_burst));
263 if (credit) {
264 /* We want to use a credit of one in average, but need to randomize
265 * it for security reasons.
266 */
267 credit = max_t(int, credit - get_random_u32_below(3), 0);
268 rc = true;
269 }
270 WRITE_ONCE(icmp_global.credit, credit);
271 spin_unlock(&icmp_global.lock);
272 return rc;
273}
274EXPORT_SYMBOL(icmp_global_allow);
275
276static bool icmpv4_mask_allow(struct net *net, int type, int code)
277{
278 if (type > NR_ICMP_TYPES)
279 return true;
280
281 /* Don't limit PMTU discovery. */
282 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
283 return true;
284
285 /* Limit if icmp type is enabled in ratemask. */
286 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
287 return true;
288
289 return false;
290}
291
292static bool icmpv4_global_allow(struct net *net, int type, int code)
293{
294 if (icmpv4_mask_allow(net, type, code))
295 return true;
296
297 if (icmp_global_allow())
298 return true;
299
300 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
301 return false;
302}
303
304/*
305 * Send an ICMP frame.
306 */
307
308static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
309 struct flowi4 *fl4, int type, int code)
310{
311 struct dst_entry *dst = &rt->dst;
312 struct inet_peer *peer;
313 bool rc = true;
314 int vif;
315
316 if (icmpv4_mask_allow(net, type, code))
317 goto out;
318
319 /* No rate limit on loopback */
320 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
321 goto out;
322
323 vif = l3mdev_master_ifindex(dst->dev);
324 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
325 rc = inet_peer_xrlim_allow(peer,
326 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
327 if (peer)
328 inet_putpeer(peer);
329out:
330 if (!rc)
331 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
332 return rc;
333}
334
335/*
336 * Maintain the counters used in the SNMP statistics for outgoing ICMP
337 */
338void icmp_out_count(struct net *net, unsigned char type)
339{
340 ICMPMSGOUT_INC_STATS(net, type);
341 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
342}
343
344/*
345 * Checksum each fragment, and on the first include the headers and final
346 * checksum.
347 */
348static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
349 struct sk_buff *skb)
350{
351 struct icmp_bxm *icmp_param = from;
352 __wsum csum;
353
354 csum = skb_copy_and_csum_bits(icmp_param->skb,
355 icmp_param->offset + offset,
356 to, len);
357
358 skb->csum = csum_block_add(skb->csum, csum, odd);
359 if (icmp_pointers[icmp_param->data.icmph.type].error)
360 nf_ct_attach(skb, icmp_param->skb);
361 return 0;
362}
363
364static void icmp_push_reply(struct sock *sk,
365 struct icmp_bxm *icmp_param,
366 struct flowi4 *fl4,
367 struct ipcm_cookie *ipc, struct rtable **rt)
368{
369 struct sk_buff *skb;
370
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;
380 struct sk_buff *skb1;
381
382 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
383 (char *)icmph,
384 icmp_param->head_len);
385 skb_queue_walk(&sk->sk_write_queue, skb1) {
386 csum = csum_add(csum, skb1->csum);
387 }
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_common(&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(sk, 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
460/*
461 * The device used for looking up which routing table to use for sending an ICMP
462 * error is preferably the source whenever it is set, which should ensure the
463 * icmp error can be sent to the source host, else lookup using the routing
464 * table of the destination device, else use the main routing table (index 0).
465 */
466static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
467{
468 struct net_device *route_lookup_dev = NULL;
469
470 if (skb->dev)
471 route_lookup_dev = skb->dev;
472 else if (skb_dst(skb))
473 route_lookup_dev = skb_dst(skb)->dev;
474 return route_lookup_dev;
475}
476
477static struct rtable *icmp_route_lookup(struct net *net,
478 struct flowi4 *fl4,
479 struct sk_buff *skb_in,
480 const struct iphdr *iph,
481 __be32 saddr, u8 tos, u32 mark,
482 int type, int code,
483 struct icmp_bxm *param)
484{
485 struct net_device *route_lookup_dev;
486 struct dst_entry *dst, *dst2;
487 struct rtable *rt, *rt2;
488 struct flowi4 fl4_dec;
489 int err;
490
491 memset(fl4, 0, sizeof(*fl4));
492 fl4->daddr = (param->replyopts.opt.opt.srr ?
