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 bool (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191};
192
193static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194
195/*
196 * The ICMP socket(s). This is the most convenient way to flow control
197 * our ICMP output as well as maintain a clean interface throughout
198 * all layers. All Socketless IP sends will soon be gone.
199 *
200 * On SMP we have one ICMP socket per-cpu.
201 */
202static struct sock *icmp_sk(struct net *net)
203{
204 return this_cpu_read(*net->ipv4.icmp_sk);
205}
206
207/* Called with BH disabled */
208static inline struct sock *icmp_xmit_lock(struct net *net)
209{
210 struct sock *sk;
211
212 sk = icmp_sk(net);
213
214 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 /* This can happen if the output path signals a
216 * dst_link_failure() for an outgoing ICMP packet.
217 */
218 return NULL;
219 }
220 return sk;
221}
222
223static inline void icmp_xmit_unlock(struct sock *sk)
224{
225 spin_unlock(&sk->sk_lock.slock);
226}
227
228int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229int sysctl_icmp_msgs_burst __read_mostly = 50;
230
231static struct {
232 spinlock_t lock;
233 u32 credit;
234 u32 stamp;
235} icmp_global = {
236 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
237};
238
239/**
240 * icmp_global_allow - Are we allowed to send one more ICMP message ?
241 *
242 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
243 * Returns false if we reached the limit and can not send another packet.
244 * Note: called with BH disabled
245 */
246bool icmp_global_allow(void)
247{
248 u32 credit, delta, incr = 0, now = (u32)jiffies;
249 bool rc = false;
250
251 /* Check if token bucket is empty and cannot be refilled
252 * without taking the spinlock. The READ_ONCE() are paired
253 * with the following WRITE_ONCE() in this same function.
254 */
255 if (!READ_ONCE(icmp_global.credit)) {
256 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
257 if (delta < HZ / 50)
258 return false;
259 }
260
261 spin_lock(&icmp_global.lock);
262 delta = min_t(u32, now - icmp_global.stamp, HZ);
263 if (delta >= HZ / 50) {
264 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
265 if (incr)
266 WRITE_ONCE(icmp_global.stamp, now);
267 }
268 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
269 if (credit) {
270 /* We want to use a credit of one in average, but need to randomize
271 * it for security reasons.
272 */
273 credit = max_t(int, credit - prandom_u32_max(3), 0);
274 rc = true;
275 }
276 WRITE_ONCE(icmp_global.credit, credit);
277 spin_unlock(&icmp_global.lock);
278 return rc;
279}
280EXPORT_SYMBOL(icmp_global_allow);
281
282static bool icmpv4_mask_allow(struct net *net, int type, int code)
283{
284 if (type > NR_ICMP_TYPES)
285 return true;
286
287 /* Don't limit PMTU discovery. */
288 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
289 return true;
290
291 /* Limit if icmp type is enabled in ratemask. */
292 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
293 return true;
294
295 return false;
296}
297
298static bool icmpv4_global_allow(struct net *net, int type, int code)
299{
300 if (icmpv4_mask_allow(net, type, code))
301 return true;
302
303 if (icmp_global_allow())
304 return true;
305
306 return false;
307}
308
309/*
310 * Send an ICMP frame.
311 */
312
313static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
314 struct flowi4 *fl4, int type, int code)
315{
316 struct dst_entry *dst = &rt->dst;
317 struct inet_peer *peer;
318 bool rc = true;
319 int vif;
320
321 if (icmpv4_mask_allow(net, type, code))
322 goto out;
323
324 /* No rate limit on loopback */
325 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
326 goto out;
327
328 vif = l3mdev_master_ifindex(dst->dev);
329 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
330 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
331 if (peer)
332 inet_putpeer(peer);
333out:
334 return rc;
335}
336
337/*
338 * Maintain the counters used in the SNMP statistics for outgoing ICMP
339 */
340void icmp_out_count(struct net *net, unsigned char type)
341{
342 ICMPMSGOUT_INC_STATS(net, type);
343 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
344}
345
346/*
347 * Checksum each fragment, and on the first include the headers and final
348 * checksum.
