Loading...
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 enum skb_drop_reason (*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
195static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
196
197/* Called with BH disabled */
198static inline struct sock *icmp_xmit_lock(struct net *net)
199{
200 struct sock *sk;
201
202 sk = this_cpu_read(ipv4_icmp_sk);
203
204 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
205 /* This can happen if the output path signals a
206 * dst_link_failure() for an outgoing ICMP packet.
207 */
208 return NULL;
209 }
210 sock_net_set(sk, net);
211 return sk;
212}
213
214static inline void icmp_xmit_unlock(struct sock *sk)
215{
216 sock_net_set(sk, &init_net);
217 spin_unlock(&sk->sk_lock.slock);
218}
219
220int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
221int sysctl_icmp_msgs_burst __read_mostly = 50;
222
223static struct {
224 spinlock_t lock;
225 u32 credit;
226 u32 stamp;
227} icmp_global = {
228 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
229};
230
231/**
232 * icmp_global_allow - Are we allowed to send one more ICMP message ?
233 *
234 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
235 * Returns false if we reached the limit and can not send another packet.
236 * Note: called with BH disabled
237 */
238bool icmp_global_allow(void)
239{
240 u32 credit, delta, incr = 0, now = (u32)jiffies;
241 bool rc = false;
242
243 /* Check if token bucket is empty and cannot be refilled
244 * without taking the spinlock. The READ_ONCE() are paired
245 * with the following WRITE_ONCE() in this same function.
246 */
247 if (!READ_ONCE(icmp_global.credit)) {
248 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
249 if (delta < HZ / 50)
250 return false;
251 }
252
253 spin_lock(&icmp_global.lock);
254 delta = min_t(u32, now - icmp_global.stamp, HZ);
255 if (delta >= HZ / 50) {
256 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
257 if (incr)
258 WRITE_ONCE(icmp_global.stamp, now);
259 }
260 credit = min_t(u32, icmp_global.credit + incr,
261 READ_ONCE(sysctl_icmp_msgs_burst));
262 if (credit) {
263 /* We want to use a credit of one in average, but need to randomize
264 * it for security reasons.
265 */
266 credit = max_t(int, credit - get_random_u32_below(3), 0);
267 rc = true;
268 }
269 WRITE_ONCE(icmp_global.credit, credit);
270 spin_unlock(&icmp_global.lock);
271 return rc;
272}
273EXPORT_SYMBOL(icmp_global_allow);
274
275static bool icmpv4_mask_allow(struct net *net, int type, int code)
276{
277 if (type > NR_ICMP_TYPES)
278 return true;
279
280 /* Don't limit PMTU discovery. */
281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
282 return true;
283
284 /* Limit if icmp type is enabled in ratemask. */
285 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
286 return true;
287
288 return false;
289}
290
291static bool icmpv4_global_allow(struct net *net, int type, int code)
292{
293 if (icmpv4_mask_allow(net, type, code))
294 return true;
295
296 if (icmp_global_allow())
297 return true;
298
299 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
300 return false;
301}
302
303/*
304 * Send an ICMP frame.
305 */
306
307static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
308 struct flowi4 *fl4, int type, int code)
309{
310 struct dst_entry *dst = &rt->dst;
311 struct inet_peer *peer;
312 bool rc = true;
313 int vif;
314
315 if (icmpv4_mask_allow(net, type, code))
316 goto out;
317
318 /* No rate limit on loopback */
319 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
320 goto out;
321
322 vif = l3mdev_master_ifindex(dst->dev);
323 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
324 rc = inet_peer_xrlim_allow(peer,
325 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
326 if (peer)
327 inet_putpeer(peer);
328out:
329 if (!rc)
330 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
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 = from;
351 __wsum csum;
352
353 csum = skb_copy_and_csum_bits(icmp_param->skb,
354 icmp_param->offset + offset,
355 to, len);
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 sock *sk,
364 struct icmp_bxm *icmp_param,
365 struct flowi4 *fl4,
366 struct ipcm_cookie *ipc, struct rtable **rt)
367{
368 struct sk_buff *skb;
369
370 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 icmp_param->data_len+icmp_param->head_len,
372 icmp_param->head_len,
373 ipc, rt, MSG_DONTWAIT) < 0) {
374 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 ip_flush_pending_frames(sk);
376 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 struct icmphdr *icmph = icmp_hdr(skb);
378 __wsum csum;
379 struct sk_buff *skb1;
380
381 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
382 (char *)icmph,
383 icmp_param->head_len);
384 skb_queue_walk(&sk->sk_write_queue, skb1) {
385 csum = csum_add(csum, skb1->csum);
386 }
387 icmph->checksum = csum_fold(csum);
388 skb->ip_summed = CHECKSUM_NONE;
389 ip_push_pending_frames(sk, fl4);
390 }
391}
392
393/*
394 * Driving logic for building and sending ICMP messages.
395 */
396
397static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
398{
399 struct ipcm_cookie ipc;
400 struct rtable *rt = skb_rtable(skb);
401 struct net *net = dev_net(rt->dst.dev);
402 struct flowi4 fl4;
403 struct sock *sk;
404 struct inet_sock *inet;
405 __be32 daddr, saddr;
406 u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 int type = icmp_param->data.icmph.type;
408 int code = icmp_param->data.icmph.code;
409
410 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 return;
412
413 /* Needed by both icmp_global_allow and icmp_xmit_lock */
414 local_bh_disable();
415
416 /* global icmp_msgs_per_sec */
417 if (!icmpv4_global_allow(net, type, code))
418 goto out_bh_enable;
419
420 sk = icmp_xmit_lock(net);
421 if (!sk)
422 goto out_bh_enable;
423 inet = inet_sk(sk);
424
425 icmp_param->data.icmph.checksum = 0;
426
427 ipcm_init(&ipc);
428 inet->tos = ip_hdr(skb)->tos;
429 ipc.sockc.mark = mark;
430 daddr = ipc.addr = ip_hdr(skb)->saddr;
431 saddr = fib_compute_spec_dst(skb);
432
433 if (icmp_param->replyopts.opt.opt.optlen) {
434 ipc.opt = &icmp_param->replyopts.opt;
435 if (ipc.opt->opt.srr)
436 daddr = icmp_param->replyopts.opt.opt.faddr;
437 }
438 memset(&fl4, 0, sizeof(fl4));
439 fl4.daddr = daddr;
440 fl4.saddr = saddr;
441 fl4.flowi4_mark = mark;
442 fl4.flowi4_uid = sock_net_uid(net, NULL);
443 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 fl4.flowi4_proto = IPPROTO_ICMP;
445 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
447 rt = ip_route_output_key(net, &fl4);
448 if (IS_ERR(rt))
449 goto out_unlock;
450 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
452 ip_rt_put(rt);
453out_unlock:
454 icmp_xmit_unlock(sk);
455out_bh_enable:
456 local_bh_enable();
457}
458
459/*
460 * The device used for looking up which routing table to use for sending an ICMP
461 * error is preferably the source whenever it is set, which should ensure the
462 * icmp error can be sent to the source host, else lookup using the routing
463 * table of the destination device, else use the main routing table (index 0).
464 */
465static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
466{
467 struct net_device *route_lookup_dev = NULL;
468
469 if (skb->dev)
470 route_lookup_dev = skb->dev;
471 else if (skb_dst(skb))
472 route_lookup_dev = skb_dst(skb)->dev;
473 return route_lookup_dev;
474}
475
476static struct rtable *icmp_route_lookup(struct net *net,
477 struct flowi4 *fl4,
478 struct sk_buff *skb_in,
479 const struct iphdr *iph,
480 __be32 saddr, u8 tos, u32 mark,
481 int type, int code,
482 struct icmp_bxm *param)
483{
484 struct net_device *route_lookup_dev;
485 struct rtable *rt, *rt2;
486 struct flowi4 fl4_dec;
487 int err;
488
489 memset(fl4, 0, sizeof(*fl4));
490 fl4->daddr = (param->replyopts.opt.opt.srr ?
