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.
253 */
254 if (!icmp_global.credit) {
255 delta = min_t(u32, now - icmp_global.stamp, HZ);
256 if (delta < HZ / 50)
257 return false;
258 }
259
260 spin_lock(&icmp_global.lock);
261 delta = min_t(u32, now - icmp_global.stamp, HZ);
262 if (delta >= HZ / 50) {
263 incr = sysctl_icmp_msgs_per_sec * delta / HZ ;
264 if (incr)
265 icmp_global.stamp = now;
266 }
267 credit = min_t(u32, icmp_global.credit + incr, sysctl_icmp_msgs_burst);
268 if (credit) {
269 credit--;
270 rc = true;
271 }
272 icmp_global.credit = credit;
273 spin_unlock(&icmp_global.lock);
274 return rc;
275}
276EXPORT_SYMBOL(icmp_global_allow);
277
278static bool icmpv4_mask_allow(struct net *net, int type, int code)
279{
280 if (type > NR_ICMP_TYPES)
281 return true;
282
283 /* Don't limit PMTU discovery. */
284 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 return true;
286
287 /* Limit if icmp type is enabled in ratemask. */
288 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
289 return true;
290
291 return false;
292}
293
294static bool icmpv4_global_allow(struct net *net, int type, int code)
295{
296 if (icmpv4_mask_allow(net, type, code))
297 return true;
298
299 if (icmp_global_allow())
300 return true;
301
302 return false;
303}
304
305/*
306 * Send an ICMP frame.
307 */
308
309static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
310 struct flowi4 *fl4, int type, int code)
311{
312 struct dst_entry *dst = &rt->dst;
313 struct inet_peer *peer;
314 bool rc = true;
315 int vif;
316
317 if (icmpv4_mask_allow(net, type, code))
318 goto out;
319
320 /* No rate limit on loopback */
321 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
322 goto out;
323
324 vif = l3mdev_master_ifindex(dst->dev);
325 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
326 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
327 if (peer)
328 inet_putpeer(peer);
329out:
330 return rc;
331}
332
333/*
334 * Maintain the counters used in the SNMP statistics for outgoing ICMP
335 */
336void icmp_out_count(struct net *net, unsigned char type)
337{
338 ICMPMSGOUT_INC_STATS(net, type);
339 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
340}
341
342/*
343 * Checksum each fragment, and on the first include the headers and final
344 * checksum.
345 */
346static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
347 struct sk_buff *skb)
348{
349 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
350 __wsum csum;
351
352 csum = skb_copy_and_csum_bits(icmp_param->skb,
353 icmp_param->offset + offset,
354 to, len, 0);
355
356 skb->csum = csum_block_add(skb->csum, csum, odd);
357 if (icmp_pointers[icmp_param->data.icmph.type].error)
358 nf_ct_attach(skb, icmp_param->skb);
359 return 0;
360}
361
362static void icmp_push_reply(struct icmp_bxm *icmp_param,
363 struct flowi4 *fl4,
364 struct ipcm_cookie *ipc, struct rtable **rt)
365{
366 struct sock *sk;
367 struct sk_buff *skb;
368
369 sk = icmp_sk(dev_net((*rt)->dst.dev));
370 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 icmp_param->data_len+icmp_param->head_len,
372 icmp_param->head_len,
373 ipc, rt, MSG_DONTWAIT) < 0) {
374 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 ip_flush_pending_frames(sk);
376 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 struct icmphdr *icmph = icmp_hdr(skb);
378 __wsum csum = 0;
379 struct sk_buff *skb1;
380
381 skb_queue_walk(&sk->sk_write_queue, skb1) {
382 csum = csum_add(csum, skb1->csum);
383 }
384 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
385 (char *)icmph,
386 icmp_param->head_len, csum);
387 icmph->checksum = csum_fold(csum);
388 skb->ip_summed = CHECKSUM_NONE;
389 ip_push_pending_frames(sk, fl4);
390 }
391}
392
393/*
394 * Driving logic for building and sending ICMP messages.
395 */
396
397static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
398{
399 struct ipcm_cookie ipc;
400 struct rtable *rt = skb_rtable(skb);
401 struct net *net = dev_net(rt->dst.dev);
402 struct flowi4 fl4;
403 struct sock *sk;
404 struct inet_sock *inet;
405 __be32 daddr, saddr;
406 u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 int type = icmp_param->data.icmph.type;
408 int code = icmp_param->data.icmph.code;
409
410 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 return;
412
413 /* Needed by both icmp_global_allow and icmp_xmit_lock */
414 local_bh_disable();
415
416 /* global icmp_msgs_per_sec */
417 if (!icmpv4_global_allow(net, type, code))
418 goto out_bh_enable;
419
420 sk = icmp_xmit_lock(net);
421 if (!sk)
422 goto out_bh_enable;
423 inet = inet_sk(sk);
424
425 icmp_param->data.icmph.checksum = 0;
426
427 ipcm_init(&ipc);
428 inet->tos = ip_hdr(skb)->tos;
429 sk->sk_mark = mark;
430 daddr = ipc.addr = ip_hdr(skb)->saddr;
431 saddr = fib_compute_spec_dst(skb);
432
433 if (icmp_param->replyopts.opt.opt.optlen) {
434 ipc.opt = &icmp_param->replyopts.opt;
435 if (ipc.opt->opt.srr)
436 daddr = icmp_param->replyopts.opt.opt.faddr;
437 }
438 memset(&fl4, 0, sizeof(fl4));
439 fl4.daddr = daddr;
440 fl4.saddr = saddr;
441 fl4.flowi4_mark = mark;
442 fl4.flowi4_uid = sock_net_uid(net, NULL);
443 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 fl4.flowi4_proto = IPPROTO_ICMP;
445 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
447 rt = ip_route_output_key(net, &fl4);
448 if (IS_ERR(rt))
449 goto out_unlock;
450 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
452 ip_rt_put(rt);
453out_unlock:
454 icmp_xmit_unlock(sk);
455out_bh_enable:
456 local_bh_enable();
457}
458
459static struct rtable *icmp_route_lookup(struct net *net,
460 struct flowi4 *fl4,
461 struct sk_buff *skb_in,
462 const struct iphdr *iph,
463 __be32 saddr, u8 tos, u32 mark,
464 int type, int code,
465 struct icmp_bxm *param)
466{
467 struct rtable *rt, *rt2;
468 struct flowi4 fl4_dec;
469 int err;
470
471 memset(fl4, 0, sizeof(*fl4));
472 fl4->daddr = (param->replyopts.opt.opt.srr ?
473 param->replyopts.opt.opt.faddr : iph->saddr);
474 fl4->saddr = saddr;
475 fl4->flowi4_mark = mark;
476 fl4->flowi4_uid = sock_net_uid(net, NULL);
477 fl4->flowi4_tos = RT_TOS(tos);
478 fl4->flowi4_proto = IPPROTO_ICMP;
479 fl4->fl4_icmp_type = type;
480 fl4->fl4_icmp_code = code;
481 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
482
483 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
484 rt = ip_route_output_key_hash(net, fl4, skb_in);
485 if (IS_ERR(rt))
486 return rt;
487
488 /* No need to clone since we're just using its address. */
489 rt2 = rt;
490
491 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
492 flowi4_to_flowi(fl4), NULL, 0);
493 if (!IS_ERR(rt)) {
494 if (rt != rt2)
495 return rt;
496 } else if (PTR_ERR(rt) == -EPERM) {
497 rt = NULL;
498 } else
499 return rt;
500
501 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
502 if (err)
503 goto relookup_failed;
504
505 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
506 fl4_dec.saddr) == RTN_LOCAL) {
507 rt2 = __ip_route_output_key(net, &fl4_dec);
508 if (IS_ERR(rt2))
509 err = PTR_ERR(rt2);
510 } else {
511 struct flowi4 fl4_2 = {};
512 unsigned long orefdst;
513
514 fl4_2.daddr = fl4_dec.saddr;
515 rt2 = ip_route_output_key(net, &fl4_2);
516 if (IS_ERR(rt2)) {
517 err = PTR_ERR(rt2);
518 goto relookup_failed;
519 }
520 /* Ugh! */
521 orefdst = skb_in->_skb_refdst; /* save old refdst */
522 skb_dst_set(skb_in, NULL);
523 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
524 RT_TOS(tos), rt2->dst.dev);
525
526 dst_release(&rt2->dst);
527 rt2 = skb_rtable(skb_in);
528 skb_in->_skb_refdst = orefdst; /* restore old refdst */
529 }
530
531 if (err)
532 goto relookup_failed;
533
534 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
535 flowi4_to_flowi(&fl4_dec), NULL,
536 XFRM_LOOKUP_ICMP);
537 if (!IS_ERR(rt2)) {
538 dst_release(&rt->dst);
539 memcpy(fl4, &fl4_dec, sizeof(*fl4));
540 rt = rt2;
541 } else if (PTR_ERR(rt2) == -EPERM) {
542 if (rt)
543 dst_release(&rt->dst);
544 return rt2;
545 } else {
546 err = PTR_ERR(rt2);
547 goto relookup_failed;
548 }
549 return rt;
550
551relookup_failed:
552 if (rt)
553 return rt;
554 return ERR_PTR(err);
555}
556
557/*
558 * Send an ICMP message in response to a situation
559 *
560 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
561 * MAY send more (we do).