349 */
350static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
351 struct sk_buff *skb)
352{
353 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
354 __wsum csum;
355
356 csum = skb_copy_and_csum_bits(icmp_param->skb,
357 icmp_param->offset + offset,
358 to, len);
359
360 skb->csum = csum_block_add(skb->csum, csum, odd);
361 if (icmp_pointers[icmp_param->data.icmph.type].error)
362 nf_ct_attach(skb, icmp_param->skb);
363 return 0;
364}
365
366static void icmp_push_reply(struct icmp_bxm *icmp_param,
367 struct flowi4 *fl4,
368 struct ipcm_cookie *ipc, struct rtable **rt)
369{
370 struct sock *sk;
371 struct sk_buff *skb;
372
373 sk = icmp_sk(dev_net((*rt)->dst.dev));
374 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
375 icmp_param->data_len+icmp_param->head_len,
376 icmp_param->head_len,
377 ipc, rt, MSG_DONTWAIT) < 0) {
378 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
379 ip_flush_pending_frames(sk);
380 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
381 struct icmphdr *icmph = icmp_hdr(skb);
382 __wsum csum;
383 struct sk_buff *skb1;
384
385 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
386 (char *)icmph,
387 icmp_param->head_len);
388 skb_queue_walk(&sk->sk_write_queue, skb1) {
389 csum = csum_add(csum, skb1->csum);
390 }
391 icmph->checksum = csum_fold(csum);
392 skb->ip_summed = CHECKSUM_NONE;
393 ip_push_pending_frames(sk, fl4);
394 }
395}
396
397/*
398 * Driving logic for building and sending ICMP messages.
399 */
400
401static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
402{
403 struct ipcm_cookie ipc;
404 struct rtable *rt = skb_rtable(skb);
405 struct net *net = dev_net(rt->dst.dev);
406 struct flowi4 fl4;
407 struct sock *sk;
408 struct inet_sock *inet;
409 __be32 daddr, saddr;
410 u32 mark = IP4_REPLY_MARK(net, skb->mark);
411 int type = icmp_param->data.icmph.type;
412 int code = icmp_param->data.icmph.code;
413
414 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
415 return;
416
417 /* Needed by both icmp_global_allow and icmp_xmit_lock */
418 local_bh_disable();
419
420 /* global icmp_msgs_per_sec */
421 if (!icmpv4_global_allow(net, type, code))
422 goto out_bh_enable;
423
424 sk = icmp_xmit_lock(net);
425 if (!sk)
426 goto out_bh_enable;
427 inet = inet_sk(sk);
428
429 icmp_param->data.icmph.checksum = 0;
430
431 ipcm_init(&ipc);
432 inet->tos = ip_hdr(skb)->tos;
433 ipc.sockc.mark = mark;
434 daddr = ipc.addr = ip_hdr(skb)->saddr;
435 saddr = fib_compute_spec_dst(skb);
436
437 if (icmp_param->replyopts.opt.opt.optlen) {
438 ipc.opt = &icmp_param->replyopts.opt;
439 if (ipc.opt->opt.srr)
440 daddr = icmp_param->replyopts.opt.opt.faddr;
441 }
442 memset(&fl4, 0, sizeof(fl4));
443 fl4.daddr = daddr;
444 fl4.saddr = saddr;
445 fl4.flowi4_mark = mark;
446 fl4.flowi4_uid = sock_net_uid(net, NULL);
447 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
448 fl4.flowi4_proto = IPPROTO_ICMP;
449 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
450 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
451 rt = ip_route_output_key(net, &fl4);
452 if (IS_ERR(rt))
453 goto out_unlock;
454 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
455 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
456 ip_rt_put(rt);
457out_unlock:
458 icmp_xmit_unlock(sk);
459out_bh_enable:
460 local_bh_enable();
461}
462
463/*
464 * The device used for looking up which routing table to use for sending an ICMP
465 * error is preferably the source whenever it is set, which should ensure the
466 * icmp error can be sent to the source host, else lookup using the routing
467 * table of the destination device, else use the main routing table (index 0).
468 */
469static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
470{
471 struct net_device *route_lookup_dev = NULL;
472
473 if (skb->dev)
474 route_lookup_dev = skb->dev;
475 else if (skb_dst(skb))
476 route_lookup_dev = skb_dst(skb)->dev;
477 return route_lookup_dev;
478}
479
480static struct rtable *icmp_route_lookup(struct net *net,
481 struct flowi4 *fl4,
482 struct sk_buff *skb_in,
483 const struct iphdr *iph,
484 __be32 saddr, u8 tos, u32 mark,
485 int type, int code,
486 struct icmp_bxm *param)
487{
488 struct net_device *route_lookup_dev;
489 struct rtable *rt, *rt2;
490 struct flowi4 fl4_dec;
491 int err;
492
493 memset(fl4, 0, sizeof(*fl4));
494 fl4->daddr = (param->replyopts.opt.opt.srr ?