493 param->replyopts.opt.opt.faddr : iph->saddr);
494 fl4->saddr = saddr;
495 fl4->flowi4_mark = mark;
496 fl4->flowi4_uid = sock_net_uid(net, NULL);
497 fl4->flowi4_tos = RT_TOS(tos);
498 fl4->flowi4_proto = IPPROTO_ICMP;
499 fl4->fl4_icmp_type = type;
500 fl4->fl4_icmp_code = code;
501 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
502 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
503
504 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
505 rt = ip_route_output_key_hash(net, fl4, skb_in);
506 if (IS_ERR(rt))
507 return rt;
508
509 /* No need to clone since we're just using its address. */
510 rt2 = rt;
511
512 dst = xfrm_lookup(net, &rt->dst,
513 flowi4_to_flowi(fl4), NULL, 0);
514 rt = dst_rtable(dst);
515 if (!IS_ERR(dst)) {
516 if (rt != rt2)
517 return rt;
518 } else if (PTR_ERR(dst) == -EPERM) {
519 rt = NULL;
520 } else {
521 return rt;
522 }
523 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
524 if (err)
525 goto relookup_failed;
526
527 if (inet_addr_type_dev_table(net, route_lookup_dev,
528 fl4_dec.saddr) == RTN_LOCAL) {
529 rt2 = __ip_route_output_key(net, &fl4_dec);
530 if (IS_ERR(rt2))
531 err = PTR_ERR(rt2);
532 } else {
533 struct flowi4 fl4_2 = {};
534 unsigned long orefdst;
535
536 fl4_2.daddr = fl4_dec.saddr;
537 rt2 = ip_route_output_key(net, &fl4_2);
538 if (IS_ERR(rt2)) {
539 err = PTR_ERR(rt2);
540 goto relookup_failed;
541 }
542 /* Ugh! */
543 orefdst = skb_in->_skb_refdst; /* save old refdst */
544 skb_dst_set(skb_in, NULL);
545 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
546 RT_TOS(tos), rt2->dst.dev);
547
548 dst_release(&rt2->dst);
549 rt2 = skb_rtable(skb_in);
550 skb_in->_skb_refdst = orefdst; /* restore old refdst */
551 }
552
553 if (err)
554 goto relookup_failed;
555
556 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
557 XFRM_LOOKUP_ICMP);
558 rt2 = dst_rtable(dst2);
559 if (!IS_ERR(dst2)) {
560 dst_release(&rt->dst);
561 memcpy(fl4, &fl4_dec, sizeof(*fl4));
562 rt = rt2;
563 } else if (PTR_ERR(dst2) == -EPERM) {
564 if (rt)
565 dst_release(&rt->dst);
566 return rt2;
567 } else {
568 err = PTR_ERR(dst2);
569 goto relookup_failed;
570 }
571 return rt;
572
573relookup_failed:
574 if (rt)
575 return rt;
576 return ERR_PTR(err);
577}
578
579/*
580 * Send an ICMP message in response to a situation
581 *
582 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
583 * MAY send more (we do).
584 * MUST NOT change this header information.
585 * MUST NOT reply to a multicast/broadcast IP address.
586 * MUST NOT reply to a multicast/broadcast MAC address.
587 * MUST reply to only the first fragment.
588 */
589
590void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
591 const struct ip_options *opt)
592{
593 struct iphdr *iph;
594 int room;
595 struct icmp_bxm icmp_param;
596 struct rtable *rt = skb_rtable(skb_in);
597 struct ipcm_cookie ipc;
598 struct flowi4 fl4;
599 __be32 saddr;
600 u8 tos;
601 u32 mark;
602 struct net *net;
603 struct sock *sk;
604
605 if (!rt)
606 goto out;
607
608 if (rt->dst.dev)
609 net = dev_net(rt->dst.dev);
610 else if (skb_in->dev)
611 net = dev_net(skb_in->dev);
612 else
613 goto out;
614
615 /*
616 * Find the original header. It is expected to be valid, of course.
617 * Check this, icmp_send is called from the most obscure devices
618 * sometimes.