491 param->replyopts.opt.opt.faddr : iph->saddr);
492 fl4->saddr = saddr;
493 fl4->flowi4_mark = mark;
494 fl4->flowi4_uid = sock_net_uid(net, NULL);
495 fl4->flowi4_tos = RT_TOS(tos);
496 fl4->flowi4_proto = IPPROTO_ICMP;
497 fl4->fl4_icmp_type = type;
498 fl4->fl4_icmp_code = code;
499 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
500 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
501
502 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
503 rt = ip_route_output_key_hash(net, fl4, skb_in);
504 if (IS_ERR(rt))
505 return rt;
506
507 /* No need to clone since we're just using its address. */
508 rt2 = rt;
509
510 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
511 flowi4_to_flowi(fl4), NULL, 0);
512 if (!IS_ERR(rt)) {
513 if (rt != rt2)
514 return rt;
515 } else if (PTR_ERR(rt) == -EPERM) {
516 rt = NULL;
517 } else
518 return rt;
519
520 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
521 if (err)
522 goto relookup_failed;
523
524 if (inet_addr_type_dev_table(net, route_lookup_dev,
525 fl4_dec.saddr) == RTN_LOCAL) {
526 rt2 = __ip_route_output_key(net, &fl4_dec);
527 if (IS_ERR(rt2))
528 err = PTR_ERR(rt2);
529 } else {
530 struct flowi4 fl4_2 = {};
531 unsigned long orefdst;
532
533 fl4_2.daddr = fl4_dec.saddr;
534 rt2 = ip_route_output_key(net, &fl4_2);
535 if (IS_ERR(rt2)) {
536 err = PTR_ERR(rt2);
537 goto relookup_failed;
538 }
539 /* Ugh! */
540 orefdst = skb_in->_skb_refdst; /* save old refdst */
541 skb_dst_set(skb_in, NULL);
542 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
543 RT_TOS(tos), rt2->dst.dev);
544
545 dst_release(&rt2->dst);
546 rt2 = skb_rtable(skb_in);
547 skb_in->_skb_refdst = orefdst; /* restore old refdst */
548 }
549
550 if (err)
551 goto relookup_failed;
552
553 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
554 flowi4_to_flowi(&fl4_dec), NULL,
555 XFRM_LOOKUP_ICMP);
556 if (!IS_ERR(rt2)) {
557 dst_release(&rt->dst);
558 memcpy(fl4, &fl4_dec, sizeof(*fl4));
559 rt = rt2;
560 } else if (PTR_ERR(rt2) == -EPERM) {
561 if (rt)
562 dst_release(&rt->dst);
563 return rt2;
564 } else {
565 err = PTR_ERR(rt2);
566 goto relookup_failed;
567 }
568 return rt;
569
570relookup_failed:
571 if (rt)
572 return rt;
573 return ERR_PTR(err);
574}
575
576/*
577 * Send an ICMP message in response to a situation
578 *
579 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
580 * MAY send more (we do).
581 * MUST NOT change this header information.
582 * MUST NOT reply to a multicast/broadcast IP address.
583 * MUST NOT reply to a multicast/broadcast MAC address.
584 * MUST reply to only the first fragment.
585 */
586
587void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
588 const struct ip_options *opt)
589{
590 struct iphdr *iph;
591 int room;
592 struct icmp_bxm icmp_param;
593 struct rtable *rt = skb_rtable(skb_in);
594 struct ipcm_cookie ipc;
595 struct flowi4 fl4;
596 __be32 saddr;
597 u8 tos;
598 u32 mark;
599 struct net *net;
600 struct sock *sk;
601
602 if (!rt)
603 goto out;
604
605 if (rt->dst.dev)
606 net = dev_net(rt->dst.dev);
607 else if (skb_in->dev)
608 net = dev_net(skb_in->dev);
609 else
610 goto out;
611
612 /*
613 * Find the original header. It is expected to be valid, of course.
614 * Check this, icmp_send is called from the most obscure devices
615 * sometimes.
616 */
617 iph = ip_hdr(skb_in);
618
619 if ((u8 *)iph < skb_in->head ||
620 (skb_network_header(skb_in) + sizeof(*iph)) >
621 skb_tail_pointer(skb_in))
622 goto out;
623
624 /*
625 * No replies to physical multicast/broadcast
626 */
627 if (skb_in->pkt_type != PACKET_HOST)
628 goto out;
629
630 /*
631 * Now check at the protocol level
632 */
633 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
634 goto out;
635
636 /*
637 * Only reply to fragment 0. We byte re-order the constant
638 * mask for efficiency.
639 */
640 if (iph->frag_off & htons(IP_OFFSET))
641 goto out;
642
643 /*
644 * If we send an ICMP error to an ICMP error a mess would result..
645 */
646 if (icmp_pointers[type].error) {
647 /*
648 * We are an error, check if we are replying to an
649 * ICMP error
650 */
651 if (iph->protocol == IPPROTO_ICMP) {
652 u8 _inner_type, *itp;
653
654 itp = skb_header_pointer(skb_in,
655 skb_network_header(skb_in) +
656 (iph->ihl << 2) +
657 offsetof(struct icmphdr,
658 type) -
659 skb_in->data,
660 sizeof(_inner_type),
661 &_inner_type);
662 if (!itp)
663 goto out;
664
665 /*
666 * Assume any unknown ICMP type is an error. This
667 * isn't specified by the RFC, but think about it..
668 */
669 if (*itp > NR_ICMP_TYPES ||
670 icmp_pointers[*itp].error)
671 goto out;
672 }
673 }
674
675 /* Needed by both icmp_global_allow and icmp_xmit_lock */
676 local_bh_disable();
677
678 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
679 * incoming dev is loopback. If outgoing dev change to not be
680 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
681 */
682 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
683 !icmpv4_global_allow(net, type, code))
684 goto out_bh_enable;
685
686 sk = icmp_xmit_lock(net);
687 if (!sk)
688 goto out_bh_enable;
689
690 /*
691 * Construct source address and options.
692 */
693
694 saddr = iph->daddr;
695 if (!(rt->rt_flags & RTCF_LOCAL)) {
696 struct net_device *dev = NULL;
697
698 rcu_read_lock();
699 if (rt_is_input_route(rt) &&
700 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
701 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
702
703 if (dev)
704 saddr = inet_select_addr(dev, iph->saddr,
705 RT_SCOPE_LINK);
706 else
707 saddr = 0;
708 rcu_read_unlock();
709 }
710
711 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
712 IPTOS_PREC_INTERNETCONTROL) :
713 iph->tos;
714 mark = IP4_REPLY_MARK(net, skb_in->mark);
715
716 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
717 goto out_unlock;
718
719
720 /*
721 * Prepare data for ICMP header.
722 */
723
724 icmp_param.data.icmph.type = type;
725 icmp_param.data.icmph.code = code;
726 icmp_param.data.icmph.un.gateway = info;
727 icmp_param.data.icmph.checksum = 0;
728 icmp_param.skb = skb_in;
729 icmp_param.offset = skb_network_offset(skb_in);
730 inet_sk(sk)->tos = tos;
731 ipcm_init(&ipc);
732 ipc.addr = iph->saddr;
733 ipc.opt = &icmp_param.replyopts.opt;
734 ipc.sockc.mark = mark;
735
736 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
737 type, code, &icmp_param);
738 if (IS_ERR(rt))
739 goto out_unlock;
740
741 /* peer icmp_ratelimit */
742 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
743 goto ende;
744
745 /* RFC says return as much as we can without exceeding 576 bytes. */
746
747 room = dst_mtu(&rt->dst);
748 if (room > 576)
749 room = 576;
750 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
751 room -= sizeof(struct icmphdr);
752 /* Guard against tiny mtu. We need to include at least one
753 * IP network header for this message to make any sense.
754 */
755 if (room <= (int)sizeof(struct iphdr))
756 goto ende;
757
758 icmp_param.data_len = skb_in->len - icmp_param.offset;
759 if (icmp_param.data_len > room)
760 icmp_param.data_len = room;
761 icmp_param.head_len = sizeof(struct icmphdr);
762
763 /* if we don't have a source address at this point, fall back to the
764 * dummy address instead of sending out a packet with a source address
765 * of 0.0.0.0
766 */
767 if (!fl4.saddr)
768 fl4.saddr = htonl(INADDR_DUMMY);
769
770 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
771ende:
772 ip_rt_put(rt);
773out_unlock:
774 icmp_xmit_unlock(sk);
775out_bh_enable:
776 local_bh_enable();
777out:;
778}
779EXPORT_SYMBOL(__icmp_send);
780
781#if IS_ENABLED(CONFIG_NF_NAT)
782#include <net/netfilter/nf_conntrack.h>
783void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
784{
785 struct sk_buff *cloned_skb = NULL;
786 struct ip_options opts = { 0 };
787 enum ip_conntrack_info ctinfo;
788 struct nf_conn *ct;
789 __be32 orig_ip;
790
791 ct = nf_ct_get(skb_in, &ctinfo);
792 if (!ct || !(ct->status & IPS_SRC_NAT)) {
793 __icmp_send(skb_in, type, code, info, &opts);
794 return;
795 }
796
797 if (skb_shared(skb_in))
798 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
799
800 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
801 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
802 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
803 skb_network_offset(skb_in) + sizeof(struct iphdr))))
804 goto out;
805
806 orig_ip = ip_hdr(skb_in)->saddr;
807 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
808 __icmp_send(skb_in, type, code, info, &opts);
809 ip_hdr(skb_in)->saddr = orig_ip;
810out:
811 consume_skb(cloned_skb);
812}
813EXPORT_SYMBOL(icmp_ndo_send);
814#endif
815
816static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
817{
818 const struct iphdr *iph = (const struct iphdr *)skb->data;
819 const struct net_protocol *ipprot;
820 int protocol = iph->protocol;
821
822 /* Checkin full IP header plus 8 bytes of protocol to
823 * avoid additional coding at protocol handlers.