495 param->replyopts.opt.opt.faddr : iph->saddr);
496 fl4->saddr = saddr;
497 fl4->flowi4_mark = mark;
498 fl4->flowi4_uid = sock_net_uid(net, NULL);
499 fl4->flowi4_tos = RT_TOS(tos);
500 fl4->flowi4_proto = IPPROTO_ICMP;
501 fl4->fl4_icmp_type = type;
502 fl4->fl4_icmp_code = code;
503 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
504 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
505
506 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
507 rt = ip_route_output_key_hash(net, fl4, skb_in);
508 if (IS_ERR(rt))
509 return rt;
510
511 /* No need to clone since we're just using its address. */
512 rt2 = rt;
513
514 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
515 flowi4_to_flowi(fl4), NULL, 0);
516 if (!IS_ERR(rt)) {
517 if (rt != rt2)
518 return rt;
519 } else if (PTR_ERR(rt) == -EPERM) {
520 rt = NULL;
521 } else
522 return rt;
523
524 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
525 if (err)
526 goto relookup_failed;
527
528 if (inet_addr_type_dev_table(net, route_lookup_dev,
529 fl4_dec.saddr) == RTN_LOCAL) {
530 rt2 = __ip_route_output_key(net, &fl4_dec);
531 if (IS_ERR(rt2))
532 err = PTR_ERR(rt2);
533 } else {
534 struct flowi4 fl4_2 = {};
535 unsigned long orefdst;
536
537 fl4_2.daddr = fl4_dec.saddr;
538 rt2 = ip_route_output_key(net, &fl4_2);
539 if (IS_ERR(rt2)) {
540 err = PTR_ERR(rt2);
541 goto relookup_failed;
542 }
543 /* Ugh! */
544 orefdst = skb_in->_skb_refdst; /* save old refdst */
545 skb_dst_set(skb_in, NULL);
546 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
547 RT_TOS(tos), rt2->dst.dev);
548
549 dst_release(&rt2->dst);
550 rt2 = skb_rtable(skb_in);
551 skb_in->_skb_refdst = orefdst; /* restore old refdst */
552 }
553
554 if (err)
555 goto relookup_failed;
556
557 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
558 flowi4_to_flowi(&fl4_dec), NULL,
559 XFRM_LOOKUP_ICMP);
560 if (!IS_ERR(rt2)) {
561 dst_release(&rt->dst);
562 memcpy(fl4, &fl4_dec, sizeof(*fl4));
563 rt = rt2;
564 } else if (PTR_ERR(rt2) == -EPERM) {
565 if (rt)
566 dst_release(&rt->dst);
567 return rt2;
568 } else {
569 err = PTR_ERR(rt2);
570 goto relookup_failed;
571 }
572 return rt;
573
574relookup_failed:
575 if (rt)
576 return rt;
577 return ERR_PTR(err);
578}
579
580/*
581 * Send an ICMP message in response to a situation
582 *
583 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
584 * MAY send more (we do).
585 * MUST NOT change this header information.
586 * MUST NOT reply to a multicast/broadcast IP address.
587 * MUST NOT reply to a multicast/broadcast MAC address.
588 * MUST reply to only the first fragment.
589 */
590
591void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
592 const struct ip_options *opt)
593{
594 struct iphdr *iph;
595 int room;
596 struct icmp_bxm icmp_param;
597 struct rtable *rt = skb_rtable(skb_in);
598 struct ipcm_cookie ipc;
599 struct flowi4 fl4;
600 __be32 saddr;
601 u8 tos;
602 u32 mark;
603 struct net *net;
604 struct sock *sk;
605
606 if (!rt)
607 goto out;
608
609 if (rt->dst.dev)
610 net = dev_net(rt->dst.dev);
611 else if (skb_in->dev)
612 net = dev_net(skb_in->dev);
613 else
614 goto out;
615
616 /*
617 * Find the original header. It is expected to be valid, of course.
618 * Check this, icmp_send is called from the most obscure devices
619 * sometimes.
620 */
621 iph = ip_hdr(skb_in);
622
623 if ((u8 *)iph < skb_in->head ||
624 (skb_network_header(skb_in) + sizeof(*iph)) >
625 skb_tail_pointer(skb_in))
626 goto out;
627
628 /*
629 * No replies to physical multicast/broadcast
630 */
631 if (skb_in->pkt_type != PACKET_HOST)
632 goto out;
633
634 /*
635 * Now check at the protocol level
636 */
637 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
638 goto out;
639
640 /*
641 * Only reply to fragment 0. We byte re-order the constant
642 * mask for efficiency.
643 */
644 if (iph->frag_off & htons(IP_OFFSET))
645 goto out;
646
647 /*
648 * If we send an ICMP error to an ICMP error a mess would result..
649 */
650 if (icmp_pointers[type].error) {
651 /*
652 * We are an error, check if we are replying to an
653 * ICMP error
654 */
655 if (iph->protocol == IPPROTO_ICMP) {
656 u8 _inner_type, *itp;
657
658 itp = skb_header_pointer(skb_in,
659 skb_network_header(skb_in) +
660 (iph->ihl << 2) +
661 offsetof(struct icmphdr,
662 type) -
663 skb_in->data,
664 sizeof(_inner_type),
665 &_inner_type);
666 if (!itp)
667 goto out;
668
669 /*
670 * Assume any unknown ICMP type is an error. This
671 * isn't specified by the RFC, but think about it..