619 */
620 iph = ip_hdr(skb_in);
621
622 if ((u8 *)iph < skb_in->head ||
623 (skb_network_header(skb_in) + sizeof(*iph)) >
624 skb_tail_pointer(skb_in))
625 goto out;
626
627 /*
628 * No replies to physical multicast/broadcast
629 */
630 if (skb_in->pkt_type != PACKET_HOST)
631 goto out;
632
633 /*
634 * Now check at the protocol level
635 */
636 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
637 goto out;
638
639 /*
640 * Only reply to fragment 0. We byte re-order the constant
641 * mask for efficiency.
642 */
643 if (iph->frag_off & htons(IP_OFFSET))
644 goto out;
645
646 /*
647 * If we send an ICMP error to an ICMP error a mess would result..
648 */
649 if (icmp_pointers[type].error) {
650 /*
651 * We are an error, check if we are replying to an
652 * ICMP error
653 */
654 if (iph->protocol == IPPROTO_ICMP) {
655 u8 _inner_type, *itp;
656
657 itp = skb_header_pointer(skb_in,
658 skb_network_header(skb_in) +
659 (iph->ihl << 2) +
660 offsetof(struct icmphdr,
661 type) -
662 skb_in->data,
663 sizeof(_inner_type),
664 &_inner_type);
665 if (!itp)
666 goto out;
667
668 /*
669 * Assume any unknown ICMP type is an error. This
670 * isn't specified by the RFC, but think about it..
671 */
672 if (*itp > NR_ICMP_TYPES ||
673 icmp_pointers[*itp].error)
674 goto out;
675 }
676 }
677
678 /* Needed by both icmp_global_allow and icmp_xmit_lock */
679 local_bh_disable();
680
681 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
682 * incoming dev is loopback. If outgoing dev change to not be
683 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
684 */
685 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
686 !icmpv4_global_allow(net, type, code))
687 goto out_bh_enable;
688
689 sk = icmp_xmit_lock(net);
690 if (!sk)
691 goto out_bh_enable;
692
693 /*
694 * Construct source address and options.
695 */
696
697 saddr = iph->daddr;
698 if (!(rt->rt_flags & RTCF_LOCAL)) {
699 struct net_device *dev = NULL;
700
701 rcu_read_lock();
702 if (rt_is_input_route(rt) &&
703 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
704 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
705
706 if (dev)
707 saddr = inet_select_addr(dev, iph->saddr,
708 RT_SCOPE_LINK);
709 else
710 saddr = 0;
711 rcu_read_unlock();
712 }
713
714 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
715 IPTOS_PREC_INTERNETCONTROL) :
716 iph->tos;
717 mark = IP4_REPLY_MARK(net, skb_in->mark);
718
719 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
720 goto out_unlock;
721
722
723 /*
724 * Prepare data for ICMP header.
725 */
726
727 icmp_param.data.icmph.type = type;
728 icmp_param.data.icmph.code = code;
729 icmp_param.data.icmph.un.gateway = info;
730 icmp_param.data.icmph.checksum = 0;
731 icmp_param.skb = skb_in;
732 icmp_param.offset = skb_network_offset(skb_in);
733 inet_sk(sk)->tos = tos;
734 ipcm_init(&ipc);
735 ipc.addr = iph->saddr;
736 ipc.opt = &icmp_param.replyopts.opt;
737 ipc.sockc.mark = mark;
738
739 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
740 type, code, &icmp_param);
741 if (IS_ERR(rt))
742 goto out_unlock;
743
744 /* peer icmp_ratelimit */
745 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
746 goto ende;
747
748 /* RFC says return as much as we can without exceeding 576 bytes. */
749
750 room = dst_mtu(&rt->dst);
751 if (room > 576)
752 room = 576;
753 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
754 room -= sizeof(struct icmphdr);
755 /* Guard against tiny mtu. We need to include at least one
756 * IP network header for this message to make any sense.