824 */
825 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
826 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
827 return;
828 }
829
830 raw_icmp_error(skb, protocol, info);
831
832 ipprot = rcu_dereference(inet_protos[protocol]);
833 if (ipprot && ipprot->err_handler)
834 ipprot->err_handler(skb, info);
835}
836
837static bool icmp_tag_validation(int proto)
838{
839 bool ok;
840
841 rcu_read_lock();
842 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
843 rcu_read_unlock();
844 return ok;
845}
846
847/*
848 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
849 * ICMP_PARAMETERPROB.
850 */
851
852static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
853{
854 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
855 const struct iphdr *iph;
856 struct icmphdr *icmph;
857 struct net *net;
858 u32 info = 0;
859
860 net = dev_net(skb_dst(skb)->dev);
861
862 /*
863 * Incomplete header ?
864 * Only checks for the IP header, there should be an
865 * additional check for longer headers in upper levels.
866 */
867
868 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
869 goto out_err;
870
871 icmph = icmp_hdr(skb);
872 iph = (const struct iphdr *)skb->data;
873
874 if (iph->ihl < 5) { /* Mangled header, drop. */
875 reason = SKB_DROP_REASON_IP_INHDR;
876 goto out_err;
877 }
878
879 switch (icmph->type) {
880 case ICMP_DEST_UNREACH:
881 switch (icmph->code & 15) {
882 case ICMP_NET_UNREACH:
883 case ICMP_HOST_UNREACH:
884 case ICMP_PROT_UNREACH:
885 case ICMP_PORT_UNREACH:
886 break;
887 case ICMP_FRAG_NEEDED:
888 /* for documentation of the ip_no_pmtu_disc
889 * values please see
890 * Documentation/networking/ip-sysctl.rst
891 */
892 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
893 default:
894 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
895 &iph->daddr);
896 break;
897 case 2:
898 goto out;
899 case 3:
900 if (!icmp_tag_validation(iph->protocol))
901 goto out;
902 fallthrough;
903 case 0:
904 info = ntohs(icmph->un.frag.mtu);
905 }
906 break;
907 case ICMP_SR_FAILED:
908 net_dbg_ratelimited("%pI4: Source Route Failed\n",
909 &iph->daddr);
910 break;
911 default:
912 break;
913 }
914 if (icmph->code > NR_ICMP_UNREACH)
915 goto out;
916 break;
917 case ICMP_PARAMETERPROB:
918 info = ntohl(icmph->un.gateway) >> 24;
919 break;
920 case ICMP_TIME_EXCEEDED:
921 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
922 if (icmph->code == ICMP_EXC_FRAGTIME)
923 goto out;
924 break;
925 }
926
927 /*
928 * Throw it at our lower layers
929 *
930 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
931 * header.
932 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
933 * transport layer.
934 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
935 * transport layer.
936 */
937
938 /*
939 * Check the other end isn't violating RFC 1122. Some routers send
940 * bogus responses to broadcast frames. If you see this message
941 * first check your netmask matches at both ends, if it does then
942 * get the other vendor to fix their kit.
943 */
944
945 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
946 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
947 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
948 &ip_hdr(skb)->saddr,
949 icmph->type, icmph->code,
950 &iph->daddr, skb->dev->name);
951 goto out;
952 }
953
954 icmp_socket_deliver(skb, info);
955
956out:
957 return reason;
958out_err:
959 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
960 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
961}
962
963
964/*
965 * Handle ICMP_REDIRECT.
966 */
967
968static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
969{
970 if (skb->len < sizeof(struct iphdr)) {
971 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
972 return SKB_DROP_REASON_PKT_TOO_SMALL;
973 }
974
975 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
976 /* there aught to be a stat */
977 return SKB_DROP_REASON_NOMEM;
978 }
979
980 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
981 return SKB_NOT_DROPPED_YET;
982}
983
984/*
985 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
986 *
987 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
988 * requests.
989 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
990 * included in the reply.
991 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
992 * echo requests, MUST have default=NOT.
993 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
994 * Extended Echo Functionality, MUST be disabled by default
995 * See also WRT handling of options once they are done and working.
996 */
997
998static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
999{
1000 struct icmp_bxm icmp_param;
1001 struct net *net;
1002
1003 net = dev_net(skb_dst(skb)->dev);
1004 /* should there be an ICMP stat for ignored echos? */
1005 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1006 return SKB_NOT_DROPPED_YET;
1007
1008 icmp_param.data.icmph = *icmp_hdr(skb);
1009 icmp_param.skb = skb;
1010 icmp_param.offset = 0;
1011 icmp_param.data_len = skb->len;
1012 icmp_param.head_len = sizeof(struct icmphdr);
1013
1014 if (icmp_param.data.icmph.type == ICMP_ECHO)
1015 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1016 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1017 return SKB_NOT_DROPPED_YET;
1018
1019 icmp_reply(&icmp_param, skb);
1020 return SKB_NOT_DROPPED_YET;
1021}
1022
1023/* Helper for icmp_echo and icmpv6_echo_reply.
1024 * Searches for net_device that matches PROBE interface identifier
1025 * and builds PROBE reply message in icmphdr.
1026 *
1027 * Returns false if PROBE responses are disabled via sysctl
1028 */
1029
1030bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1031{
1032 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1033 struct icmp_ext_echo_iio *iio, _iio;
1034 struct net *net = dev_net(skb->dev);
1035 struct net_device *dev;
1036 char buff[IFNAMSIZ];
1037 u16 ident_len;
1038 u8 status;
1039
1040 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1041 return false;
1042
1043 /* We currently only support probing interfaces on the proxy node
1044 * Check to ensure L-bit is set
1045 */
1046 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1047 return false;
1048 /* Clear status bits in reply message */
1049 icmphdr->un.echo.sequence &= htons(0xFF00);
1050 if (icmphdr->type == ICMP_EXT_ECHO)
1051 icmphdr->type = ICMP_EXT_ECHOREPLY;
1052 else
1053 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1054 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1055 /* Size of iio is class_type dependent.
1056 * Only check header here and assign length based on ctype in the switch statement
1057 */
1058 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1059 if (!ext_hdr || !iio)
1060 goto send_mal_query;
1061 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1062 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1063 goto send_mal_query;
1064 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1065 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1066 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1067 if (!iio)
1068 goto send_mal_query;
1069
1070 status = 0;
1071 dev = NULL;
1072 switch (iio->extobj_hdr.class_type) {
1073 case ICMP_EXT_ECHO_CTYPE_NAME:
1074 if (ident_len >= IFNAMSIZ)
1075 goto send_mal_query;
1076 memset(buff, 0, sizeof(buff));
1077 memcpy(buff, &iio->ident.name, ident_len);
1078 dev = dev_get_by_name(net, buff);
1079 break;
1080 case ICMP_EXT_ECHO_CTYPE_INDEX:
1081 if (ident_len != sizeof(iio->ident.ifindex))
1082 goto send_mal_query;
1083 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1084 break;
1085 case ICMP_EXT_ECHO_CTYPE_ADDR:
1086 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1087 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1088 iio->ident.addr.ctype3_hdr.addrlen)
1089 goto send_mal_query;
1090 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1091 case ICMP_AFI_IP:
1092 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1093 goto send_mal_query;
1094 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1095 break;
1096#if IS_ENABLED(CONFIG_IPV6)
1097 case ICMP_AFI_IP6:
1098 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1099 goto send_mal_query;
1100 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1101 dev_hold(dev);
1102 break;
1103#endif
1104 default:
1105 goto send_mal_query;
1106 }
1107 break;
1108 default:
1109 goto send_mal_query;
1110 }
1111 if (!dev) {
1112 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1113 return true;
1114 }
1115 /* Fill bits in reply message */
1116 if (dev->flags & IFF_UP)
1117 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1118 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1119 status |= ICMP_EXT_ECHOREPLY_IPV4;
1120 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1121 status |= ICMP_EXT_ECHOREPLY_IPV6;
1122 dev_put(dev);
1123 icmphdr->un.echo.sequence |= htons(status);
1124 return true;
1125send_mal_query:
1126 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1127 return true;
1128}
1129EXPORT_SYMBOL_GPL(icmp_build_probe);
1130
1131/*
1132 * Handle ICMP Timestamp requests.
1133 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1134 * SHOULD be in the kernel for minimum random latency.