562 * MUST NOT change this header information.
563 * MUST NOT reply to a multicast/broadcast IP address.
564 * MUST NOT reply to a multicast/broadcast MAC address.
565 * MUST reply to only the first fragment.
566 */
567
568void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
569 const struct ip_options *opt)
570{
571 struct iphdr *iph;
572 int room;
573 struct icmp_bxm icmp_param;
574 struct rtable *rt = skb_rtable(skb_in);
575 struct ipcm_cookie ipc;
576 struct flowi4 fl4;
577 __be32 saddr;
578 u8 tos;
579 u32 mark;
580 struct net *net;
581 struct sock *sk;
582
583 if (!rt)
584 goto out;
585
586 if (rt->dst.dev)
587 net = dev_net(rt->dst.dev);
588 else if (skb_in->dev)
589 net = dev_net(skb_in->dev);
590 else
591 goto out;
592
593 /*
594 * Find the original header. It is expected to be valid, of course.
595 * Check this, icmp_send is called from the most obscure devices
596 * sometimes.
597 */
598 iph = ip_hdr(skb_in);
599
600 if ((u8 *)iph < skb_in->head ||
601 (skb_network_header(skb_in) + sizeof(*iph)) >
602 skb_tail_pointer(skb_in))
603 goto out;
604
605 /*
606 * No replies to physical multicast/broadcast
607 */
608 if (skb_in->pkt_type != PACKET_HOST)
609 goto out;
610
611 /*
612 * Now check at the protocol level
613 */
614 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
615 goto out;
616
617 /*
618 * Only reply to fragment 0. We byte re-order the constant
619 * mask for efficiency.
620 */
621 if (iph->frag_off & htons(IP_OFFSET))
622 goto out;
623
624 /*
625 * If we send an ICMP error to an ICMP error a mess would result..
626 */
627 if (icmp_pointers[type].error) {
628 /*
629 * We are an error, check if we are replying to an
630 * ICMP error
631 */
632 if (iph->protocol == IPPROTO_ICMP) {
633 u8 _inner_type, *itp;
634
635 itp = skb_header_pointer(skb_in,
636 skb_network_header(skb_in) +
637 (iph->ihl << 2) +
638 offsetof(struct icmphdr,
639 type) -
640 skb_in->data,
641 sizeof(_inner_type),
642 &_inner_type);
643 if (!itp)
644 goto out;
645
646 /*
647 * Assume any unknown ICMP type is an error. This
648 * isn't specified by the RFC, but think about it..
649 */
650 if (*itp > NR_ICMP_TYPES ||
651 icmp_pointers[*itp].error)
652 goto out;
653 }
654 }
655
656 /* Needed by both icmp_global_allow and icmp_xmit_lock */
657 local_bh_disable();
658
659 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
660 * incoming dev is loopback. If outgoing dev change to not be
661 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
662 */
663 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
664 !icmpv4_global_allow(net, type, code))
665 goto out_bh_enable;
666
667 sk = icmp_xmit_lock(net);
668 if (!sk)
669 goto out_bh_enable;
670
671 /*
672 * Construct source address and options.
673 */
674
675 saddr = iph->daddr;
676 if (!(rt->rt_flags & RTCF_LOCAL)) {
677 struct net_device *dev = NULL;
678
679 rcu_read_lock();
680 if (rt_is_input_route(rt) &&
681 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
682 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
683
684 if (dev)
685 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
686 else
687 saddr = 0;
688 rcu_read_unlock();
689 }
690
691 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
692 IPTOS_PREC_INTERNETCONTROL) :
693 iph->tos;
694 mark = IP4_REPLY_MARK(net, skb_in->mark);
695
696 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
697 goto out_unlock;
698
699
700 /*
701 * Prepare data for ICMP header.
702 */
703
704 icmp_param.data.icmph.type = type;
705 icmp_param.data.icmph.code = code;
706 icmp_param.data.icmph.un.gateway = info;
707 icmp_param.data.icmph.checksum = 0;
708 icmp_param.skb = skb_in;
709 icmp_param.offset = skb_network_offset(skb_in);
710 inet_sk(sk)->tos = tos;
711 sk->sk_mark = mark;
712 ipcm_init(&ipc);
713 ipc.addr = iph->saddr;
714 ipc.opt = &icmp_param.replyopts.opt;
715
716 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
717 type, code, &icmp_param);
718 if (IS_ERR(rt))
719 goto out_unlock;
720
721 /* peer icmp_ratelimit */
722 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
723 goto ende;
724
725 /* RFC says return as much as we can without exceeding 576 bytes. */
726
727 room = dst_mtu(&rt->dst);
728 if (room > 576)
729 room = 576;
730 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
731 room -= sizeof(struct icmphdr);
732
733 icmp_param.data_len = skb_in->len - icmp_param.offset;
734 if (icmp_param.data_len > room)
735 icmp_param.data_len = room;
736 icmp_param.head_len = sizeof(struct icmphdr);
737
738 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
739ende:
740 ip_rt_put(rt);
741out_unlock:
742 icmp_xmit_unlock(sk);
743out_bh_enable:
744 local_bh_enable();
745out:;
746}
747EXPORT_SYMBOL(__icmp_send);
748
749
750static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
751{
752 const struct iphdr *iph = (const struct iphdr *) skb->data;
753 const struct net_protocol *ipprot;
754 int protocol = iph->protocol;
755
756 /* Checkin full IP header plus 8 bytes of protocol to
757 * avoid additional coding at protocol handlers.
758 */
759 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
760 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
761 return;
762 }
763
764 raw_icmp_error(skb, protocol, info);
765
766 ipprot = rcu_dereference(inet_protos[protocol]);
767 if (ipprot && ipprot->err_handler)
768 ipprot->err_handler(skb, info);
769}
770
771static bool icmp_tag_validation(int proto)
772{
773 bool ok;
774
775 rcu_read_lock();
776 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
777 rcu_read_unlock();
778 return ok;
779}
780
781/*
782 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
783 * ICMP_PARAMETERPROB.
784 */
785
786static bool icmp_unreach(struct sk_buff *skb)
787{
788 const struct iphdr *iph;
789 struct icmphdr *icmph;
790 struct net *net;
791 u32 info = 0;
792
793 net = dev_net(skb_dst(skb)->dev);
794
795 /*
796 * Incomplete header ?
797 * Only checks for the IP header, there should be an
798 * additional check for longer headers in upper levels.