672 */
673 if (*itp > NR_ICMP_TYPES ||
674 icmp_pointers[*itp].error)
675 goto out;
676 }
677 }
678
679 /* Needed by both icmp_global_allow and icmp_xmit_lock */
680 local_bh_disable();
681
682 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
683 * incoming dev is loopback. If outgoing dev change to not be
684 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
685 */
686 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
687 !icmpv4_global_allow(net, type, code))
688 goto out_bh_enable;
689
690 sk = icmp_xmit_lock(net);
691 if (!sk)
692 goto out_bh_enable;
693
694 /*
695 * Construct source address and options.
696 */
697
698 saddr = iph->daddr;
699 if (!(rt->rt_flags & RTCF_LOCAL)) {
700 struct net_device *dev = NULL;
701
702 rcu_read_lock();
703 if (rt_is_input_route(rt) &&
704 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
705 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
706
707 if (dev)
708 saddr = inet_select_addr(dev, iph->saddr,
709 RT_SCOPE_LINK);
710 else
711 saddr = 0;
712 rcu_read_unlock();
713 }
714
715 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
716 IPTOS_PREC_INTERNETCONTROL) :
717 iph->tos;
718 mark = IP4_REPLY_MARK(net, skb_in->mark);
719
720 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
721 goto out_unlock;
722
723
724 /*
725 * Prepare data for ICMP header.
726 */
727
728 icmp_param.data.icmph.type = type;
729 icmp_param.data.icmph.code = code;
730 icmp_param.data.icmph.un.gateway = info;
731 icmp_param.data.icmph.checksum = 0;
732 icmp_param.skb = skb_in;
733 icmp_param.offset = skb_network_offset(skb_in);
734 inet_sk(sk)->tos = tos;
735 ipcm_init(&ipc);
736 ipc.addr = iph->saddr;
737 ipc.opt = &icmp_param.replyopts.opt;
738 ipc.sockc.mark = mark;
739
740 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
741 type, code, &icmp_param);
742 if (IS_ERR(rt))
743 goto out_unlock;
744
745 /* peer icmp_ratelimit */
746 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
747 goto ende;
748
749 /* RFC says return as much as we can without exceeding 576 bytes. */
750
751 room = dst_mtu(&rt->dst);
752 if (room > 576)
753 room = 576;
754 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
755 room -= sizeof(struct icmphdr);
756
757 icmp_param.data_len = skb_in->len - icmp_param.offset;
758 if (icmp_param.data_len > room)
759 icmp_param.data_len = room;
760 icmp_param.head_len = sizeof(struct icmphdr);
761
762 /* if we don't have a source address at this point, fall back to the
763 * dummy address instead of sending out a packet with a source address
764 * of 0.0.0.0
765 */
766 if (!fl4.saddr)
767 fl4.saddr = htonl(INADDR_DUMMY);
768
769 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
770ende:
771 ip_rt_put(rt);
772out_unlock:
773 icmp_xmit_unlock(sk);
774out_bh_enable:
775 local_bh_enable();
776out:;
777}
778EXPORT_SYMBOL(__icmp_send);
779
780#if IS_ENABLED(CONFIG_NF_NAT)
781#include <net/netfilter/nf_conntrack.h>
782void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
783{
784 struct sk_buff *cloned_skb = NULL;
785 struct ip_options opts = { 0 };
786 enum ip_conntrack_info ctinfo;
787 struct nf_conn *ct;
788 __be32 orig_ip;
789
790 ct = nf_ct_get(skb_in, &ctinfo);
791 if (!ct || !(ct->status & IPS_SRC_NAT)) {
792 __icmp_send(skb_in, type, code, info, &opts);
793 return;
794 }
795
796 if (skb_shared(skb_in))
797 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
798
799 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
800 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
801 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
802 skb_network_offset(skb_in) + sizeof(struct iphdr))))
803 goto out;
804
805 orig_ip = ip_hdr(skb_in)->saddr;
806 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
807 __icmp_send(skb_in, type, code, info, &opts);
808 ip_hdr(skb_in)->saddr = orig_ip;
809out:
810 consume_skb(cloned_skb);
811}
812EXPORT_SYMBOL(icmp_ndo_send);
813#endif
814
815static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
816{
817 const struct iphdr *iph = (const struct iphdr *)skb->data;
818 const struct net_protocol *ipprot;
819 int protocol = iph->protocol;
820
821 /* Checkin full IP header plus 8 bytes of protocol to
822 * avoid additional coding at protocol handlers.