1135 * MUST be accurate to a few minutes.
1136 * MUST be updated at least at 15Hz.
1137 */
1138static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1139{
1140 struct icmp_bxm icmp_param;
1141 /*
1142 * Too short.
1143 */
1144 if (skb->len < 4)
1145 goto out_err;
1146
1147 /*
1148 * Fill in the current time as ms since midnight UT:
1149 */
1150 icmp_param.data.times[1] = inet_current_timestamp();
1151 icmp_param.data.times[2] = icmp_param.data.times[1];
1152
1153 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1154
1155 icmp_param.data.icmph = *icmp_hdr(skb);
1156 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1157 icmp_param.data.icmph.code = 0;
1158 icmp_param.skb = skb;
1159 icmp_param.offset = 0;
1160 icmp_param.data_len = 0;
1161 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1162 icmp_reply(&icmp_param, skb);
1163 return SKB_NOT_DROPPED_YET;
1164
1165out_err:
1166 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1167 return SKB_DROP_REASON_PKT_TOO_SMALL;
1168}
1169
1170static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1171{
1172 /* pretend it was a success */
1173 return SKB_NOT_DROPPED_YET;
1174}
1175
1176/*
1177 * Deal with incoming ICMP packets.
1178 */
1179int icmp_rcv(struct sk_buff *skb)
1180{
1181 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1182 struct rtable *rt = skb_rtable(skb);
1183 struct net *net = dev_net(rt->dst.dev);
1184 struct icmphdr *icmph;
1185
1186 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1187 struct sec_path *sp = skb_sec_path(skb);
1188 int nh;
1189
1190 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1191 XFRM_STATE_ICMP)) {
1192 reason = SKB_DROP_REASON_XFRM_POLICY;
1193 goto drop;
1194 }
1195
1196 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1197 goto drop;
1198
1199 nh = skb_network_offset(skb);
1200 skb_set_network_header(skb, sizeof(*icmph));
1201
1202 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1203 skb)) {
1204 reason = SKB_DROP_REASON_XFRM_POLICY;
1205 goto drop;
1206 }
1207
1208 skb_set_network_header(skb, nh);
1209 }
1210
1211 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1212
1213 if (skb_checksum_simple_validate(skb))
1214 goto csum_error;
1215
1216 if (!pskb_pull(skb, sizeof(*icmph)))
1217 goto error;
1218
1219 icmph = icmp_hdr(skb);
1220
1221 ICMPMSGIN_INC_STATS(net, icmph->type);
1222
1223 /* Check for ICMP Extended Echo (PROBE) messages */
1224 if (icmph->type == ICMP_EXT_ECHO) {
1225 /* We can't use icmp_pointers[].handler() because it is an array of
1226 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1227 */
1228 reason = icmp_echo(skb);
1229 goto reason_check;
1230 }
1231
1232 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1233 reason = ping_rcv(skb);
1234 goto reason_check;
1235 }
1236
1237 /*
1238 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1239 *
1240 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1241 * discarded.
1242 */
1243 if (icmph->type > NR_ICMP_TYPES) {
1244 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1245 goto error;
1246 }
1247
1248 /*
1249 * Parse the ICMP message
1250 */
1251
1252 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1253 /*
1254 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1255 * silently ignored (we let user decide with a sysctl).
1256 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1257 * discarded if to broadcast/multicast.
1258 */
1259 if ((icmph->type == ICMP_ECHO ||
1260 icmph->type == ICMP_TIMESTAMP) &&
1261 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1262 reason = SKB_DROP_REASON_INVALID_PROTO;
1263 goto error;
1264 }
1265 if (icmph->type != ICMP_ECHO &&
1266 icmph->type != ICMP_TIMESTAMP &&
1267 icmph->type != ICMP_ADDRESS &&
1268 icmph->type != ICMP_ADDRESSREPLY) {
1269 reason = SKB_DROP_REASON_INVALID_PROTO;
1270 goto error;
1271 }
1272 }
1273
1274 reason = icmp_pointers[icmph->type].handler(skb);
1275reason_check:
1276 if (!reason) {
1277 consume_skb(skb);
1278 return NET_RX_SUCCESS;
1279 }
1280
1281drop:
1282 kfree_skb_reason(skb, reason);
1283 return NET_RX_DROP;
1284csum_error:
1285 reason = SKB_DROP_REASON_ICMP_CSUM;
1286 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1287error:
1288 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1289 goto drop;
1290}
1291
1292static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1293{
1294 struct icmp_extobj_hdr *objh, _objh;
1295 struct icmp_ext_hdr *exth, _exth;
1296 u16 olen;
1297
1298 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1299 if (!exth)
1300 return false;
1301 if (exth->version != 2)
1302 return true;
1303
1304 if (exth->checksum &&
1305 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1306 return false;
1307
1308 off += sizeof(_exth);
1309 while (off < skb->len) {
1310 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1311 if (!objh)
1312 return false;
1313
1314 olen = ntohs(objh->length);
1315 if (olen < sizeof(_objh))
1316 return false;
1317
1318 off += olen;
1319 if (off > skb->len)
1320 return false;
1321 }
1322
1323 return true;
1324}
1325
1326void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1327 struct sock_ee_data_rfc4884 *out,
1328 int thlen, int off)
1329{
1330 int hlen;
1331
1332 /* original datagram headers: end of icmph to payload (skb->data) */
1333 hlen = -skb_transport_offset(skb) - thlen;
1334
1335 /* per rfc 4884: minimal datagram length of 128 bytes */
1336 if (off < 128 || off < hlen)
1337 return;
1338
1339 /* kernel has stripped headers: return payload offset in bytes */
1340 off -= hlen;
1341 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1342 return;
1343
1344 out->len = off;
1345
1346 if (!ip_icmp_error_rfc4884_validate(skb, off))
1347 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1348}
1349EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1350
1351int icmp_err(struct sk_buff *skb, u32 info)
1352{
1353 struct iphdr *iph = (struct iphdr *)skb->data;
1354 int offset = iph->ihl<<2;
1355 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1356 int type = icmp_hdr(skb)->type;
1357 int code = icmp_hdr(skb)->code;
1358 struct net *net = dev_net(skb->dev);
1359
1360 /*
1361 * Use ping_err to handle all icmp errors except those
1362 * triggered by ICMP_ECHOREPLY which sent from kernel.
1363 */
1364 if (icmph->type != ICMP_ECHOREPLY) {
1365 ping_err(skb, offset, info);
1366 return 0;
1367 }
1368
1369 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1370 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1371 else if (type == ICMP_REDIRECT)
1372 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1373
1374 return 0;
1375}
1376
1377/*
1378 * This table is the definition of how we handle ICMP.
1379 */
1380static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1381 [ICMP_ECHOREPLY] = {
1382 .handler = ping_rcv,
1383 },
1384 [1] = {
1385 .handler = icmp_discard,
1386 .error = 1,
1387 },
1388 [2] = {
1389 .handler = icmp_discard,
1390 .error = 1,
1391 },
1392 [ICMP_DEST_UNREACH] = {
1393 .handler = icmp_unreach,
1394 .error = 1,
1395 },
1396 [ICMP_SOURCE_QUENCH] = {
1397 .handler = icmp_unreach,
1398 .error = 1,
1399 },
1400 [ICMP_REDIRECT] = {
1401 .handler = icmp_redirect,
1402 .error = 1,
1403 },
1404 [6] = {
1405 .handler = icmp_discard,
1406 .error = 1,
1407 },
1408 [7] = {
1409 .handler = icmp_discard,
1410 .error = 1,
1411 },
1412 [ICMP_ECHO] = {
1413 .handler = icmp_echo,
1414 },
1415 [9] = {
1416 .handler = icmp_discard,
1417 .error = 1,
1418 },
1419 [10] = {
1420 .handler = icmp_discard,
1421 .error = 1,
1422 },
1423 [ICMP_TIME_EXCEEDED] = {
1424 .handler = icmp_unreach,
1425 .error = 1,
1426 },
1427 [ICMP_PARAMETERPROB] = {
1428 .handler = icmp_unreach,
1429 .error = 1,
1430 },
1431 [ICMP_TIMESTAMP] = {
1432 .handler = icmp_timestamp,
1433 },
1434 [ICMP_TIMESTAMPREPLY] = {
1435 .handler = icmp_discard,
1436 },
1437 [ICMP_INFO_REQUEST] = {
1438 .handler = icmp_discard,
1439 },
1440 [ICMP_INFO_REPLY] = {
1441 .handler = icmp_discard,
1442 },
1443 [ICMP_ADDRESS] = {
1444 .handler = icmp_discard,
1445 },
1446 [ICMP_ADDRESSREPLY] = {
1447 .handler = icmp_discard,
1448 },
1449};
1450
1451static int __net_init icmp_sk_init(struct net *net)
1452{
1453 /* Control parameters for ECHO replies. */
1454 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1455 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1456 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1457
1458 /* Control parameter - ignore bogus broadcast responses? */
1459 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1460
1461 /*
1462 * Configurable global rate limit.