799 */
800
801 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
802 goto out_err;
803
804 icmph = icmp_hdr(skb);
805 iph = (const struct iphdr *)skb->data;
806
807 if (iph->ihl < 5) /* Mangled header, drop. */
808 goto out_err;
809
810 switch (icmph->type) {
811 case ICMP_DEST_UNREACH:
812 switch (icmph->code & 15) {
813 case ICMP_NET_UNREACH:
814 case ICMP_HOST_UNREACH:
815 case ICMP_PROT_UNREACH:
816 case ICMP_PORT_UNREACH:
817 break;
818 case ICMP_FRAG_NEEDED:
819 /* for documentation of the ip_no_pmtu_disc
820 * values please see
821 * Documentation/networking/ip-sysctl.txt
822 */
823 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
824 default:
825 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
826 &iph->daddr);
827 break;
828 case 2:
829 goto out;
830 case 3:
831 if (!icmp_tag_validation(iph->protocol))
832 goto out;
833 /* fall through */
834 case 0:
835 info = ntohs(icmph->un.frag.mtu);
836 }
837 break;
838 case ICMP_SR_FAILED:
839 net_dbg_ratelimited("%pI4: Source Route Failed\n",
840 &iph->daddr);
841 break;
842 default:
843 break;
844 }
845 if (icmph->code > NR_ICMP_UNREACH)
846 goto out;
847 break;
848 case ICMP_PARAMETERPROB:
849 info = ntohl(icmph->un.gateway) >> 24;
850 break;
851 case ICMP_TIME_EXCEEDED:
852 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
853 if (icmph->code == ICMP_EXC_FRAGTIME)
854 goto out;
855 break;
856 }
857
858 /*
859 * Throw it at our lower layers
860 *
861 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
862 * header.
863 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
864 * transport layer.
865 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
866 * transport layer.
867 */
868
869 /*
870 * Check the other end isn't violating RFC 1122. Some routers send
871 * bogus responses to broadcast frames. If you see this message
872 * first check your netmask matches at both ends, if it does then
873 * get the other vendor to fix their kit.
874 */
875
876 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
877 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
878 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
879 &ip_hdr(skb)->saddr,
880 icmph->type, icmph->code,
881 &iph->daddr, skb->dev->name);
882 goto out;
883 }
884
885 icmp_socket_deliver(skb, info);
886
887out:
888 return true;
889out_err:
890 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
891 return false;
892}
893
894
895/*
896 * Handle ICMP_REDIRECT.
897 */
898
899static bool icmp_redirect(struct sk_buff *skb)
900{
901 if (skb->len < sizeof(struct iphdr)) {
902 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
903 return false;
904 }
905
906 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
907 /* there aught to be a stat */
908 return false;
909 }
910
911 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
912 return true;
913}
914
915/*
916 * Handle ICMP_ECHO ("ping") requests.
917 *
918 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
919 * requests.
920 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
921 * included in the reply.
922 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
923 * echo requests, MUST have default=NOT.
924 * See also WRT handling of options once they are done and working.
925 */
926
927static bool icmp_echo(struct sk_buff *skb)
928{
929 struct net *net;
930
931 net = dev_net(skb_dst(skb)->dev);
932 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
933 struct icmp_bxm icmp_param;
934
935 icmp_param.data.icmph = *icmp_hdr(skb);
936 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
937 icmp_param.skb = skb;
938 icmp_param.offset = 0;
939 icmp_param.data_len = skb->len;
940 icmp_param.head_len = sizeof(struct icmphdr);
941 icmp_reply(&icmp_param, skb);
942 }
943 /* should there be an ICMP stat for ignored echos? */
944 return true;
945}
946
947/*
948 * Handle ICMP Timestamp requests.
949 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
950 * SHOULD be in the kernel for minimum random latency.
951 * MUST be accurate to a few minutes.
952 * MUST be updated at least at 15Hz.
953 */
954static bool icmp_timestamp(struct sk_buff *skb)
955{
956 struct icmp_bxm icmp_param;
957 /*
958 * Too short.
959 */
960 if (skb->len < 4)
961 goto out_err;
962
963 /*
964 * Fill in the current time as ms since midnight UT:
965 */
966 icmp_param.data.times[1] = inet_current_timestamp();
967 icmp_param.data.times[2] = icmp_param.data.times[1];
968
969 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
970
971 icmp_param.data.icmph = *icmp_hdr(skb);
972 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
973 icmp_param.data.icmph.code = 0;
974 icmp_param.skb = skb;
975 icmp_param.offset = 0;
976 icmp_param.data_len = 0;
977 icmp_param.head_len = sizeof(struct icmphdr) + 12;
978 icmp_reply(&icmp_param, skb);
979 return true;
980
981out_err:
982 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
983 return false;
984}
985
986static bool icmp_discard(struct sk_buff *skb)
987{
988 /* pretend it was a success */
989 return true;
990}
991
992/*
993 * Deal with incoming ICMP packets.
994 */
995int icmp_rcv(struct sk_buff *skb)
996{
997 struct icmphdr *icmph;
998 struct rtable *rt = skb_rtable(skb);
999 struct net *net = dev_net(rt->dst.dev);
1000 bool success;
1001
1002 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1003 struct sec_path *sp = skb_sec_path(skb);
1004 int nh;
1005
1006 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1007 XFRM_STATE_ICMP))
1008 goto drop;
1009
1010 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1011 goto drop;
1012
1013 nh = skb_network_offset(skb);
1014 skb_set_network_header(skb, sizeof(*icmph));
1015
1016 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1017 goto drop;
1018
1019 skb_set_network_header(skb, nh);
1020 }
1021
1022 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1023
1024 if (skb_checksum_simple_validate(skb))
1025 goto csum_error;
1026
1027 if (!pskb_pull(skb, sizeof(*icmph)))
1028 goto error;
1029
1030 icmph = icmp_hdr(skb);
1031
1032 ICMPMSGIN_INC_STATS(net, icmph->type);
1033 /*
1034 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1035 *
1036 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1037 * discarded.
1038 */
1039 if (icmph->type > NR_ICMP_TYPES)
1040 goto error;
1041
1042
1043 /*
1044 * Parse the ICMP message
1045 */
1046
1047 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1048 /*
1049 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1050 * silently ignored (we let user decide with a sysctl).
1051 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1052 * discarded if to broadcast/multicast.
1053 */
1054 if ((icmph->type == ICMP_ECHO ||
1055 icmph->type == ICMP_TIMESTAMP) &&
1056 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1057 goto error;
1058 }
1059 if (icmph->type != ICMP_ECHO &&
1060 icmph->type != ICMP_TIMESTAMP &&
1061 icmph->type != ICMP_ADDRESS &&
1062 icmph->type != ICMP_ADDRESSREPLY) {
1063 goto error;
1064 }
1065 }
1066
1067 success = icmp_pointers[icmph->type].handler(skb);
1068
1069 if (success) {
1070 consume_skb(skb);
1071 return NET_RX_SUCCESS;
1072 }
1073
1074drop:
1075 kfree_skb(skb);
1076 return NET_RX_DROP;
1077csum_error:
1078 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1079error:
1080 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1081 goto drop;
1082}
1083
1084int icmp_err(struct sk_buff *skb, u32 info)
1085{
1086 struct iphdr *iph = (struct iphdr *)skb->data;
1087 int offset = iph->ihl<<2;
1088 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1089 int type = icmp_hdr(skb)->type;
1090 int code = icmp_hdr(skb)->code;
1091 struct net *net = dev_net(skb->dev);
1092
1093 /*
1094 * Use ping_err to handle all icmp errors except those
1095 * triggered by ICMP_ECHOREPLY which sent from kernel.
1096 */
1097 if (icmph->type != ICMP_ECHOREPLY) {
1098 ping_err(skb, offset, info);
1099 return 0;
1100 }
1101
1102 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1103 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1104 else if (type == ICMP_REDIRECT)
1105 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1106
1107 return 0;
1108}
1109
1110/*
1111 * This table is the definition of how we handle ICMP.