823 */
824 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
825 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
826 return;
827 }
828
829 raw_icmp_error(skb, protocol, info);
830
831 ipprot = rcu_dereference(inet_protos[protocol]);
832 if (ipprot && ipprot->err_handler)
833 ipprot->err_handler(skb, info);
834}
835
836static bool icmp_tag_validation(int proto)
837{
838 bool ok;
839
840 rcu_read_lock();
841 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
842 rcu_read_unlock();
843 return ok;
844}
845
846/*
847 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
848 * ICMP_PARAMETERPROB.
849 */
850
851static bool icmp_unreach(struct sk_buff *skb)
852{
853 const struct iphdr *iph;
854 struct icmphdr *icmph;
855 struct net *net;
856 u32 info = 0;
857
858 net = dev_net(skb_dst(skb)->dev);
859
860 /*
861 * Incomplete header ?
862 * Only checks for the IP header, there should be an
863 * additional check for longer headers in upper levels.
864 */
865
866 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
867 goto out_err;
868
869 icmph = icmp_hdr(skb);
870 iph = (const struct iphdr *)skb->data;
871
872 if (iph->ihl < 5) /* Mangled header, drop. */
873 goto out_err;
874
875 switch (icmph->type) {
876 case ICMP_DEST_UNREACH:
877 switch (icmph->code & 15) {
878 case ICMP_NET_UNREACH:
879 case ICMP_HOST_UNREACH:
880 case ICMP_PROT_UNREACH:
881 case ICMP_PORT_UNREACH:
882 break;
883 case ICMP_FRAG_NEEDED:
884 /* for documentation of the ip_no_pmtu_disc
885 * values please see
886 * Documentation/networking/ip-sysctl.rst
887 */
888 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
889 default:
890 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
891 &iph->daddr);
892 break;
893 case 2:
894 goto out;
895 case 3:
896 if (!icmp_tag_validation(iph->protocol))
897 goto out;
898 fallthrough;
899 case 0:
900 info = ntohs(icmph->un.frag.mtu);
901 }
902 break;
903 case ICMP_SR_FAILED:
904 net_dbg_ratelimited("%pI4: Source Route Failed\n",
905 &iph->daddr);
906 break;
907 default:
908 break;
909 }
910 if (icmph->code > NR_ICMP_UNREACH)
911 goto out;
912 break;
913 case ICMP_PARAMETERPROB:
914 info = ntohl(icmph->un.gateway) >> 24;
915 break;
916 case ICMP_TIME_EXCEEDED:
917 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
918 if (icmph->code == ICMP_EXC_FRAGTIME)
919 goto out;
920 break;
921 }
922
923 /*
924 * Throw it at our lower layers
925 *
926 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
927 * header.
928 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
929 * transport layer.
930 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
931 * transport layer.
932 */
933
934 /*
935 * Check the other end isn't violating RFC 1122. Some routers send
936 * bogus responses to broadcast frames. If you see this message
937 * first check your netmask matches at both ends, if it does then
938 * get the other vendor to fix their kit.
939 */
940
941 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
942 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
943 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
944 &ip_hdr(skb)->saddr,
945 icmph->type, icmph->code,
946 &iph->daddr, skb->dev->name);
947 goto out;
948 }
949
950 icmp_socket_deliver(skb, info);
951
952out:
953 return true;
954out_err:
955 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
956 return false;
957}
958
959
960/*
961 * Handle ICMP_REDIRECT.
962 */
963
964static bool icmp_redirect(struct sk_buff *skb)
965{
966 if (skb->len < sizeof(struct iphdr)) {
967 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
968 return false;
969 }
970
971 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
972 /* there aught to be a stat */
973 return false;
974 }
975
976 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
977 return true;
978}
979
980/*
981 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
982 *
983 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
984 * requests.
985 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
986 * included in the reply.
987 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
988 * echo requests, MUST have default=NOT.
989 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
990 * Extended Echo Functionality, MUST be disabled by default
991 * See also WRT handling of options once they are done and working.
992 */
993
994static bool icmp_echo(struct sk_buff *skb)
995{
996 struct icmp_bxm icmp_param;
997 struct net *net;
998
999 net = dev_net(skb_dst(skb)->dev);
1000 /* should there be an ICMP stat for ignored echos? */
1001 if (net->ipv4.sysctl_icmp_echo_ignore_all)
1002 return true;
1003
1004 icmp_param.data.icmph = *icmp_hdr(skb);
1005 icmp_param.skb = skb;
1006 icmp_param.offset = 0;
1007 icmp_param.data_len = skb->len;
1008 icmp_param.head_len = sizeof(struct icmphdr);
1009
1010 if (icmp_param.data.icmph.type == ICMP_ECHO)
1011 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1012 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1013 return true;
1014
1015 icmp_reply(&icmp_param, skb);
1016 return true;
1017}
1018
1019/* Helper for icmp_echo and icmpv6_echo_reply.