1463 *
1464 * ratelimit defines tokens/packet consumed for dst->rate_token
1465 * bucket ratemask defines which icmp types are ratelimited by
1466 * setting it's bit position.
1467 *
1468 * default:
1469 * dest unreachable (3), source quench (4),
1470 * time exceeded (11), parameter problem (12)
1471 */
1472
1473 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1474 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1475 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1476
1477 return 0;
1478}
1479
1480static struct pernet_operations __net_initdata icmp_sk_ops = {
1481 .init = icmp_sk_init,
1482};
1483
1484int __init icmp_init(void)
1485{
1486 int err, i;
1487
1488 for_each_possible_cpu(i) {
1489 struct sock *sk;
1490
1491 err = inet_ctl_sock_create(&sk, PF_INET,
1492 SOCK_RAW, IPPROTO_ICMP, &init_net);
1493 if (err < 0)
1494 return err;
1495
1496 per_cpu(ipv4_icmp_sk, i) = sk;
1497
1498 /* Enough space for 2 64K ICMP packets, including
1499 * sk_buff/skb_shared_info struct overhead.
1500 */
1501 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1502
1503 /*
1504 * Speedup sock_wfree()
1505 */
1506 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1507 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1508 }
1509 return register_pernet_subsys(&icmp_sk_ops);
1510}
1/*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * Some of the function names and the icmp unreach table for this
12 * module were derived from [icmp.c 1.0.11 06/02/93] by
13 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
14 * Other than that this module is a complete rewrite.
15 *
16 * Fixes:
17 * Clemens Fruhwirth : introduce global icmp rate limiting
18 * with icmp type masking ability instead
19 * of broken per type icmp timeouts.
20 * Mike Shaver : RFC1122 checks.
21 * Alan Cox : Multicast ping reply as self.
22 * Alan Cox : Fix atomicity lockup in ip_build_xmit
23 * call.
24 * Alan Cox : Added 216,128 byte paths to the MTU
25 * code.
26 * Martin Mares : RFC1812 checks.
27 * Martin Mares : Can be configured to follow redirects
28 * if acting as a router _without_ a
29 * routing protocol (RFC 1812).
30 * Martin Mares : Echo requests may be configured to
31 * be ignored (RFC 1812).
32 * Martin Mares : Limitation of ICMP error message
33 * transmit rate (RFC 1812).
34 * Martin Mares : TOS and Precedence set correctly
35 * (RFC 1812).
36 * Martin Mares : Now copying as much data from the
37 * original packet as we can without
38 * exceeding 576 bytes (RFC 1812).
39 * Willy Konynenberg : Transparent proxying support.
40 * Keith Owens : RFC1191 correction for 4.2BSD based
41 * path MTU bug.
42 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
43 * valid (RFC 1812).
44 * Andi Kleen : Check all packet lengths properly
45 * and moved all kfree_skb() up to
46 * icmp_rcv.
47 * Andi Kleen : Move the rate limit bookkeeping
48 * into the dest entry and use a token
49 * bucket filter (thanks to ANK). Make
50 * the rates sysctl configurable.
51 * Yu Tianli : Fixed two ugly bugs in icmp_send
52 * - IP option length was accounted wrongly
53 * - ICMP header length was not accounted
54 * at all.
55 * Tristan Greaves : Added sysctl option to ignore bogus
56 * broadcast responses from broken routers.
57 *
58 * To Fix:
59 *
60 * - Should use skb_pull() instead of all the manual checking.
61 * This would also greatly simply some upper layer error handlers. --AK
62 *
63 */
64
65#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
66
67#include <linux/module.h>
68#include <linux/types.h>
69#include <linux/jiffies.h>
70#include <linux/kernel.h>
71#include <linux/fcntl.h>
72#include <linux/socket.h>
73#include <linux/in.h>
74#include <linux/inet.h>
75#include <linux/inetdevice.h>
76#include <linux/netdevice.h>
77#include <linux/string.h>
78#include <linux/netfilter_ipv4.h>
79#include <linux/slab.h>
80#include <net/snmp.h>
81#include <net/ip.h>
82#include <net/route.h>
83#include <net/protocol.h>
84#include <net/icmp.h>
85#include <net/tcp.h>
86#include <net/udp.h>
87#include <net/raw.h>
88#include <net/ping.h>
89#include <linux/skbuff.h>
90#include <net/sock.h>
91#include <linux/errno.h>
92#include <linux/timer.h>
93#include <linux/init.h>
94#include <asm/uaccess.h>
95#include <net/checksum.h>
96#include <net/xfrm.h>
97#include <net/inet_common.h>
98
99/*
100 * Build xmit assembly blocks
101 */
102
103struct icmp_bxm {
104 struct sk_buff *skb;
105 int offset;
106 int data_len;
107
108 struct {
109 struct icmphdr icmph;
110 __be32 times[3];
111 } data;
112 int head_len;
113 struct ip_options_data replyopts;
114};
115
116/* An array of errno for error messages from dest unreach. */
117/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
118
119const struct icmp_err icmp_err_convert[] = {
120 {
121 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
122 .fatal = 0,
123 },
124 {
125 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
126 .fatal = 0,
127 },
128 {
129 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
130 .fatal = 1,
131 },
132 {
133 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
134 .fatal = 1,
135 },
136 {
137 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
138 .fatal = 0,
139 },
140 {
141 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
142 .fatal = 0,
143 },
144 {
145 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
146 .fatal = 1,
147 },
148 {
149 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
150 .fatal = 1,
151 },
152 {
153 .errno = ENONET, /* ICMP_HOST_ISOLATED */
154 .fatal = 1,
155 },
156 {
157 .errno = ENETUNREACH, /* ICMP_NET_ANO */
158 .fatal = 1,
159 },
160 {
161 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
162 .fatal = 1,
163 },
164 {
165 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
166 .fatal = 0,
167 },
168 {
169 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
170 .fatal = 0,
171 },
172 {
173 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
174 .fatal = 1,
175 },
176 {
177 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
178 .fatal = 1,
179 },
180 {
181 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
182 .fatal = 1,
183 },
184};
185EXPORT_SYMBOL(icmp_err_convert);
186
187/*
188 * ICMP control array. This specifies what to do with each ICMP.
189 */
190
191struct icmp_control {
192 void (*handler)(struct sk_buff *skb);
193 short error; /* This ICMP is classed as an error message */
194};
195
196static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
197
198/*
199 * The ICMP socket(s). This is the most convenient way to flow control
200 * our ICMP output as well as maintain a clean interface throughout
201 * all layers. All Socketless IP sends will soon be gone.
202 *
203 * On SMP we have one ICMP socket per-cpu.
204 */
205static struct sock *icmp_sk(struct net *net)
206{
207 return net->ipv4.icmp_sk[smp_processor_id()];
208}
209
210static inline struct sock *icmp_xmit_lock(struct net *net)
211{
212 struct sock *sk;
213
214 local_bh_disable();
215
216 sk = icmp_sk(net);
217
218 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 /* This can happen if the output path signals a
220 * dst_link_failure() for an outgoing ICMP packet.
221 */
222 local_bh_enable();
223 return NULL;
224 }
225 return sk;
226}
227
228static inline void icmp_xmit_unlock(struct sock *sk)
229{
230 spin_unlock_bh(&sk->sk_lock.slock);
231}
232
233/*
234 * Send an ICMP frame.
235 */
236
237static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
238 struct flowi4 *fl4, int type, int code)
239{
240 struct dst_entry *dst = &rt->dst;
241 bool rc = true;
242
243 if (type > NR_ICMP_TYPES)
244 goto out;
245
246 /* Don't limit PMTU discovery. */
247 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
248 goto out;
249
250 /* No rate limit on loopback */
251 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
252 goto out;
253
254 /* Limit if icmp type is enabled in ratemask. */
255 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
256 if (!rt->peer)
257 rt_bind_peer(rt, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(rt->peer,
259 net->ipv4.sysctl_icmp_ratelimit);
260 }
261out:
262 return rc;
263}
264
265/*
266 * Maintain the counters used in the SNMP statistics for outgoing ICMP
267 */
268void icmp_out_count(struct net *net, unsigned char type)
269{
270 ICMPMSGOUT_INC_STATS(net, type);
271 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
272}
273
274/*
275 * Checksum each fragment, and on the first include the headers and final
276 * checksum.