1112 */
1113static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1114 [ICMP_ECHOREPLY] = {
1115 .handler = ping_rcv,
1116 },
1117 [1] = {
1118 .handler = icmp_discard,
1119 .error = 1,
1120 },
1121 [2] = {
1122 .handler = icmp_discard,
1123 .error = 1,
1124 },
1125 [ICMP_DEST_UNREACH] = {
1126 .handler = icmp_unreach,
1127 .error = 1,
1128 },
1129 [ICMP_SOURCE_QUENCH] = {
1130 .handler = icmp_unreach,
1131 .error = 1,
1132 },
1133 [ICMP_REDIRECT] = {
1134 .handler = icmp_redirect,
1135 .error = 1,
1136 },
1137 [6] = {
1138 .handler = icmp_discard,
1139 .error = 1,
1140 },
1141 [7] = {
1142 .handler = icmp_discard,
1143 .error = 1,
1144 },
1145 [ICMP_ECHO] = {
1146 .handler = icmp_echo,
1147 },
1148 [9] = {
1149 .handler = icmp_discard,
1150 .error = 1,
1151 },
1152 [10] = {
1153 .handler = icmp_discard,
1154 .error = 1,
1155 },
1156 [ICMP_TIME_EXCEEDED] = {
1157 .handler = icmp_unreach,
1158 .error = 1,
1159 },
1160 [ICMP_PARAMETERPROB] = {
1161 .handler = icmp_unreach,
1162 .error = 1,
1163 },
1164 [ICMP_TIMESTAMP] = {
1165 .handler = icmp_timestamp,
1166 },
1167 [ICMP_TIMESTAMPREPLY] = {
1168 .handler = icmp_discard,
1169 },
1170 [ICMP_INFO_REQUEST] = {
1171 .handler = icmp_discard,
1172 },
1173 [ICMP_INFO_REPLY] = {
1174 .handler = icmp_discard,
1175 },
1176 [ICMP_ADDRESS] = {
1177 .handler = icmp_discard,
1178 },
1179 [ICMP_ADDRESSREPLY] = {
1180 .handler = icmp_discard,
1181 },
1182};
1183
1184static void __net_exit icmp_sk_exit(struct net *net)
1185{
1186 int i;
1187
1188 for_each_possible_cpu(i)
1189 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1190 free_percpu(net->ipv4.icmp_sk);
1191 net->ipv4.icmp_sk = NULL;
1192}
1193
1194static int __net_init icmp_sk_init(struct net *net)
1195{
1196 int i, err;
1197
1198 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1199 if (!net->ipv4.icmp_sk)
1200 return -ENOMEM;
1201
1202 for_each_possible_cpu(i) {
1203 struct sock *sk;
1204
1205 err = inet_ctl_sock_create(&sk, PF_INET,
1206 SOCK_RAW, IPPROTO_ICMP, net);
1207 if (err < 0)
1208 goto fail;
1209
1210 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1211
1212 /* Enough space for 2 64K ICMP packets, including
1213 * sk_buff/skb_shared_info struct overhead.
1214 */
1215 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1216
1217 /*
1218 * Speedup sock_wfree()
1219 */
1220 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1221 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1222 }
1223
1224 /* Control parameters for ECHO replies. */
1225 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1226 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1227
1228 /* Control parameter - ignore bogus broadcast responses? */
1229 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1230
1231 /*
1232 * Configurable global rate limit.
1233 *
1234 * ratelimit defines tokens/packet consumed for dst->rate_token
1235 * bucket ratemask defines which icmp types are ratelimited by
1236 * setting it's bit position.
1237 *
1238 * default:
1239 * dest unreachable (3), source quench (4),
1240 * time exceeded (11), parameter problem (12)
1241 */
1242
1243 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1244 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1245 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1246
1247 return 0;
1248
1249fail:
1250 icmp_sk_exit(net);
1251 return err;
1252}
1253
1254static struct pernet_operations __net_initdata icmp_sk_ops = {
1255 .init = icmp_sk_init,
1256 .exit = icmp_sk_exit,
1257};
1258
1259int __init icmp_init(void)
1260{
1261 return register_pernet_subsys(&icmp_sk_ops);
1262}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NET3: Implementation of the ICMP protocol layer.
4 *
5 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Some of the function names and the icmp unreach table for this
8 * module were derived from [icmp.c 1.0.11 06/02/93] by
9 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10 * Other than that this module is a complete rewrite.
11 *
12 * Fixes:
13 * Clemens Fruhwirth : introduce global icmp rate limiting
14 * with icmp type masking ability instead
15 * of broken per type icmp timeouts.
16 * Mike Shaver : RFC1122 checks.
17 * Alan Cox : Multicast ping reply as self.
18 * Alan Cox : Fix atomicity lockup in ip_build_xmit
19 * call.
20 * Alan Cox : Added 216,128 byte paths to the MTU
21 * code.
22 * Martin Mares : RFC1812 checks.
23 * Martin Mares : Can be configured to follow redirects
24 * if acting as a router _without_ a
25 * routing protocol (RFC 1812).
26 * Martin Mares : Echo requests may be configured to
27 * be ignored (RFC 1812).
28 * Martin Mares : Limitation of ICMP error message
29 * transmit rate (RFC 1812).
30 * Martin Mares : TOS and Precedence set correctly
31 * (RFC 1812).
32 * Martin Mares : Now copying as much data from the
33 * original packet as we can without
34 * exceeding 576 bytes (RFC 1812).
35 * Willy Konynenberg : Transparent proxying support.
36 * Keith Owens : RFC1191 correction for 4.2BSD based
37 * path MTU bug.
38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
39 * valid (RFC 1812).
40 * Andi Kleen : Check all packet lengths properly
41 * and moved all kfree_skb() up to
42 * icmp_rcv.
43 * Andi Kleen : Move the rate limit bookkeeping
44 * into the dest entry and use a token
45 * bucket filter (thanks to ANK). Make
46 * the rates sysctl configurable.
47 * Yu Tianli : Fixed two ugly bugs in icmp_send
48 * - IP option length was accounted wrongly
49 * - ICMP header length was not accounted
50 * at all.
51 * Tristan Greaves : Added sysctl option to ignore bogus
52 * broadcast responses from broken routers.
53 *
54 * To Fix:
55 *
56 * - Should use skb_pull() instead of all the manual checking.
57 * This would also greatly simply some upper layer error handlers. --AK
58 */
59
60#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61
62#include <linux/module.h>
63#include <linux/types.h>
64#include <linux/jiffies.h>
65#include <linux/kernel.h>
66#include <linux/fcntl.h>
67#include <linux/socket.h>
68#include <linux/in.h>
69#include <linux/inet.h>
70#include <linux/inetdevice.h>
71#include <linux/netdevice.h>
72#include <linux/string.h>
73#include <linux/netfilter_ipv4.h>
74#include <linux/slab.h>
75#include <net/snmp.h>
76#include <net/ip.h>
77#include <net/route.h>
78#include <net/protocol.h>
79#include <net/icmp.h>
80#include <net/tcp.h>
81#include <net/udp.h>
82#include <net/raw.h>
83#include <net/ping.h>
84#include <linux/skbuff.h>
85#include <net/sock.h>
86#include <linux/errno.h>
87#include <linux/timer.h>
88#include <linux/init.h>
89#include <linux/uaccess.h>
90#include <net/checksum.h>
91#include <net/xfrm.h>
92#include <net/inet_common.h>
93#include <net/ip_fib.h>
94#include <net/l3mdev.h>
95
96/*
97 * Build xmit assembly blocks
98 */
99
100struct icmp_bxm {
101 struct sk_buff *skb;
102 int offset;
103 int data_len;
104
105 struct {
106 struct icmphdr icmph;
107 __be32 times[3];
108 } data;
109 int head_len;
110 struct ip_options_data replyopts;
111};
112
113/* An array of errno for error messages from dest unreach. */
114/* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115
116const struct icmp_err icmp_err_convert[] = {
117 {
118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
119 .fatal = 0,
120 },
121 {
122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
123 .fatal = 0,
124 },
125 {
126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
127 .fatal = 1,
128 },
129 {
130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
131 .fatal = 1,
132 },
133 {
134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
135 .fatal = 0,
136 },
137 {
138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
139 .fatal = 0,
140 },
141 {
142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
143 .fatal = 1,
144 },
145 {
146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
147 .fatal = 1,
148 },
149 {
150 .errno = ENONET, /* ICMP_HOST_ISOLATED */
151 .fatal = 1,
152 },
153 {
154 .errno = ENETUNREACH, /* ICMP_NET_ANO */
155 .fatal = 1,
156 },
157 {
158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
163 .fatal = 0,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
167 .fatal = 0,
168 },
169 {
170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
171 .fatal = 1,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
175 .fatal = 1,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
179 .fatal = 1,
180 },
181};
182EXPORT_SYMBOL(icmp_err_convert);
183
184/*
185 * ICMP control array. This specifies what to do with each ICMP.