1020 * Searches for net_device that matches PROBE interface identifier
1021 * and builds PROBE reply message in icmphdr.
1022 *
1023 * Returns false if PROBE responses are disabled via sysctl
1024 */
1025
1026bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1027{
1028 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1029 struct icmp_ext_echo_iio *iio, _iio;
1030 struct net *net = dev_net(skb->dev);
1031 struct net_device *dev;
1032 char buff[IFNAMSIZ];
1033 u16 ident_len;
1034 u8 status;
1035
1036 if (!net->ipv4.sysctl_icmp_echo_enable_probe)
1037 return false;
1038
1039 /* We currently only support probing interfaces on the proxy node
1040 * Check to ensure L-bit is set
1041 */
1042 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1043 return false;
1044 /* Clear status bits in reply message */
1045 icmphdr->un.echo.sequence &= htons(0xFF00);
1046 if (icmphdr->type == ICMP_EXT_ECHO)
1047 icmphdr->type = ICMP_EXT_ECHOREPLY;
1048 else
1049 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1050 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1051 /* Size of iio is class_type dependent.
1052 * Only check header here and assign length based on ctype in the switch statement
1053 */
1054 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1055 if (!ext_hdr || !iio)
1056 goto send_mal_query;
1057 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr))
1058 goto send_mal_query;
1059 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1060 status = 0;
1061 dev = NULL;
1062 switch (iio->extobj_hdr.class_type) {
1063 case ICMP_EXT_ECHO_CTYPE_NAME:
1064 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(_iio), &_iio);
1065 if (ident_len >= IFNAMSIZ)
1066 goto send_mal_query;
1067 memset(buff, 0, sizeof(buff));
1068 memcpy(buff, &iio->ident.name, ident_len);
1069 dev = dev_get_by_name(net, buff);
1070 break;
1071 case ICMP_EXT_ECHO_CTYPE_INDEX:
1072 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr) +
1073 sizeof(iio->ident.ifindex), &_iio);
1074 if (ident_len != sizeof(iio->ident.ifindex))
1075 goto send_mal_query;
1076 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1077 break;
1078 case ICMP_EXT_ECHO_CTYPE_ADDR:
1079 if (ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1080 iio->ident.addr.ctype3_hdr.addrlen)
1081 goto send_mal_query;
1082 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1083 case ICMP_AFI_IP:
1084 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr) +
1085 sizeof(struct in_addr), &_iio);
1086 if (ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1087 sizeof(struct in_addr))
1088 goto send_mal_query;
1089 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1090 break;
1091#if IS_ENABLED(CONFIG_IPV6)
1092 case ICMP_AFI_IP6:
1093 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(_iio), &_iio);
1094 if (ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1095 sizeof(struct in6_addr))
1096 goto send_mal_query;
1097 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1098 if (dev)
1099 dev_hold(dev);
1100 break;
1101#endif
1102 default:
1103 goto send_mal_query;
1104 }
1105 break;
1106 default:
1107 goto send_mal_query;
1108 }
1109 if (!dev) {
1110 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1111 return true;
1112 }
1113 /* Fill bits in reply message */
1114 if (dev->flags & IFF_UP)
1115 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1116 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1117 status |= ICMP_EXT_ECHOREPLY_IPV4;
1118 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1119 status |= ICMP_EXT_ECHOREPLY_IPV6;
1120 dev_put(dev);
1121 icmphdr->un.echo.sequence |= htons(status);
1122 return true;
1123send_mal_query:
1124 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1125 return true;
1126}
1127EXPORT_SYMBOL_GPL(icmp_build_probe);
1128
1129/*
1130 * Handle ICMP Timestamp requests.
1131 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1132 * SHOULD be in the kernel for minimum random latency.
1133 * MUST be accurate to a few minutes.
1134 * MUST be updated at least at 15Hz.
1135 */
1136static bool icmp_timestamp(struct sk_buff *skb)
1137{
1138 struct icmp_bxm icmp_param;
1139 /*
1140 * Too short.
1141 */
1142 if (skb->len < 4)
1143 goto out_err;
1144
1145 /*
1146 * Fill in the current time as ms since midnight UT:
1147 */
1148 icmp_param.data.times[1] = inet_current_timestamp();
1149 icmp_param.data.times[2] = icmp_param.data.times[1];
1150
1151 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1152
1153 icmp_param.data.icmph = *icmp_hdr(skb);
1154 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1155 icmp_param.data.icmph.code = 0;
1156 icmp_param.skb = skb;
1157 icmp_param.offset = 0;
1158 icmp_param.data_len = 0;
1159 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1160 icmp_reply(&icmp_param, skb);
1161 return true;
1162
1163out_err:
1164 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1165 return false;
1166}
1167
1168static bool icmp_discard(struct sk_buff *skb)
1169{
1170 /* pretend it was a success */
1171 return true;
1172}
1173
1174/*
1175 * Deal with incoming ICMP packets.