757 */
758 if (room <= (int)sizeof(struct iphdr))
759 goto ende;
760
761 icmp_param.data_len = skb_in->len - icmp_param.offset;
762 if (icmp_param.data_len > room)
763 icmp_param.data_len = room;
764 icmp_param.head_len = sizeof(struct icmphdr);
765
766 /* if we don't have a source address at this point, fall back to the
767 * dummy address instead of sending out a packet with a source address
768 * of 0.0.0.0
769 */
770 if (!fl4.saddr)
771 fl4.saddr = htonl(INADDR_DUMMY);
772
773 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
774ende:
775 ip_rt_put(rt);
776out_unlock:
777 icmp_xmit_unlock(sk);
778out_bh_enable:
779 local_bh_enable();
780out:;
781}
782EXPORT_SYMBOL(__icmp_send);
783
784#if IS_ENABLED(CONFIG_NF_NAT)
785#include <net/netfilter/nf_conntrack.h>
786void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
787{
788 struct sk_buff *cloned_skb = NULL;
789 struct ip_options opts = { 0 };
790 enum ip_conntrack_info ctinfo;
791 struct nf_conn *ct;
792 __be32 orig_ip;
793
794 ct = nf_ct_get(skb_in, &ctinfo);
795 if (!ct || !(ct->status & IPS_SRC_NAT)) {
796 __icmp_send(skb_in, type, code, info, &opts);
797 return;
798 }
799
800 if (skb_shared(skb_in))
801 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
802
803 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
804 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
805 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
806 skb_network_offset(skb_in) + sizeof(struct iphdr))))
807 goto out;
808
809 orig_ip = ip_hdr(skb_in)->saddr;
810 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
811 __icmp_send(skb_in, type, code, info, &opts);
812 ip_hdr(skb_in)->saddr = orig_ip;
813out:
814 consume_skb(cloned_skb);
815}
816EXPORT_SYMBOL(icmp_ndo_send);
817#endif
818
819static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
820{
821 const struct iphdr *iph = (const struct iphdr *)skb->data;
822 const struct net_protocol *ipprot;
823 int protocol = iph->protocol;
824
825 /* Checkin full IP header plus 8 bytes of protocol to
826 * avoid additional coding at protocol handlers.
827 */
828 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
829 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
830 return;
831 }
832
833 raw_icmp_error(skb, protocol, info);
834
835 ipprot = rcu_dereference(inet_protos[protocol]);
836 if (ipprot && ipprot->err_handler)
837 ipprot->err_handler(skb, info);
838}
839
840static bool icmp_tag_validation(int proto)
841{
842 bool ok;
843
844 rcu_read_lock();
845 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
846 rcu_read_unlock();
847 return ok;
848}
849
850/*
851 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
852 * ICMP_PARAMETERPROB.
853 */
854
855static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
856{
857 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
858 const struct iphdr *iph;
859 struct icmphdr *icmph;
860 struct net *net;
861 u32 info = 0;
862
863 net = dev_net(skb_dst(skb)->dev);
864
865 /*
866 * Incomplete header ?
867 * Only checks for the IP header, there should be an
868 * additional check for longer headers in upper levels.
869 */
870
871 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
872 goto out_err;
873
874 icmph = icmp_hdr(skb);
875 iph = (const struct iphdr *)skb->data;
876
877 if (iph->ihl < 5) { /* Mangled header, drop. */
878 reason = SKB_DROP_REASON_IP_INHDR;
879 goto out_err;
880 }
881
882 switch (icmph->type) {
883 case ICMP_DEST_UNREACH:
884 switch (icmph->code & 15) {
885 case ICMP_NET_UNREACH:
886 case ICMP_HOST_UNREACH:
887 case ICMP_PROT_UNREACH:
888 case ICMP_PORT_UNREACH:
889 break;
890 case ICMP_FRAG_NEEDED:
891 /* for documentation of the ip_no_pmtu_disc
892 * values please see
893 * Documentation/networking/ip-sysctl.rst
894 */
895 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
896 default:
897 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
898 &iph->daddr);
899 break;
900 case 2:
901 goto out;
902 case 3:
903 if (!icmp_tag_validation(iph->protocol))
904 goto out;
905 fallthrough;
906 case 0:
907 info = ntohs(icmph->un.frag.mtu);
908 }
909 break;
910 case ICMP_SR_FAILED:
911 net_dbg_ratelimited("%pI4: Source Route Failed\n",
912 &iph->daddr);
913 break;
914 default:
915 break;
916 }
917 if (icmph->code > NR_ICMP_UNREACH)
918 goto out;
919 break;
920 case ICMP_PARAMETERPROB:
921 info = ntohl(icmph->un.gateway) >> 24;
922 break;
923 case ICMP_TIME_EXCEEDED:
924 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
925 if (icmph->code == ICMP_EXC_FRAGTIME)
926 goto out;
927 break;
928 }
929
930 /*
931 * Throw it at our lower layers
932 *
933 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
934 * header.