186 */
187
188struct icmp_control {
189 enum skb_drop_reason (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191};
192
193static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194
195static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
196
197/* Called with BH disabled */
198static inline struct sock *icmp_xmit_lock(struct net *net)
199{
200 struct sock *sk;
201
202 sk = this_cpu_read(ipv4_icmp_sk);
203
204 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
205 /* This can happen if the output path signals a
206 * dst_link_failure() for an outgoing ICMP packet.
207 */
208 return NULL;
209 }
210 sock_net_set(sk, net);
211 return sk;
212}
213
214static inline void icmp_xmit_unlock(struct sock *sk)
215{
216 sock_net_set(sk, &init_net);
217 spin_unlock(&sk->sk_lock.slock);
218}
219
220int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
221int sysctl_icmp_msgs_burst __read_mostly = 50;
222
223static struct {
224 spinlock_t lock;
225 u32 credit;
226 u32 stamp;
227} icmp_global = {
228 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
229};
230
231/**
232 * icmp_global_allow - Are we allowed to send one more ICMP message ?
233 *
234 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
235 * Returns false if we reached the limit and can not send another packet.
236 * Note: called with BH disabled
237 */
238bool icmp_global_allow(void)
239{
240 u32 credit, delta, incr = 0, now = (u32)jiffies;
241 bool rc = false;
242
243 /* Check if token bucket is empty and cannot be refilled
244 * without taking the spinlock. The READ_ONCE() are paired
245 * with the following WRITE_ONCE() in this same function.
246 */
247 if (!READ_ONCE(icmp_global.credit)) {
248 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
249 if (delta < HZ / 50)
250 return false;
251 }
252
253 spin_lock(&icmp_global.lock);
254 delta = min_t(u32, now - icmp_global.stamp, HZ);
255 if (delta >= HZ / 50) {
256 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
257 if (incr)
258 WRITE_ONCE(icmp_global.stamp, now);
259 }
260 credit = min_t(u32, icmp_global.credit + incr,
261 READ_ONCE(sysctl_icmp_msgs_burst));
262 if (credit) {
263 /* We want to use a credit of one in average, but need to randomize
264 * it for security reasons.
265 */
266 credit = max_t(int, credit - get_random_u32_below(3), 0);
267 rc = true;
268 }
269 WRITE_ONCE(icmp_global.credit, credit);
270 spin_unlock(&icmp_global.lock);
271 return rc;
272}
273EXPORT_SYMBOL(icmp_global_allow);
274
275static bool icmpv4_mask_allow(struct net *net, int type, int code)
276{
277 if (type > NR_ICMP_TYPES)
278 return true;
279
280 /* Don't limit PMTU discovery. */
281 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
282 return true;
283
284 /* Limit if icmp type is enabled in ratemask. */
285 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
286 return true;
287
288 return false;
289}
290
291static bool icmpv4_global_allow(struct net *net, int type, int code)
292{
293 if (icmpv4_mask_allow(net, type, code))
294 return true;
295
296 if (icmp_global_allow())
297 return true;
298
299 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
300 return false;
301}
302
303/*
304 * Send an ICMP frame.
305 */
306
307static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
308 struct flowi4 *fl4, int type, int code)
309{
310 struct dst_entry *dst = &rt->dst;
311 struct inet_peer *peer;
312 bool rc = true;
313 int vif;
314
315 if (icmpv4_mask_allow(net, type, code))
316 goto out;
317
318 /* No rate limit on loopback */
319 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
320 goto out;
321
322 vif = l3mdev_master_ifindex(dst->dev);
323 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
324 rc = inet_peer_xrlim_allow(peer,
325 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
326 if (peer)
327 inet_putpeer(peer);
328out:
329 if (!rc)
330 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
331 return rc;
332}
333
334/*
335 * Maintain the counters used in the SNMP statistics for outgoing ICMP
336 */
337void icmp_out_count(struct net *net, unsigned char type)
338{
339 ICMPMSGOUT_INC_STATS(net, type);
340 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
341}
342
343/*
344 * Checksum each fragment, and on the first include the headers and final
345 * checksum.
346 */
347static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
348 struct sk_buff *skb)
349{
350 struct icmp_bxm *icmp_param = from;
351 __wsum csum;
352
353 csum = skb_copy_and_csum_bits(icmp_param->skb,
354 icmp_param->offset + offset,
355 to, len);
356
357 skb->csum = csum_block_add(skb->csum, csum, odd);
358 if (icmp_pointers[icmp_param->data.icmph.type].error)
359 nf_ct_attach(skb, icmp_param->skb);
360 return 0;
361}
362
363static void icmp_push_reply(struct sock *sk,
364 struct icmp_bxm *icmp_param,
365 struct flowi4 *fl4,
366 struct ipcm_cookie *ipc, struct rtable **rt)
367{
368 struct sk_buff *skb;
369
370 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
371 icmp_param->data_len+icmp_param->head_len,
372 icmp_param->head_len,
373 ipc, rt, MSG_DONTWAIT) < 0) {
374 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
375 ip_flush_pending_frames(sk);
376 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
377 struct icmphdr *icmph = icmp_hdr(skb);
378 __wsum csum;
379 struct sk_buff *skb1;
380
381 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
382 (char *)icmph,
383 icmp_param->head_len);
384 skb_queue_walk(&sk->sk_write_queue, skb1) {
385 csum = csum_add(csum, skb1->csum);
386 }
387 icmph->checksum = csum_fold(csum);
388 skb->ip_summed = CHECKSUM_NONE;
389 ip_push_pending_frames(sk, fl4);
390 }
391}
392
393/*
394 * Driving logic for building and sending ICMP messages.
395 */
396
397static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
398{
399 struct ipcm_cookie ipc;
400 struct rtable *rt = skb_rtable(skb);
401 struct net *net = dev_net(rt->dst.dev);
402 struct flowi4 fl4;
403 struct sock *sk;
404 struct inet_sock *inet;
405 __be32 daddr, saddr;
406 u32 mark = IP4_REPLY_MARK(net, skb->mark);
407 int type = icmp_param->data.icmph.type;
408 int code = icmp_param->data.icmph.code;
409
410 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
411 return;
412
413 /* Needed by both icmp_global_allow and icmp_xmit_lock */
414 local_bh_disable();
415
416 /* global icmp_msgs_per_sec */
417 if (!icmpv4_global_allow(net, type, code))
418 goto out_bh_enable;
419
420 sk = icmp_xmit_lock(net);
421 if (!sk)
422 goto out_bh_enable;
423 inet = inet_sk(sk);
424
425 icmp_param->data.icmph.checksum = 0;
426
427 ipcm_init(&ipc);
428 inet->tos = ip_hdr(skb)->tos;
429 ipc.sockc.mark = mark;
430 daddr = ipc.addr = ip_hdr(skb)->saddr;
431 saddr = fib_compute_spec_dst(skb);
432
433 if (icmp_param->replyopts.opt.opt.optlen) {
434 ipc.opt = &icmp_param->replyopts.opt;
435 if (ipc.opt->opt.srr)
436 daddr = icmp_param->replyopts.opt.opt.faddr;
437 }
438 memset(&fl4, 0, sizeof(fl4));
439 fl4.daddr = daddr;
440 fl4.saddr = saddr;
441 fl4.flowi4_mark = mark;
442 fl4.flowi4_uid = sock_net_uid(net, NULL);
443 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
444 fl4.flowi4_proto = IPPROTO_ICMP;
445 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
446 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
447 rt = ip_route_output_key(net, &fl4);
448 if (IS_ERR(rt))
449 goto out_unlock;
450 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
451 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
452 ip_rt_put(rt);
453out_unlock:
454 icmp_xmit_unlock(sk);
455out_bh_enable:
456 local_bh_enable();
457}
458
459/*
460 * The device used for looking up which routing table to use for sending an ICMP
461 * error is preferably the source whenever it is set, which should ensure the
462 * icmp error can be sent to the source host, else lookup using the routing
463 * table of the destination device, else use the main routing table (index 0).