277 */
278static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
279 struct sk_buff *skb)
280{
281 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
282 __wsum csum;
283
284 csum = skb_copy_and_csum_bits(icmp_param->skb,
285 icmp_param->offset + offset,
286 to, len, 0);
287
288 skb->csum = csum_block_add(skb->csum, csum, odd);
289 if (icmp_pointers[icmp_param->data.icmph.type].error)
290 nf_ct_attach(skb, icmp_param->skb);
291 return 0;
292}
293
294static void icmp_push_reply(struct icmp_bxm *icmp_param,
295 struct flowi4 *fl4,
296 struct ipcm_cookie *ipc, struct rtable **rt)
297{
298 struct sock *sk;
299 struct sk_buff *skb;
300
301 sk = icmp_sk(dev_net((*rt)->dst.dev));
302 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
303 icmp_param->data_len+icmp_param->head_len,
304 icmp_param->head_len,
305 ipc, rt, MSG_DONTWAIT) < 0) {
306 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
307 ip_flush_pending_frames(sk);
308 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
309 struct icmphdr *icmph = icmp_hdr(skb);
310 __wsum csum = 0;
311 struct sk_buff *skb1;
312
313 skb_queue_walk(&sk->sk_write_queue, skb1) {
314 csum = csum_add(csum, skb1->csum);
315 }
316 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
317 (char *)icmph,
318 icmp_param->head_len, csum);
319 icmph->checksum = csum_fold(csum);
320 skb->ip_summed = CHECKSUM_NONE;
321 ip_push_pending_frames(sk, fl4);
322 }
323}
324
325/*
326 * Driving logic for building and sending ICMP messages.
327 */
328
329static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
330{
331 struct ipcm_cookie ipc;
332 struct rtable *rt = skb_rtable(skb);
333 struct net *net = dev_net(rt->dst.dev);
334 struct flowi4 fl4;
335 struct sock *sk;
336 struct inet_sock *inet;
337 __be32 daddr;
338
339 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
340 return;
341
342 sk = icmp_xmit_lock(net);
343 if (sk == NULL)
344 return;
345 inet = inet_sk(sk);
346
347 icmp_param->data.icmph.checksum = 0;
348
349 inet->tos = ip_hdr(skb)->tos;
350 daddr = ipc.addr = ip_hdr(skb)->saddr;
351 ipc.opt = NULL;
352 ipc.tx_flags = 0;
353 if (icmp_param->replyopts.opt.opt.optlen) {
354 ipc.opt = &icmp_param->replyopts.opt;
355 if (ipc.opt->opt.srr)
356 daddr = icmp_param->replyopts.opt.opt.faddr;
357 }
358 memset(&fl4, 0, sizeof(fl4));
359 fl4.daddr = daddr;
360 fl4.saddr = rt->rt_spec_dst;
361 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
362 fl4.flowi4_proto = IPPROTO_ICMP;
363 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
364 rt = ip_route_output_key(net, &fl4);
365 if (IS_ERR(rt))
366 goto out_unlock;
367 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
368 icmp_param->data.icmph.code))
369 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
370 ip_rt_put(rt);
371out_unlock:
372 icmp_xmit_unlock(sk);
373}
374
375static struct rtable *icmp_route_lookup(struct net *net,
376 struct flowi4 *fl4,
377 struct sk_buff *skb_in,
378 const struct iphdr *iph,
379 __be32 saddr, u8 tos,
380 int type, int code,
381 struct icmp_bxm *param)
382{
383 struct rtable *rt, *rt2;
384 struct flowi4 fl4_dec;
385 int err;
386
387 memset(fl4, 0, sizeof(*fl4));
388 fl4->daddr = (param->replyopts.opt.opt.srr ?
389 param->replyopts.opt.opt.faddr : iph->saddr);
390 fl4->saddr = saddr;
391 fl4->flowi4_tos = RT_TOS(tos);
392 fl4->flowi4_proto = IPPROTO_ICMP;
393 fl4->fl4_icmp_type = type;
394 fl4->fl4_icmp_code = code;
395 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
396 rt = __ip_route_output_key(net, fl4);
397 if (IS_ERR(rt))
398 return rt;
399
400 /* No need to clone since we're just using its address. */
401 rt2 = rt;
402
403 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
404 flowi4_to_flowi(fl4), NULL, 0);
405 if (!IS_ERR(rt)) {
406 if (rt != rt2)
407 return rt;
408 } else if (PTR_ERR(rt) == -EPERM) {
409 rt = NULL;
410 } else
411 return rt;
412
413 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
414 if (err)
415 goto relookup_failed;
416
417 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
418 rt2 = __ip_route_output_key(net, &fl4_dec);
419 if (IS_ERR(rt2))
420 err = PTR_ERR(rt2);
421 } else {
422 struct flowi4 fl4_2 = {};
423 unsigned long orefdst;
424
425 fl4_2.daddr = fl4_dec.saddr;
426 rt2 = ip_route_output_key(net, &fl4_2);
427 if (IS_ERR(rt2)) {
428 err = PTR_ERR(rt2);
429 goto relookup_failed;
430 }
431 /* Ugh! */
432 orefdst = skb_in->_skb_refdst; /* save old refdst */
433 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
434 RT_TOS(tos), rt2->dst.dev);
435
436 dst_release(&rt2->dst);
437 rt2 = skb_rtable(skb_in);
438 skb_in->_skb_refdst = orefdst; /* restore old refdst */
439 }
440
441 if (err)
442 goto relookup_failed;
443
444 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
445 flowi4_to_flowi(&fl4_dec), NULL,
446 XFRM_LOOKUP_ICMP);
447 if (!IS_ERR(rt2)) {
448 dst_release(&rt->dst);
449 memcpy(fl4, &fl4_dec, sizeof(*fl4));
450 rt = rt2;
451 } else if (PTR_ERR(rt2) == -EPERM) {
452 if (rt)
453 dst_release(&rt->dst);
454 return rt2;
455 } else {
456 err = PTR_ERR(rt2);
457 goto relookup_failed;
458 }
459 return rt;
460
461relookup_failed:
462 if (rt)
463 return rt;
464 return ERR_PTR(err);
465}
466
467/*
468 * Send an ICMP message in response to a situation
469 *
470 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
471 * MAY send more (we do).
472 * MUST NOT change this header information.
473 * MUST NOT reply to a multicast/broadcast IP address.
474 * MUST NOT reply to a multicast/broadcast MAC address.
475 * MUST reply to only the first fragment.
476 */
477
478void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
479{
480 struct iphdr *iph;
481 int room;
482 struct icmp_bxm icmp_param;
483 struct rtable *rt = skb_rtable(skb_in);
484 struct ipcm_cookie ipc;
485 struct flowi4 fl4;
486 __be32 saddr;
487 u8 tos;
488 struct net *net;
489 struct sock *sk;
490
491 if (!rt)
492 goto out;
493 net = dev_net(rt->dst.dev);
494
495 /*
496 * Find the original header. It is expected to be valid, of course.
497 * Check this, icmp_send is called from the most obscure devices
498 * sometimes.
499 */
500 iph = ip_hdr(skb_in);
501
502 if ((u8 *)iph < skb_in->head ||
503 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
504 goto out;
505
506 /*
507 * No replies to physical multicast/broadcast
508 */
509 if (skb_in->pkt_type != PACKET_HOST)
510 goto out;
511
512 /*
513 * Now check at the protocol level
514 */
515 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
516 goto out;
517
518 /*
519 * Only reply to fragment 0. We byte re-order the constant
520 * mask for efficiency.
521 */
522 if (iph->frag_off & htons(IP_OFFSET))
523 goto out;
524
525 /*
526 * If we send an ICMP error to an ICMP error a mess would result..
527 */
528 if (icmp_pointers[type].error) {
529 /*
530 * We are an error, check if we are replying to an
531 * ICMP error
532 */
533 if (iph->protocol == IPPROTO_ICMP) {
534 u8 _inner_type, *itp;
535
536 itp = skb_header_pointer(skb_in,
537 skb_network_header(skb_in) +
538 (iph->ihl << 2) +
539 offsetof(struct icmphdr,
540 type) -
541 skb_in->data,
542 sizeof(_inner_type),
543 &_inner_type);
544 if (itp == NULL)
545 goto out;
546
547 /*
548 * Assume any unknown ICMP type is an error. This
549 * isn't specified by the RFC, but think about it..
550 */
551 if (*itp > NR_ICMP_TYPES ||
552 icmp_pointers[*itp].error)
553 goto out;
554 }
555 }
556
557 sk = icmp_xmit_lock(net);
558 if (sk == NULL)
559 return;
560
561 /*
562 * Construct source address and options.
