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