464 */
465static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
466{
467 struct net_device *route_lookup_dev = NULL;
468
469 if (skb->dev)
470 route_lookup_dev = skb->dev;
471 else if (skb_dst(skb))
472 route_lookup_dev = skb_dst(skb)->dev;
473 return route_lookup_dev;
474}
475
476static struct rtable *icmp_route_lookup(struct net *net,
477 struct flowi4 *fl4,
478 struct sk_buff *skb_in,
479 const struct iphdr *iph,
480 __be32 saddr, u8 tos, u32 mark,
481 int type, int code,
482 struct icmp_bxm *param)
483{
484 struct net_device *route_lookup_dev;
485 struct rtable *rt, *rt2;
486 struct flowi4 fl4_dec;
487 int err;
488
489 memset(fl4, 0, sizeof(*fl4));
490 fl4->daddr = (param->replyopts.opt.opt.srr ?
491 param->replyopts.opt.opt.faddr : iph->saddr);
492 fl4->saddr = saddr;
493 fl4->flowi4_mark = mark;
494 fl4->flowi4_uid = sock_net_uid(net, NULL);
495 fl4->flowi4_tos = RT_TOS(tos);
496 fl4->flowi4_proto = IPPROTO_ICMP;
497 fl4->fl4_icmp_type = type;
498 fl4->fl4_icmp_code = code;
499 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
500 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
501
502 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
503 rt = ip_route_output_key_hash(net, fl4, skb_in);
504 if (IS_ERR(rt))
505 return rt;
506
507 /* No need to clone since we're just using its address. */
508 rt2 = rt;
509
510 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
511 flowi4_to_flowi(fl4), NULL, 0);
512 if (!IS_ERR(rt)) {
513 if (rt != rt2)
514 return rt;
515 } else if (PTR_ERR(rt) == -EPERM) {
516 rt = NULL;
517 } else
518 return rt;
519
520 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
521 if (err)
522 goto relookup_failed;
523
524 if (inet_addr_type_dev_table(net, route_lookup_dev,
525 fl4_dec.saddr) == RTN_LOCAL) {
526 rt2 = __ip_route_output_key(net, &fl4_dec);
527 if (IS_ERR(rt2))
528 err = PTR_ERR(rt2);
529 } else {
530 struct flowi4 fl4_2 = {};
531 unsigned long orefdst;
532
533 fl4_2.daddr = fl4_dec.saddr;
534 rt2 = ip_route_output_key(net, &fl4_2);
535 if (IS_ERR(rt2)) {
536 err = PTR_ERR(rt2);
537 goto relookup_failed;
538 }
539 /* Ugh! */
540 orefdst = skb_in->_skb_refdst; /* save old refdst */
541 skb_dst_set(skb_in, NULL);
542 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
543 RT_TOS(tos), rt2->dst.dev);
544
545 dst_release(&rt2->dst);
546 rt2 = skb_rtable(skb_in);
547 skb_in->_skb_refdst = orefdst; /* restore old refdst */
548 }
549
550 if (err)
551 goto relookup_failed;
552
553 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
554 flowi4_to_flowi(&fl4_dec), NULL,
555 XFRM_LOOKUP_ICMP);
556 if (!IS_ERR(rt2)) {
557 dst_release(&rt->dst);
558 memcpy(fl4, &fl4_dec, sizeof(*fl4));
559 rt = rt2;
560 } else if (PTR_ERR(rt2) == -EPERM) {
561 if (rt)
562 dst_release(&rt->dst);
563 return rt2;
564 } else {
565 err = PTR_ERR(rt2);
566 goto relookup_failed;
567 }
568 return rt;
569
570relookup_failed:
571 if (rt)
572 return rt;
573 return ERR_PTR(err);
574}
575
576/*
577 * Send an ICMP message in response to a situation
578 *
579 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
580 * MAY send more (we do).
581 * MUST NOT change this header information.
582 * MUST NOT reply to a multicast/broadcast IP address.
583 * MUST NOT reply to a multicast/broadcast MAC address.
584 * MUST reply to only the first fragment.
585 */
586
587void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
588 const struct ip_options *opt)
589{
590 struct iphdr *iph;
591 int room;
592 struct icmp_bxm icmp_param;
593 struct rtable *rt = skb_rtable(skb_in);
594 struct ipcm_cookie ipc;
595 struct flowi4 fl4;
596 __be32 saddr;
597 u8 tos;
598 u32 mark;
599 struct net *net;
600 struct sock *sk;
601
602 if (!rt)
603 goto out;
604
605 if (rt->dst.dev)
606 net = dev_net(rt->dst.dev);
607 else if (skb_in->dev)
608 net = dev_net(skb_in->dev);
609 else
610 goto out;
611
612 /*
613 * Find the original header. It is expected to be valid, of course.
614 * Check this, icmp_send is called from the most obscure devices
615 * sometimes.
616 */
617 iph = ip_hdr(skb_in);
618
619 if ((u8 *)iph < skb_in->head ||
620 (skb_network_header(skb_in) + sizeof(*iph)) >
621 skb_tail_pointer(skb_in))
622 goto out;
623
624 /*
625 * No replies to physical multicast/broadcast
626 */
627 if (skb_in->pkt_type != PACKET_HOST)
628 goto out;
629
630 /*
631 * Now check at the protocol level
632 */
633 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
634 goto out;
635
636 /*
637 * Only reply to fragment 0. We byte re-order the constant
638 * mask for efficiency.
639 */
640 if (iph->frag_off & htons(IP_OFFSET))
641 goto out;
642
643 /*
644 * If we send an ICMP error to an ICMP error a mess would result..
645 */
646 if (icmp_pointers[type].error) {
647 /*
648 * We are an error, check if we are replying to an
649 * ICMP error
650 */
651 if (iph->protocol == IPPROTO_ICMP) {
652 u8 _inner_type, *itp;
653
654 itp = skb_header_pointer(skb_in,
655 skb_network_header(skb_in) +
656 (iph->ihl << 2) +
657 offsetof(struct icmphdr,
658 type) -
659 skb_in->data,
660 sizeof(_inner_type),
661 &_inner_type);
662 if (!itp)
663 goto out;
664
665 /*
666 * Assume any unknown ICMP type is an error. This
667 * isn't specified by the RFC, but think about it..
668 */
669 if (*itp > NR_ICMP_TYPES ||
670 icmp_pointers[*itp].error)
671 goto out;
672 }
673 }
674
675 /* Needed by both icmp_global_allow and icmp_xmit_lock */
676 local_bh_disable();
677
678 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
679 * incoming dev is loopback. If outgoing dev change to not be
680 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
681 */
682 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
683 !icmpv4_global_allow(net, type, code))
684 goto out_bh_enable;
685
686 sk = icmp_xmit_lock(net);
687 if (!sk)
688 goto out_bh_enable;
689
690 /*
691 * Construct source address and options.
692 */
693
694 saddr = iph->daddr;
695 if (!(rt->rt_flags & RTCF_LOCAL)) {
696 struct net_device *dev = NULL;
697
698 rcu_read_lock();
699 if (rt_is_input_route(rt) &&
700 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
701 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
702
703 if (dev)
704 saddr = inet_select_addr(dev, iph->saddr,
705 RT_SCOPE_LINK);
706 else
707 saddr = 0;
708 rcu_read_unlock();
709 }
710
711 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
712 IPTOS_PREC_INTERNETCONTROL) :
713 iph->tos;
714 mark = IP4_REPLY_MARK(net, skb_in->mark);
715
716 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
717 goto out_unlock;
718
719
720 /*
721 * Prepare data for ICMP header.
722 */
723
724 icmp_param.data.icmph.type = type;
725 icmp_param.data.icmph.code = code;
726 icmp_param.data.icmph.un.gateway = info;
727 icmp_param.data.icmph.checksum = 0;
728 icmp_param.skb = skb_in;
729 icmp_param.offset = skb_network_offset(skb_in);
730 inet_sk(sk)->tos = tos;
731 ipcm_init(&ipc);
732 ipc.addr = iph->saddr;
733 ipc.opt = &icmp_param.replyopts.opt;
734 ipc.sockc.mark = mark;
735
736 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
737 type, code, &icmp_param);
738 if (IS_ERR(rt))
739 goto out_unlock;
740
741 /* peer icmp_ratelimit */
742 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
743 goto ende;
744
745 /* RFC says return as much as we can without exceeding 576 bytes. */
746
747 room = dst_mtu(&rt->dst);
748 if (room > 576)
749 room = 576;
750 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
751 room -= sizeof(struct icmphdr);
752 /* Guard against tiny mtu. We need to include at least one
753 * IP network header for this message to make any sense.