563 */
564
565 saddr = iph->daddr;
566 if (!(rt->rt_flags & RTCF_LOCAL)) {
567 struct net_device *dev = NULL;
568
569 rcu_read_lock();
570 if (rt_is_input_route(rt) &&
571 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
572 dev = dev_get_by_index_rcu(net, rt->rt_iif);
573
574 if (dev)
575 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
576 else
577 saddr = 0;
578 rcu_read_unlock();
579 }
580
581 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
582 IPTOS_PREC_INTERNETCONTROL) :
583 iph->tos;
584
585 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
586 goto out_unlock;
587
588
589 /*
590 * Prepare data for ICMP header.
591 */
592
593 icmp_param.data.icmph.type = type;
594 icmp_param.data.icmph.code = code;
595 icmp_param.data.icmph.un.gateway = info;
596 icmp_param.data.icmph.checksum = 0;
597 icmp_param.skb = skb_in;
598 icmp_param.offset = skb_network_offset(skb_in);
599 inet_sk(sk)->tos = tos;
600 ipc.addr = iph->saddr;
601 ipc.opt = &icmp_param.replyopts.opt;
602 ipc.tx_flags = 0;
603
604 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos,
605 type, code, &icmp_param);
606 if (IS_ERR(rt))
607 goto out_unlock;
608
609 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
610 goto ende;
611
612 /* RFC says return as much as we can without exceeding 576 bytes. */
613
614 room = dst_mtu(&rt->dst);
615 if (room > 576)
616 room = 576;
617 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
618 room -= sizeof(struct icmphdr);
619
620 icmp_param.data_len = skb_in->len - icmp_param.offset;
621 if (icmp_param.data_len > room)
622 icmp_param.data_len = room;
623 icmp_param.head_len = sizeof(struct icmphdr);
624
625 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
626ende:
627 ip_rt_put(rt);
628out_unlock:
629 icmp_xmit_unlock(sk);
630out:;
631}
632EXPORT_SYMBOL(icmp_send);
633
634
635/*
636 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
637 */
638
639static void icmp_unreach(struct sk_buff *skb)
640{
641 const struct iphdr *iph;
642 struct icmphdr *icmph;
643 int hash, protocol;
644 const struct net_protocol *ipprot;
645 u32 info = 0;
646 struct net *net;
647
648 net = dev_net(skb_dst(skb)->dev);
649
650 /*
651 * Incomplete header ?
652 * Only checks for the IP header, there should be an
653 * additional check for longer headers in upper levels.
654 */
655
656 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
657 goto out_err;
658
659 icmph = icmp_hdr(skb);
660 iph = (const struct iphdr *)skb->data;
661
662 if (iph->ihl < 5) /* Mangled header, drop. */
663 goto out_err;
664
665 if (icmph->type == ICMP_DEST_UNREACH) {
666 switch (icmph->code & 15) {
667 case ICMP_NET_UNREACH:
668 case ICMP_HOST_UNREACH:
669 case ICMP_PROT_UNREACH:
670 case ICMP_PORT_UNREACH:
671 break;
672 case ICMP_FRAG_NEEDED:
673 if (ipv4_config.no_pmtu_disc) {
674 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
675 &iph->daddr);
676 } else {
677 info = ip_rt_frag_needed(net, iph,
678 ntohs(icmph->un.frag.mtu),
679 skb->dev);
680 if (!info)
681 goto out;
682 }
683 break;
684 case ICMP_SR_FAILED:
685 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
686 &iph->daddr);
687 break;
688 default:
689 break;
690 }
691 if (icmph->code > NR_ICMP_UNREACH)
692 goto out;
693 } else if (icmph->type == ICMP_PARAMETERPROB)
694 info = ntohl(icmph->un.gateway) >> 24;
695
696 /*
697 * Throw it at our lower layers
698 *
699 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
700 * header.
701 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
702 * transport layer.
703 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
704 * transport layer.
705 */
706
707 /*
708 * Check the other end isn't violating RFC 1122. Some routers send
709 * bogus responses to broadcast frames. If you see this message
710 * first check your netmask matches at both ends, if it does then
711 * get the other vendor to fix their kit.
712 */
713
714 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
715 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
716 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
717 &ip_hdr(skb)->saddr,
718 icmph->type, icmph->code,
719 &iph->daddr, skb->dev->name);
720 goto out;
721 }
722
723 /* Checkin full IP header plus 8 bytes of protocol to
724 * avoid additional coding at protocol handlers.
725 */
726 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
727 goto out;
728
729 iph = (const struct iphdr *)skb->data;
730 protocol = iph->protocol;
731
732 /*
733 * Deliver ICMP message to raw sockets. Pretty useless feature?
734 */
735 raw_icmp_error(skb, protocol, info);
736
737 hash = protocol & (MAX_INET_PROTOS - 1);
738 rcu_read_lock();
739 ipprot = rcu_dereference(inet_protos[hash]);
740 if (ipprot && ipprot->err_handler)
741 ipprot->err_handler(skb, info);
742 rcu_read_unlock();
743
744out:
745 return;
746out_err:
747 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
748 goto out;
749}
750
751
752/*
753 * Handle ICMP_REDIRECT.
754 */
755
756static void icmp_redirect(struct sk_buff *skb)
757{
758 const struct iphdr *iph;
759
760 if (skb->len < sizeof(struct iphdr))
761 goto out_err;
762
763 /*
764 * Get the copied header of the packet that caused the redirect
765 */
766 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
767 goto out;
768
769 iph = (const struct iphdr *)skb->data;
770
771 switch (icmp_hdr(skb)->code & 7) {
772 case ICMP_REDIR_NET:
773 case ICMP_REDIR_NETTOS:
774 /*
775 * As per RFC recommendations now handle it as a host redirect.
776 */
777 case ICMP_REDIR_HOST:
778 case ICMP_REDIR_HOSTTOS:
779 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
780 icmp_hdr(skb)->un.gateway,
781 iph->saddr, skb->dev);
782 break;
783 }
784
785 /* Ping wants to see redirects.
786 * Let's pretend they are errors of sorts... */
787 if (iph->protocol == IPPROTO_ICMP &&
788 iph->ihl >= 5 &&
789 pskb_may_pull(skb, (iph->ihl<<2)+8)) {
790 ping_err(skb, icmp_hdr(skb)->un.gateway);
791 }
792
793out:
794 return;
795out_err:
796 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
797 goto out;
798}
799
800/*
801 * Handle ICMP_ECHO ("ping") requests.
802 *
803 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
804 * requests.
805 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
806 * included in the reply.
807 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
808 * echo requests, MUST have default=NOT.
809 * See also WRT handling of options once they are done and working.
810 */
811
812static void icmp_echo(struct sk_buff *skb)
813{
814 struct net *net;
815
816 net = dev_net(skb_dst(skb)->dev);
817 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
818 struct icmp_bxm icmp_param;
819
820 icmp_param.data.icmph = *icmp_hdr(skb);
821 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
822 icmp_param.skb = skb;
823 icmp_param.offset = 0;
824 icmp_param.data_len = skb->len;
825 icmp_param.head_len = sizeof(struct icmphdr);
826 icmp_reply(&icmp_param, skb);
827 }
828}
829
830/*
831 * Handle ICMP Timestamp requests.
832 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
833 * SHOULD be in the kernel for minimum random latency.
834 * MUST be accurate to a few minutes.
835 * MUST be updated at least at 15Hz.
836 */
837static void icmp_timestamp(struct sk_buff *skb)
838{
839 struct timespec tv;
840 struct icmp_bxm icmp_param;
841 /*
842 * Too short.
843 */
844 if (skb->len < 4)
845 goto out_err;
846
847 /*
848 * Fill in the current time as ms since midnight UT:
849 */
850 getnstimeofday(&tv);
851 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
852 tv.tv_nsec / NSEC_PER_MSEC);
853 icmp_param.data.times[2] = icmp_param.data.times[1];
854 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
855 BUG();
856 icmp_param.data.icmph = *icmp_hdr(skb);
857 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
858 icmp_param.data.icmph.code = 0;
859 icmp_param.skb = skb;
860 icmp_param.offset = 0;
861 icmp_param.data_len = 0;
862 icmp_param.head_len = sizeof(struct icmphdr) + 12;
863 icmp_reply(&icmp_param, skb);
864out:
865 return;
866out_err:
867 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
868 goto out;
869}
870
871
872/*
873 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
874 *
875 * RFC1122 (3.2.2.9). A host MUST only send replies to
876 * ADDRESS_MASK requests if it's been configured as an address mask
877 * agent. Receiving a request doesn't constitute implicit permission to
878 * act as one. Of course, implementing this correctly requires (SHOULD)
879 * a way to turn the functionality on and off. Another one for sysctl(),
880 * I guess. -- MS
881 *
882 * RFC1812 (4.3.3.9). A router MUST implement it.
883 * A router SHOULD have switch turning it on/off.
884 * This switch MUST be ON by default.