935 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
936 * transport layer.
937 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
938 * transport layer.
939 */
940
941 /*
942 * Check the other end isn't violating RFC 1122. Some routers send
943 * bogus responses to broadcast frames. If you see this message
944 * first check your netmask matches at both ends, if it does then
945 * get the other vendor to fix their kit.
946 */
947
948 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
949 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
950 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
951 &ip_hdr(skb)->saddr,
952 icmph->type, icmph->code,
953 &iph->daddr, skb->dev->name);
954 goto out;
955 }
956
957 icmp_socket_deliver(skb, info);
958
959out:
960 return reason;
961out_err:
962 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
963 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
964}
965
966
967/*
968 * Handle ICMP_REDIRECT.
969 */
970
971static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
972{
973 if (skb->len < sizeof(struct iphdr)) {
974 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
975 return SKB_DROP_REASON_PKT_TOO_SMALL;
976 }
977
978 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
979 /* there aught to be a stat */
980 return SKB_DROP_REASON_NOMEM;
981 }
982
983 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
984 return SKB_NOT_DROPPED_YET;
985}
986
987/*
988 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
989 *
990 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
991 * requests.
992 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
993 * included in the reply.
994 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
995 * echo requests, MUST have default=NOT.
996 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
997 * Extended Echo Functionality, MUST be disabled by default
998 * See also WRT handling of options once they are done and working.
999 */
1000
1001static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1002{
1003 struct icmp_bxm icmp_param;
1004 struct net *net;
1005
1006 net = dev_net(skb_dst(skb)->dev);
1007 /* should there be an ICMP stat for ignored echos? */
1008 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1009 return SKB_NOT_DROPPED_YET;
1010
1011 icmp_param.data.icmph = *icmp_hdr(skb);
1012 icmp_param.skb = skb;
1013 icmp_param.offset = 0;
1014 icmp_param.data_len = skb->len;
1015 icmp_param.head_len = sizeof(struct icmphdr);
1016
1017 if (icmp_param.data.icmph.type == ICMP_ECHO)
1018 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1019 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1020 return SKB_NOT_DROPPED_YET;
1021
1022 icmp_reply(&icmp_param, skb);
1023 return SKB_NOT_DROPPED_YET;
1024}
1025
1026/* Helper for icmp_echo and icmpv6_echo_reply.
1027 * Searches for net_device that matches PROBE interface identifier
1028 * and builds PROBE reply message in icmphdr.
1029 *
1030 * Returns false if PROBE responses are disabled via sysctl
1031 */
1032
1033bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1034{
1035 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1036 struct icmp_ext_echo_iio *iio, _iio;
1037 struct net *net = dev_net(skb->dev);
1038 struct inet6_dev *in6_dev;
1039 struct in_device *in_dev;
1040 struct net_device *dev;
1041 char buff[IFNAMSIZ];
1042 u16 ident_len;
1043 u8 status;
1044
1045 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1046 return false;
1047
1048 /* We currently only support probing interfaces on the proxy node
1049 * Check to ensure L-bit is set
1050 */
1051 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1052 return false;
1053 /* Clear status bits in reply message */
1054 icmphdr->un.echo.sequence &= htons(0xFF00);
1055 if (icmphdr->type == ICMP_EXT_ECHO)
1056 icmphdr->type = ICMP_EXT_ECHOREPLY;
1057 else
1058 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1059 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1060 /* Size of iio is class_type dependent.