1176 */
1177int icmp_rcv(struct sk_buff *skb)
1178{
1179 struct icmphdr *icmph;
1180 struct rtable *rt = skb_rtable(skb);
1181 struct net *net = dev_net(rt->dst.dev);
1182 bool success;
1183
1184 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1185 struct sec_path *sp = skb_sec_path(skb);
1186 int nh;
1187
1188 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1189 XFRM_STATE_ICMP))
1190 goto drop;
1191
1192 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1193 goto drop;
1194
1195 nh = skb_network_offset(skb);
1196 skb_set_network_header(skb, sizeof(*icmph));
1197
1198 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1199 goto drop;
1200
1201 skb_set_network_header(skb, nh);
1202 }
1203
1204 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1205
1206 if (skb_checksum_simple_validate(skb))
1207 goto csum_error;
1208
1209 if (!pskb_pull(skb, sizeof(*icmph)))
1210 goto error;
1211
1212 icmph = icmp_hdr(skb);
1213
1214 ICMPMSGIN_INC_STATS(net, icmph->type);
1215
1216 /* Check for ICMP Extended Echo (PROBE) messages */
1217 if (icmph->type == ICMP_EXT_ECHO) {
1218 /* We can't use icmp_pointers[].handler() because it is an array of
1219 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1220 */
1221 success = icmp_echo(skb);
1222 goto success_check;
1223 }
1224
1225 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1226 success = ping_rcv(skb);
1227 goto success_check;
1228 }
1229
1230 /*
1231 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1232 *
1233 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1234 * discarded.
1235 */
1236 if (icmph->type > NR_ICMP_TYPES)
1237 goto error;
1238
1239 /*
1240 * Parse the ICMP message
1241 */
1242
1243 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1244 /*
1245 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1246 * silently ignored (we let user decide with a sysctl).
1247 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1248 * discarded if to broadcast/multicast.
1249 */
1250 if ((icmph->type == ICMP_ECHO ||
1251 icmph->type == ICMP_TIMESTAMP) &&
1252 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1253 goto error;
1254 }
1255 if (icmph->type != ICMP_ECHO &&
1256 icmph->type != ICMP_TIMESTAMP &&
1257 icmph->type != ICMP_ADDRESS &&
1258 icmph->type != ICMP_ADDRESSREPLY) {
1259 goto error;
1260 }
1261 }
1262
1263 success = icmp_pointers[icmph->type].handler(skb);
1264success_check:
1265 if (success) {
1266 consume_skb(skb);
1267 return NET_RX_SUCCESS;
1268 }
1269
1270drop:
1271 kfree_skb(skb);
1272 return NET_RX_DROP;
1273csum_error:
1274 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1275error:
1276 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1277 goto drop;
1278}
1279
1280static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1281{
1282 struct icmp_extobj_hdr *objh, _objh;
1283 struct icmp_ext_hdr *exth, _exth;
1284 u16 olen;
1285
1286 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1287 if (!exth)
1288 return false;
1289 if (exth->version != 2)
1290 return true;
1291
1292 if (exth->checksum &&
1293 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1294 return false;
1295
1296 off += sizeof(_exth);
1297 while (off < skb->len) {
1298 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1299 if (!objh)
1300 return false;
1301
1302 olen = ntohs(objh->length);
1303 if (olen < sizeof(_objh))
1304 return false;
1305
1306 off += olen;
1307 if (off > skb->len)
1308 return false;
1309 }
1310
1311 return true;
1312}
1313
1314void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1315 struct sock_ee_data_rfc4884 *out,
1316 int thlen, int off)
1317{
1318 int hlen;
1319
1320 /* original datagram headers: end of icmph to payload (skb->data) */
1321 hlen = -skb_transport_offset(skb) - thlen;
1322
1323 /* per rfc 4884: minimal datagram length of 128 bytes */
1324 if (off < 128 || off < hlen)
1325 return;
1326
1327 /* kernel has stripped headers: return payload offset in bytes */
1328 off -= hlen;
1329 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1330 return;
1331
1332 out->len = off;
1333
1334 if (!ip_icmp_error_rfc4884_validate(skb, off))
1335 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1336}
1337EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1338
1339int icmp_err(struct sk_buff *skb, u32 info)
1340{
1341 struct iphdr *iph = (struct iphdr *)skb->data;
1342 int offset = iph->ihl<<2;
1343 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1344 int type = icmp_hdr(skb)->type;
1345 int code = icmp_hdr(skb)->code;
1346 struct net *net = dev_net(skb->dev);
1347
1348 /*
1349 * Use ping_err to handle all icmp errors except those
1350 * triggered by ICMP_ECHOREPLY which sent from kernel.