754 */
755 if (room <= (int)sizeof(struct iphdr))
756 goto ende;
757
758 icmp_param.data_len = skb_in->len - icmp_param.offset;
759 if (icmp_param.data_len > room)
760 icmp_param.data_len = room;
761 icmp_param.head_len = sizeof(struct icmphdr);
762
763 /* if we don't have a source address at this point, fall back to the
764 * dummy address instead of sending out a packet with a source address
765 * of 0.0.0.0
766 */
767 if (!fl4.saddr)
768 fl4.saddr = htonl(INADDR_DUMMY);
769
770 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
771ende:
772 ip_rt_put(rt);
773out_unlock:
774 icmp_xmit_unlock(sk);
775out_bh_enable:
776 local_bh_enable();
777out:;
778}
779EXPORT_SYMBOL(__icmp_send);
780
781#if IS_ENABLED(CONFIG_NF_NAT)
782#include <net/netfilter/nf_conntrack.h>
783void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
784{
785 struct sk_buff *cloned_skb = NULL;
786 struct ip_options opts = { 0 };
787 enum ip_conntrack_info ctinfo;
788 struct nf_conn *ct;
789 __be32 orig_ip;
790
791 ct = nf_ct_get(skb_in, &ctinfo);
792 if (!ct || !(ct->status & IPS_SRC_NAT)) {
793 __icmp_send(skb_in, type, code, info, &opts);
794 return;
795 }
796
797 if (skb_shared(skb_in))
798 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
799
800 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
801 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
802 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
803 skb_network_offset(skb_in) + sizeof(struct iphdr))))
804 goto out;
805
806 orig_ip = ip_hdr(skb_in)->saddr;
807 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
808 __icmp_send(skb_in, type, code, info, &opts);
809 ip_hdr(skb_in)->saddr = orig_ip;
810out:
811 consume_skb(cloned_skb);
812}
813EXPORT_SYMBOL(icmp_ndo_send);
814#endif
815
816static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
817{
818 const struct iphdr *iph = (const struct iphdr *)skb->data;
819 const struct net_protocol *ipprot;
820 int protocol = iph->protocol;
821
822 /* Checkin full IP header plus 8 bytes of protocol to
823 * avoid additional coding at protocol handlers.
824 */
825 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
826 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
827 return;
828 }
829
830 raw_icmp_error(skb, protocol, info);
831
832 ipprot = rcu_dereference(inet_protos[protocol]);
833 if (ipprot && ipprot->err_handler)
834 ipprot->err_handler(skb, info);
835}
836
837static bool icmp_tag_validation(int proto)
838{
839 bool ok;
840
841 rcu_read_lock();
842 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
843 rcu_read_unlock();
844 return ok;
845}
846
847/*
848 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
849 * ICMP_PARAMETERPROB.
850 */
851
852static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
853{
854 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
855 const struct iphdr *iph;
856 struct icmphdr *icmph;
857 struct net *net;
858 u32 info = 0;
859
860 net = dev_net(skb_dst(skb)->dev);
861
862 /*
863 * Incomplete header ?
864 * Only checks for the IP header, there should be an
865 * additional check for longer headers in upper levels.
866 */
867
868 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
869 goto out_err;
870
871 icmph = icmp_hdr(skb);
872 iph = (const struct iphdr *)skb->data;
873
874 if (iph->ihl < 5) { /* Mangled header, drop. */
875 reason = SKB_DROP_REASON_IP_INHDR;
876 goto out_err;
877 }
878
879 switch (icmph->type) {
880 case ICMP_DEST_UNREACH:
881 switch (icmph->code & 15) {
882 case ICMP_NET_UNREACH:
883 case ICMP_HOST_UNREACH:
884 case ICMP_PROT_UNREACH:
885 case ICMP_PORT_UNREACH:
886 break;
887 case ICMP_FRAG_NEEDED:
888 /* for documentation of the ip_no_pmtu_disc
889 * values please see
890 * Documentation/networking/ip-sysctl.rst
891 */
892 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
893 default:
894 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
895 &iph->daddr);
896 break;
897 case 2:
898 goto out;
899 case 3:
900 if (!icmp_tag_validation(iph->protocol))
901 goto out;
902 fallthrough;
903 case 0:
904 info = ntohs(icmph->un.frag.mtu);
905 }
906 break;
907 case ICMP_SR_FAILED:
908 net_dbg_ratelimited("%pI4: Source Route Failed\n",
909 &iph->daddr);
910 break;
911 default:
912 break;
913 }
914 if (icmph->code > NR_ICMP_UNREACH)
915 goto out;
916 break;
917 case ICMP_PARAMETERPROB:
918 info = ntohl(icmph->un.gateway) >> 24;
919 break;
920 case ICMP_TIME_EXCEEDED:
921 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
922 if (icmph->code == ICMP_EXC_FRAGTIME)
923 goto out;
924 break;
925 }
926
927 /*
928 * Throw it at our lower layers
929 *
930 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
931 * header.
932 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
933 * transport layer.
934 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
935 * transport layer.
936 */
937
938 /*
939 * Check the other end isn't violating RFC 1122. Some routers send
940 * bogus responses to broadcast frames. If you see this message
941 * first check your netmask matches at both ends, if it does then
942 * get the other vendor to fix their kit.
943 */
944
945 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
946 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
947 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
948 &ip_hdr(skb)->saddr,
949 icmph->type, icmph->code,
950 &iph->daddr, skb->dev->name);
951 goto out;
952 }
953
954 icmp_socket_deliver(skb, info);
955
956out:
957 return reason;
958out_err:
959 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
960 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
961}
962
963
964/*
965 * Handle ICMP_REDIRECT.
966 */
967
968static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
969{
970 if (skb->len < sizeof(struct iphdr)) {
971 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
972 return SKB_DROP_REASON_PKT_TOO_SMALL;
973 }
974
975 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
976 /* there aught to be a stat */
977 return SKB_DROP_REASON_NOMEM;
978 }
979
980 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
981 return SKB_NOT_DROPPED_YET;
982}
983
984/*
985 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
986 *
987 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
988 * requests.
989 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
990 * included in the reply.
991 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
992 * echo requests, MUST have default=NOT.
993 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
994 * Extended Echo Functionality, MUST be disabled by default
995 * See also WRT handling of options once they are done and working.
996 */
997
998static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
999{
1000 struct icmp_bxm icmp_param;
1001 struct net *net;
1002
1003 net = dev_net(skb_dst(skb)->dev);
1004 /* should there be an ICMP stat for ignored echos? */
1005 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1006 return SKB_NOT_DROPPED_YET;
1007
1008 icmp_param.data.icmph = *icmp_hdr(skb);
1009 icmp_param.skb = skb;
1010 icmp_param.offset = 0;
1011 icmp_param.data_len = skb->len;
1012 icmp_param.head_len = sizeof(struct icmphdr);
1013
1014 if (icmp_param.data.icmph.type == ICMP_ECHO)
1015 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1016 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1017 return SKB_NOT_DROPPED_YET;
1018
1019 icmp_reply(&icmp_param, skb);
1020 return SKB_NOT_DROPPED_YET;
1021}
1022
1023/* Helper for icmp_echo and icmpv6_echo_reply.
1024 * Searches for net_device that matches PROBE interface identifier
1025 * and builds PROBE reply message in icmphdr.
1026 *
1027 * Returns false if PROBE responses are disabled via sysctl
1028 */
1029
1030bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1031{
1032 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1033 struct icmp_ext_echo_iio *iio, _iio;
1034 struct net *net = dev_net(skb->dev);
1035 struct net_device *dev;
1036 char buff[IFNAMSIZ];
1037 u16 ident_len;
1038 u8 status;
1039
1040 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1041 return false;
1042
1043 /* We currently only support probing interfaces on the proxy node
1044 * Check to ensure L-bit is set
1045 */
1046 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1047 return false;
1048 /* Clear status bits in reply message */
1049 icmphdr->un.echo.sequence &= htons(0xFF00);
1050 if (icmphdr->type == ICMP_EXT_ECHO)
1051 icmphdr->type = ICMP_EXT_ECHOREPLY;
1052 else
1053 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1054 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1055 /* Size of iio is class_type dependent.