1061 * Only check header here and assign length based on ctype in the switch statement
1062 */
1063 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1064 if (!ext_hdr || !iio)
1065 goto send_mal_query;
1066 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1067 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1068 goto send_mal_query;
1069 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1070 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1071 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1072 if (!iio)
1073 goto send_mal_query;
1074
1075 status = 0;
1076 dev = NULL;
1077 switch (iio->extobj_hdr.class_type) {
1078 case ICMP_EXT_ECHO_CTYPE_NAME:
1079 if (ident_len >= IFNAMSIZ)
1080 goto send_mal_query;
1081 memset(buff, 0, sizeof(buff));
1082 memcpy(buff, &iio->ident.name, ident_len);
1083 dev = dev_get_by_name(net, buff);
1084 break;
1085 case ICMP_EXT_ECHO_CTYPE_INDEX:
1086 if (ident_len != sizeof(iio->ident.ifindex))
1087 goto send_mal_query;
1088 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1089 break;
1090 case ICMP_EXT_ECHO_CTYPE_ADDR:
1091 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1092 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1093 iio->ident.addr.ctype3_hdr.addrlen)
1094 goto send_mal_query;
1095 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1096 case ICMP_AFI_IP:
1097 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1098 goto send_mal_query;
1099 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1100 break;
1101#if IS_ENABLED(CONFIG_IPV6)
1102 case ICMP_AFI_IP6:
1103 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1104 goto send_mal_query;
1105 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1106 dev_hold(dev);
1107 break;
1108#endif
1109 default:
1110 goto send_mal_query;
1111 }
1112 break;
1113 default:
1114 goto send_mal_query;
1115 }
1116 if (!dev) {
1117 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1118 return true;
1119 }
1120 /* Fill bits in reply message */
1121 if (dev->flags & IFF_UP)
1122 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1123
1124 in_dev = __in_dev_get_rcu(dev);
1125 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1126 status |= ICMP_EXT_ECHOREPLY_IPV4;
1127
1128 in6_dev = __in6_dev_get(dev);
1129 if (in6_dev && !list_empty(&in6_dev->addr_list))
1130 status |= ICMP_EXT_ECHOREPLY_IPV6;
1131
1132 dev_put(dev);
1133 icmphdr->un.echo.sequence |= htons(status);
1134 return true;
1135send_mal_query:
1136 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1137 return true;
1138}
1139EXPORT_SYMBOL_GPL(icmp_build_probe);
1140
1141/*
1142 * Handle ICMP Timestamp requests.
1143 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1144 * SHOULD be in the kernel for minimum random latency.
1145 * MUST be accurate to a few minutes.
1146 * MUST be updated at least at 15Hz.
1147 */
1148static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1149{
1150 struct icmp_bxm icmp_param;
1151 /*
1152 * Too short.
1153 */
1154 if (skb->len < 4)
1155 goto out_err;
1156
1157 /*
1158 * Fill in the current time as ms since midnight UT:
1159 */
1160 icmp_param.data.times[1] = inet_current_timestamp();
1161 icmp_param.data.times[2] = icmp_param.data.times[1];
1162
1163 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1164
1165 icmp_param.data.icmph = *icmp_hdr(skb);
1166 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1167 icmp_param.data.icmph.code = 0;
1168 icmp_param.skb = skb;
1169 icmp_param.offset = 0;
1170 icmp_param.data_len = 0;
1171 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1172 icmp_reply(&icmp_param, skb);
1173 return SKB_NOT_DROPPED_YET;
1174
1175out_err:
1176 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1177 return SKB_DROP_REASON_PKT_TOO_SMALL;
1178}
1179
1180static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1181{
1182 /* pretend it was a success */
1183 return SKB_NOT_DROPPED_YET;
1184}
1185
1186/*
1187 * Deal with incoming ICMP packets.