885 *
886 * Gratuitous replies, zero-source replies are not implemented,
887 * that complies with RFC. DO NOT implement them!!! All the idea
888 * of broadcast addrmask replies as specified in RFC950 is broken.
889 * The problem is that it is not uncommon to have several prefixes
890 * on one physical interface. Moreover, addrmask agent can even be
891 * not aware of existing another prefixes.
892 * If source is zero, addrmask agent cannot choose correct prefix.
893 * Gratuitous mask announcements suffer from the same problem.
894 * RFC1812 explains it, but still allows to use ADDRMASK,
895 * that is pretty silly. --ANK
896 *
897 * All these rules are so bizarre, that I removed kernel addrmask
898 * support at all. It is wrong, it is obsolete, nobody uses it in
899 * any case. --ANK
900 *
901 * Furthermore you can do it with a usermode address agent program
902 * anyway...
903 */
904
905static void icmp_address(struct sk_buff *skb)
906{
907#if 0
908 net_dbg_ratelimited("a guy asks for address mask. Who is it?\n");
909#endif
910}
911
912/*
913 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
914 * loudly if an inconsistency is found.
915 * called with rcu_read_lock()
916 */
917
918static void icmp_address_reply(struct sk_buff *skb)
919{
920 struct rtable *rt = skb_rtable(skb);
921 struct net_device *dev = skb->dev;
922 struct in_device *in_dev;
923 struct in_ifaddr *ifa;
924
925 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
926 return;
927
928 in_dev = __in_dev_get_rcu(dev);
929 if (!in_dev)
930 return;
931
932 if (in_dev->ifa_list &&
933 IN_DEV_LOG_MARTIANS(in_dev) &&
934 IN_DEV_FORWARD(in_dev)) {
935 __be32 _mask, *mp;
936
937 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
938 BUG_ON(mp == NULL);
939 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
940 if (*mp == ifa->ifa_mask &&
941 inet_ifa_match(ip_hdr(skb)->saddr, ifa))
942 break;
943 }
944 if (!ifa)
945 net_info_ratelimited("Wrong address mask %pI4 from %s/%pI4\n",
946 mp,
947 dev->name, &ip_hdr(skb)->saddr);
948 }
949}
950
951static void icmp_discard(struct sk_buff *skb)
952{
953}
954
955/*
956 * Deal with incoming ICMP packets.
957 */
958int icmp_rcv(struct sk_buff *skb)
959{
960 struct icmphdr *icmph;
961 struct rtable *rt = skb_rtable(skb);
962 struct net *net = dev_net(rt->dst.dev);
963
964 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
965 struct sec_path *sp = skb_sec_path(skb);
966 int nh;
967
968 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
969 XFRM_STATE_ICMP))
970 goto drop;
971
972 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
973 goto drop;
974
975 nh = skb_network_offset(skb);
976 skb_set_network_header(skb, sizeof(*icmph));
977
978 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
979 goto drop;
980
981 skb_set_network_header(skb, nh);
982 }
983
984 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
985
986 switch (skb->ip_summed) {
987 case CHECKSUM_COMPLETE:
988 if (!csum_fold(skb->csum))
989 break;
990 /* fall through */
991 case CHECKSUM_NONE:
992 skb->csum = 0;
993 if (__skb_checksum_complete(skb))
994 goto error;
995 }
996
997 if (!pskb_pull(skb, sizeof(*icmph)))
998 goto error;
999
1000 icmph = icmp_hdr(skb);
1001
1002 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
1003 /*
1004 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1005 *
1006 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1007 * discarded.
1008 */
1009 if (icmph->type > NR_ICMP_TYPES)
1010 goto error;
1011
1012
1013 /*
1014 * Parse the ICMP message
1015 */
1016
1017 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1018 /*
1019 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1020 * silently ignored (we let user decide with a sysctl).
1021 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1022 * discarded if to broadcast/multicast.
1023 */
1024 if ((icmph->type == ICMP_ECHO ||
1025 icmph->type == ICMP_TIMESTAMP) &&
1026 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1027 goto error;
1028 }
1029 if (icmph->type != ICMP_ECHO &&
1030 icmph->type != ICMP_TIMESTAMP &&
1031 icmph->type != ICMP_ADDRESS &&
1032 icmph->type != ICMP_ADDRESSREPLY) {
1033 goto error;
1034 }
1035 }
1036
1037 icmp_pointers[icmph->type].handler(skb);
1038
1039drop:
1040 kfree_skb(skb);
1041 return 0;
1042error:
1043 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
1044 goto drop;
1045}
1046
1047/*
1048 * This table is the definition of how we handle ICMP.
1049 */
1050static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1051 [ICMP_ECHOREPLY] = {
1052 .handler = ping_rcv,
1053 },
1054 [1] = {
1055 .handler = icmp_discard,
1056 .error = 1,
1057 },
1058 [2] = {
1059 .handler = icmp_discard,
1060 .error = 1,
1061 },
1062 [ICMP_DEST_UNREACH] = {
1063 .handler = icmp_unreach,
1064 .error = 1,
1065 },
1066 [ICMP_SOURCE_QUENCH] = {
1067 .handler = icmp_unreach,
1068 .error = 1,
1069 },
1070 [ICMP_REDIRECT] = {
1071 .handler = icmp_redirect,
1072 .error = 1,
1073 },
1074 [6] = {
1075 .handler = icmp_discard,
1076 .error = 1,
1077 },
1078 [7] = {
1079 .handler = icmp_discard,
1080 .error = 1,
1081 },
1082 [ICMP_ECHO] = {
1083 .handler = icmp_echo,
1084 },
1085 [9] = {
1086 .handler = icmp_discard,
1087 .error = 1,
1088 },
1089 [10] = {
1090 .handler = icmp_discard,
1091 .error = 1,
1092 },
1093 [ICMP_TIME_EXCEEDED] = {
1094 .handler = icmp_unreach,
1095 .error = 1,
1096 },
1097 [ICMP_PARAMETERPROB] = {
1098 .handler = icmp_unreach,
1099 .error = 1,
1100 },
1101 [ICMP_TIMESTAMP] = {
1102 .handler = icmp_timestamp,
1103 },
1104 [ICMP_TIMESTAMPREPLY] = {
1105 .handler = icmp_discard,
1106 },
1107 [ICMP_INFO_REQUEST] = {
1108 .handler = icmp_discard,
1109 },
1110 [ICMP_INFO_REPLY] = {
1111 .handler = icmp_discard,
1112 },
1113 [ICMP_ADDRESS] = {
1114 .handler = icmp_address,
1115 },
1116 [ICMP_ADDRESSREPLY] = {
1117 .handler = icmp_address_reply,
1118 },
1119};
1120
1121static void __net_exit icmp_sk_exit(struct net *net)
1122{
1123 int i;
1124
1125 for_each_possible_cpu(i)
1126 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1127 kfree(net->ipv4.icmp_sk);
1128 net->ipv4.icmp_sk = NULL;
1129}
1130
1131static int __net_init icmp_sk_init(struct net *net)
1132{
1133 int i, err;
1134
1135 net->ipv4.icmp_sk =
1136 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1137 if (net->ipv4.icmp_sk == NULL)
1138 return -ENOMEM;
1139
1140 for_each_possible_cpu(i) {
1141 struct sock *sk;
1142
1143 err = inet_ctl_sock_create(&sk, PF_INET,
1144 SOCK_RAW, IPPROTO_ICMP, net);
1145 if (err < 0)
1146 goto fail;
1147
1148 net->ipv4.icmp_sk[i] = sk;
1149
1150 /* Enough space for 2 64K ICMP packets, including
1151 * sk_buff/skb_shared_info struct overhead.
1152 */
1153 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1154
1155 /*
1156 * Speedup sock_wfree()
1157 */
1158 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1159 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1160 }
1161
1162 /* Control parameters for ECHO replies. */
1163 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1164 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1165
1166 /* Control parameter - ignore bogus broadcast responses? */
1167 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1168
1169 /*
1170 * Configurable global rate limit.
1171 *
1172 * ratelimit defines tokens/packet consumed for dst->rate_token
1173 * bucket ratemask defines which icmp types are ratelimited by
1174 * setting it's bit position.
1175 *
1176 * default:
1177 * dest unreachable (3), source quench (4),
1178 * time exceeded (11), parameter problem (12)
1179 */
1180
1181 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1182 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1183 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1184
1185 return 0;
1186
1187fail:
1188 for_each_possible_cpu(i)
1189 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1190 kfree(net->ipv4.icmp_sk);
1191 return err;
1192}
1193
1194static struct pernet_operations __net_initdata icmp_sk_ops = {
1195 .init = icmp_sk_init,
1196 .exit = icmp_sk_exit,
1197};
1198
1199int __init icmp_init(void)
1200{
1201 return register_pernet_subsys(&icmp_sk_ops);
1202}