1351 */
1352 if (icmph->type != ICMP_ECHOREPLY) {
1353 ping_err(skb, offset, info);
1354 return 0;
1355 }
1356
1357 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1358 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1359 else if (type == ICMP_REDIRECT)
1360 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1361
1362 return 0;
1363}
1364
1365/*
1366 * This table is the definition of how we handle ICMP.
1367 */
1368static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1369 [ICMP_ECHOREPLY] = {
1370 .handler = ping_rcv,
1371 },
1372 [1] = {
1373 .handler = icmp_discard,
1374 .error = 1,
1375 },
1376 [2] = {
1377 .handler = icmp_discard,
1378 .error = 1,
1379 },
1380 [ICMP_DEST_UNREACH] = {
1381 .handler = icmp_unreach,
1382 .error = 1,
1383 },
1384 [ICMP_SOURCE_QUENCH] = {
1385 .handler = icmp_unreach,
1386 .error = 1,
1387 },
1388 [ICMP_REDIRECT] = {
1389 .handler = icmp_redirect,
1390 .error = 1,
1391 },
1392 [6] = {
1393 .handler = icmp_discard,
1394 .error = 1,
1395 },
1396 [7] = {
1397 .handler = icmp_discard,
1398 .error = 1,
1399 },
1400 [ICMP_ECHO] = {
1401 .handler = icmp_echo,
1402 },
1403 [9] = {
1404 .handler = icmp_discard,
1405 .error = 1,
1406 },
1407 [10] = {
1408 .handler = icmp_discard,
1409 .error = 1,
1410 },
1411 [ICMP_TIME_EXCEEDED] = {
1412 .handler = icmp_unreach,
1413 .error = 1,
1414 },
1415 [ICMP_PARAMETERPROB] = {
1416 .handler = icmp_unreach,
1417 .error = 1,
1418 },
1419 [ICMP_TIMESTAMP] = {
1420 .handler = icmp_timestamp,
1421 },
1422 [ICMP_TIMESTAMPREPLY] = {
1423 .handler = icmp_discard,
1424 },
1425 [ICMP_INFO_REQUEST] = {
1426 .handler = icmp_discard,
1427 },
1428 [ICMP_INFO_REPLY] = {
1429 .handler = icmp_discard,
1430 },
1431 [ICMP_ADDRESS] = {
1432 .handler = icmp_discard,
1433 },
1434 [ICMP_ADDRESSREPLY] = {
1435 .handler = icmp_discard,
1436 },
1437};
1438
1439static void __net_exit icmp_sk_exit(struct net *net)
1440{
1441 int i;
1442
1443 for_each_possible_cpu(i)
1444 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1445 free_percpu(net->ipv4.icmp_sk);
1446 net->ipv4.icmp_sk = NULL;
1447}
1448
1449static int __net_init icmp_sk_init(struct net *net)
1450{
1451 int i, err;
1452
1453 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1454 if (!net->ipv4.icmp_sk)
1455 return -ENOMEM;
1456
1457 for_each_possible_cpu(i) {
1458 struct sock *sk;
1459
1460 err = inet_ctl_sock_create(&sk, PF_INET,
1461 SOCK_RAW, IPPROTO_ICMP, net);
1462 if (err < 0)
1463 goto fail;
1464
1465 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1466
1467 /* Enough space for 2 64K ICMP packets, including
1468 * sk_buff/skb_shared_info struct overhead.
1469 */
1470 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1471
1472 /*
1473 * Speedup sock_wfree()
1474 */
1475 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1476 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1477 }
1478
1479 /* Control parameters for ECHO replies. */
1480 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1481 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1482 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1483
1484 /* Control parameter - ignore bogus broadcast responses? */
1485 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1486
1487 /*
1488 * Configurable global rate limit.
1489 *
1490 * ratelimit defines tokens/packet consumed for dst->rate_token
1491 * bucket ratemask defines which icmp types are ratelimited by
1492 * setting it's bit position.
1493 *
1494 * default:
1495 * dest unreachable (3), source quench (4),
1496 * time exceeded (11), parameter problem (12)
1497 */
1498
1499 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1500 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1501 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1502
1503 return 0;
1504
1505fail:
1506 icmp_sk_exit(net);
1507 return err;
1508}
1509
1510static struct pernet_operations __net_initdata icmp_sk_ops = {
1511 .init = icmp_sk_init,
1512 .exit = icmp_sk_exit,
1513};
1514
1515int __init icmp_init(void)
1516{
1517 return register_pernet_subsys(&icmp_sk_ops);
1518}
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}