1188 */
1189int icmp_rcv(struct sk_buff *skb)
1190{
1191 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1192 struct rtable *rt = skb_rtable(skb);
1193 struct net *net = dev_net(rt->dst.dev);
1194 struct icmphdr *icmph;
1195
1196 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1197 struct sec_path *sp = skb_sec_path(skb);
1198 int nh;
1199
1200 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1201 XFRM_STATE_ICMP)) {
1202 reason = SKB_DROP_REASON_XFRM_POLICY;
1203 goto drop;
1204 }
1205
1206 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1207 goto drop;
1208
1209 nh = skb_network_offset(skb);
1210 skb_set_network_header(skb, sizeof(*icmph));
1211
1212 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1213 skb)) {
1214 reason = SKB_DROP_REASON_XFRM_POLICY;
1215 goto drop;
1216 }
1217
1218 skb_set_network_header(skb, nh);
1219 }
1220
1221 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1222
1223 if (skb_checksum_simple_validate(skb))
1224 goto csum_error;
1225
1226 if (!pskb_pull(skb, sizeof(*icmph)))
1227 goto error;
1228
1229 icmph = icmp_hdr(skb);
1230
1231 ICMPMSGIN_INC_STATS(net, icmph->type);
1232
1233 /* Check for ICMP Extended Echo (PROBE) messages */
1234 if (icmph->type == ICMP_EXT_ECHO) {
1235 /* We can't use icmp_pointers[].handler() because it is an array of
1236 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1237 */
1238 reason = icmp_echo(skb);
1239 goto reason_check;
1240 }
1241
1242 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1243 reason = ping_rcv(skb);
1244 goto reason_check;
1245 }
1246
1247 /*
1248 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1249 *
1250 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1251 * discarded.
1252 */
1253 if (icmph->type > NR_ICMP_TYPES) {
1254 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1255 goto error;
1256 }
1257
1258 /*
1259 * Parse the ICMP message
1260 */
1261
1262 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1263 /*
1264 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1265 * silently ignored (we let user decide with a sysctl).
1266 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1267 * discarded if to broadcast/multicast.
1268 */
1269 if ((icmph->type == ICMP_ECHO ||
1270 icmph->type == ICMP_TIMESTAMP) &&
1271 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1272 reason = SKB_DROP_REASON_INVALID_PROTO;
1273 goto error;
1274 }
1275 if (icmph->type != ICMP_ECHO &&
1276 icmph->type != ICMP_TIMESTAMP &&
1277 icmph->type != ICMP_ADDRESS &&
1278 icmph->type != ICMP_ADDRESSREPLY) {
1279 reason = SKB_DROP_REASON_INVALID_PROTO;
1280 goto error;
1281 }
1282 }
1283
1284 reason = icmp_pointers[icmph->type].handler(skb);
1285reason_check:
1286 if (!reason) {
1287 consume_skb(skb);
1288 return NET_RX_SUCCESS;
1289 }
1290
1291drop:
1292 kfree_skb_reason(skb, reason);
1293 return NET_RX_DROP;
1294csum_error:
1295 reason = SKB_DROP_REASON_ICMP_CSUM;
1296 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1297error:
1298 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1299 goto drop;
1300}
1301
1302static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1303{
1304 struct icmp_extobj_hdr *objh, _objh;
1305 struct icmp_ext_hdr *exth, _exth;
1306 u16 olen;
1307
1308 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1309 if (!exth)
1310 return false;
1311 if (exth->version != 2)
1312 return true;
1313
1314 if (exth->checksum &&
1315 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1316 return false;
1317
1318 off += sizeof(_exth);
1319 while (off < skb->len) {
1320 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1321 if (!objh)
1322 return false;
1323
1324 olen = ntohs(objh->length);
1325 if (olen < sizeof(_objh))
1326 return false;
1327
1328 off += olen;
1329 if (off > skb->len)
1330 return false;
1331 }
1332
1333 return true;
1334}
1335
1336void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1337 struct sock_ee_data_rfc4884 *out,
1338 int thlen, int off)
1339{
1340 int hlen;
1341
1342 /* original datagram headers: end of icmph to payload (skb->data) */
1343 hlen = -skb_transport_offset(skb) - thlen;
1344
1345 /* per rfc 4884: minimal datagram length of 128 bytes */
1346 if (off < 128 || off < hlen)
1347 return;
1348
1349 /* kernel has stripped headers: return payload offset in bytes */
1350 off -= hlen;
1351 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1352 return;
1353
1354 out->len = off;
1355
1356 if (!ip_icmp_error_rfc4884_validate(skb, off))
1357 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1358}
1359EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1360
1361int icmp_err(struct sk_buff *skb, u32 info)
1362{
1363 struct iphdr *iph = (struct iphdr *)skb->data;
1364 int offset = iph->ihl<<2;
1365 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1366 int type = icmp_hdr(skb)->type;
1367 int code = icmp_hdr(skb)->code;
1368 struct net *net = dev_net(skb->dev);
1369
1370 /*
1371 * Use ping_err to handle all icmp errors except those
1372 * triggered by ICMP_ECHOREPLY which sent from kernel.
1373 */
1374 if (icmph->type != ICMP_ECHOREPLY) {
1375 ping_err(skb, offset, info);
1376 return 0;
1377 }
1378
1379 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1380 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1381 else if (type == ICMP_REDIRECT)
1382 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1383
1384 return 0;
1385}
1386
1387/*
1388 * This table is the definition of how we handle ICMP.
1389 */
1390static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1391 [ICMP_ECHOREPLY] = {
1392 .handler = ping_rcv,
1393 },
1394 [1] = {
1395 .handler = icmp_discard,
1396 .error = 1,
1397 },
1398 [2] = {
1399 .handler = icmp_discard,
1400 .error = 1,
1401 },
1402 [ICMP_DEST_UNREACH] = {
1403 .handler = icmp_unreach,
1404 .error = 1,
1405 },
1406 [ICMP_SOURCE_QUENCH] = {
1407 .handler = icmp_unreach,
1408 .error = 1,
1409 },
1410 [ICMP_REDIRECT] = {
1411 .handler = icmp_redirect,
1412 .error = 1,
1413 },
1414 [6] = {
1415 .handler = icmp_discard,
1416 .error = 1,
1417 },
1418 [7] = {
1419 .handler = icmp_discard,
1420 .error = 1,
1421 },
1422 [ICMP_ECHO] = {
1423 .handler = icmp_echo,
1424 },
1425 [9] = {
1426 .handler = icmp_discard,
1427 .error = 1,
1428 },
1429 [10] = {
1430 .handler = icmp_discard,
1431 .error = 1,
1432 },
1433 [ICMP_TIME_EXCEEDED] = {
1434 .handler = icmp_unreach,
1435 .error = 1,
1436 },
1437 [ICMP_PARAMETERPROB] = {
1438 .handler = icmp_unreach,
1439 .error = 1,
1440 },
1441 [ICMP_TIMESTAMP] = {
1442 .handler = icmp_timestamp,
1443 },
1444 [ICMP_TIMESTAMPREPLY] = {
1445 .handler = icmp_discard,
1446 },
1447 [ICMP_INFO_REQUEST] = {
1448 .handler = icmp_discard,
1449 },
1450 [ICMP_INFO_REPLY] = {
1451 .handler = icmp_discard,
1452 },
1453 [ICMP_ADDRESS] = {
1454 .handler = icmp_discard,
1455 },
1456 [ICMP_ADDRESSREPLY] = {
1457 .handler = icmp_discard,
1458 },
1459};
1460
1461static int __net_init icmp_sk_init(struct net *net)
1462{
1463 /* Control parameters for ECHO replies. */
1464 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1465 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1466 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1467
1468 /* Control parameter - ignore bogus broadcast responses? */
1469 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1470
1471 /*
1472 * Configurable global rate limit.
1473 *
1474 * ratelimit defines tokens/packet consumed for dst->rate_token
1475 * bucket ratemask defines which icmp types are ratelimited by
1476 * setting it's bit position.
1477 *
1478 * default:
1479 * dest unreachable (3), source quench (4),
1480 * time exceeded (11), parameter problem (12)
1481 */
1482
1483 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1484 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1485 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1486
1487 return 0;
1488}
1489
1490static struct pernet_operations __net_initdata icmp_sk_ops = {
1491 .init = icmp_sk_init,
1492};
1493
1494int __init icmp_init(void)
1495{
1496 int err, i;
1497
1498 for_each_possible_cpu(i) {
1499 struct sock *sk;
1500
1501 err = inet_ctl_sock_create(&sk, PF_INET,
1502 SOCK_RAW, IPPROTO_ICMP, &init_net);
1503 if (err < 0)
1504 return err;
1505
1506 per_cpu(ipv4_icmp_sk, i) = sk;
1507
1508 /* Enough space for 2 64K ICMP packets, including
1509 * sk_buff/skb_shared_info struct overhead.
1510 */
1511 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1512
1513 /*
1514 * Speedup sock_wfree()
1515 */
1516 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1517 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1518 }
1519 return register_pernet_subsys(&icmp_sk_ops);
1520}