1056 * Only check header here and assign length based on ctype in the switch statement
1057 */
1058 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1059 if (!ext_hdr || !iio)
1060 goto send_mal_query;
1061 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1062 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1063 goto send_mal_query;
1064 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1065 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1066 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1067 if (!iio)
1068 goto send_mal_query;
1069
1070 status = 0;
1071 dev = NULL;
1072 switch (iio->extobj_hdr.class_type) {
1073 case ICMP_EXT_ECHO_CTYPE_NAME:
1074 if (ident_len >= IFNAMSIZ)
1075 goto send_mal_query;
1076 memset(buff, 0, sizeof(buff));
1077 memcpy(buff, &iio->ident.name, ident_len);
1078 dev = dev_get_by_name(net, buff);
1079 break;
1080 case ICMP_EXT_ECHO_CTYPE_INDEX:
1081 if (ident_len != sizeof(iio->ident.ifindex))
1082 goto send_mal_query;
1083 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1084 break;
1085 case ICMP_EXT_ECHO_CTYPE_ADDR:
1086 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1087 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1088 iio->ident.addr.ctype3_hdr.addrlen)
1089 goto send_mal_query;
1090 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1091 case ICMP_AFI_IP:
1092 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1093 goto send_mal_query;
1094 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1095 break;
1096#if IS_ENABLED(CONFIG_IPV6)
1097 case ICMP_AFI_IP6:
1098 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1099 goto send_mal_query;
1100 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1101 dev_hold(dev);
1102 break;
1103#endif
1104 default:
1105 goto send_mal_query;
1106 }
1107 break;
1108 default:
1109 goto send_mal_query;
1110 }
1111 if (!dev) {
1112 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1113 return true;
1114 }
1115 /* Fill bits in reply message */
1116 if (dev->flags & IFF_UP)
1117 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1118 if (__in_dev_get_rcu(dev) && __in_dev_get_rcu(dev)->ifa_list)
1119 status |= ICMP_EXT_ECHOREPLY_IPV4;
1120 if (!list_empty(&rcu_dereference(dev->ip6_ptr)->addr_list))
1121 status |= ICMP_EXT_ECHOREPLY_IPV6;
1122 dev_put(dev);
1123 icmphdr->un.echo.sequence |= htons(status);
1124 return true;
1125send_mal_query:
1126 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1127 return true;
1128}
1129EXPORT_SYMBOL_GPL(icmp_build_probe);
1130
1131/*
1132 * Handle ICMP Timestamp requests.
1133 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1134 * SHOULD be in the kernel for minimum random latency.
1135 * MUST be accurate to a few minutes.
1136 * MUST be updated at least at 15Hz.
1137 */
1138static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1139{
1140 struct icmp_bxm icmp_param;
1141 /*
1142 * Too short.
1143 */
1144 if (skb->len < 4)
1145 goto out_err;
1146
1147 /*
1148 * Fill in the current time as ms since midnight UT:
1149 */
1150 icmp_param.data.times[1] = inet_current_timestamp();
1151 icmp_param.data.times[2] = icmp_param.data.times[1];
1152
1153 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1154
1155 icmp_param.data.icmph = *icmp_hdr(skb);
1156 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1157 icmp_param.data.icmph.code = 0;
1158 icmp_param.skb = skb;
1159 icmp_param.offset = 0;
1160 icmp_param.data_len = 0;
1161 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1162 icmp_reply(&icmp_param, skb);
1163 return SKB_NOT_DROPPED_YET;
1164
1165out_err:
1166 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1167 return SKB_DROP_REASON_PKT_TOO_SMALL;
1168}
1169
1170static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1171{
1172 /* pretend it was a success */
1173 return SKB_NOT_DROPPED_YET;
1174}
1175
1176/*
1177 * Deal with incoming ICMP packets.
1178 */
1179int icmp_rcv(struct sk_buff *skb)
1180{
1181 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1182 struct rtable *rt = skb_rtable(skb);
1183 struct net *net = dev_net(rt->dst.dev);
1184 struct icmphdr *icmph;
1185
1186 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1187 struct sec_path *sp = skb_sec_path(skb);
1188 int nh;
1189
1190 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1191 XFRM_STATE_ICMP)) {
1192 reason = SKB_DROP_REASON_XFRM_POLICY;
1193 goto drop;
1194 }
1195
1196 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1197 goto drop;
1198
1199 nh = skb_network_offset(skb);
1200 skb_set_network_header(skb, sizeof(*icmph));
1201
1202 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1203 skb)) {
1204 reason = SKB_DROP_REASON_XFRM_POLICY;
1205 goto drop;
1206 }
1207
1208 skb_set_network_header(skb, nh);
1209 }
1210
1211 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1212
1213 if (skb_checksum_simple_validate(skb))
1214 goto csum_error;
1215
1216 if (!pskb_pull(skb, sizeof(*icmph)))
1217 goto error;
1218
1219 icmph = icmp_hdr(skb);
1220
1221 ICMPMSGIN_INC_STATS(net, icmph->type);
1222
1223 /* Check for ICMP Extended Echo (PROBE) messages */
1224 if (icmph->type == ICMP_EXT_ECHO) {
1225 /* We can't use icmp_pointers[].handler() because it is an array of
1226 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1227 */
1228 reason = icmp_echo(skb);
1229 goto reason_check;
1230 }
1231
1232 if (icmph->type == ICMP_EXT_ECHOREPLY) {
1233 reason = ping_rcv(skb);
1234 goto reason_check;
1235 }
1236
1237 /*
1238 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1239 *
1240 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1241 * discarded.
1242 */
1243 if (icmph->type > NR_ICMP_TYPES) {
1244 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1245 goto error;
1246 }
1247
1248 /*
1249 * Parse the ICMP message
1250 */
1251
1252 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1253 /*
1254 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1255 * silently ignored (we let user decide with a sysctl).
1256 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1257 * discarded if to broadcast/multicast.
1258 */
1259 if ((icmph->type == ICMP_ECHO ||
1260 icmph->type == ICMP_TIMESTAMP) &&
1261 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1262 reason = SKB_DROP_REASON_INVALID_PROTO;
1263 goto error;
1264 }
1265 if (icmph->type != ICMP_ECHO &&
1266 icmph->type != ICMP_TIMESTAMP &&
1267 icmph->type != ICMP_ADDRESS &&
1268 icmph->type != ICMP_ADDRESSREPLY) {
1269 reason = SKB_DROP_REASON_INVALID_PROTO;
1270 goto error;
1271 }
1272 }
1273
1274 reason = icmp_pointers[icmph->type].handler(skb);
1275reason_check:
1276 if (!reason) {
1277 consume_skb(skb);
1278 return NET_RX_SUCCESS;
1279 }
1280
1281drop:
1282 kfree_skb_reason(skb, reason);
1283 return NET_RX_DROP;
1284csum_error:
1285 reason = SKB_DROP_REASON_ICMP_CSUM;
1286 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1287error:
1288 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1289 goto drop;
1290}
1291
1292static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1293{
1294 struct icmp_extobj_hdr *objh, _objh;
1295 struct icmp_ext_hdr *exth, _exth;
1296 u16 olen;
1297
1298 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1299 if (!exth)
1300 return false;
1301 if (exth->version != 2)
1302 return true;
1303
1304 if (exth->checksum &&
1305 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1306 return false;
1307
1308 off += sizeof(_exth);
1309 while (off < skb->len) {
1310 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1311 if (!objh)
1312 return false;
1313
1314 olen = ntohs(objh->length);
1315 if (olen < sizeof(_objh))
1316 return false;
1317
1318 off += olen;
1319 if (off > skb->len)
1320 return false;
1321 }
1322
1323 return true;
1324}
1325
1326void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1327 struct sock_ee_data_rfc4884 *out,
1328 int thlen, int off)
1329{
1330 int hlen;
1331
1332 /* original datagram headers: end of icmph to payload (skb->data) */
1333 hlen = -skb_transport_offset(skb) - thlen;
1334
1335 /* per rfc 4884: minimal datagram length of 128 bytes */
1336 if (off < 128 || off < hlen)
1337 return;
1338
1339 /* kernel has stripped headers: return payload offset in bytes */
1340 off -= hlen;
1341 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1342 return;
1343
1344 out->len = off;
1345
1346 if (!ip_icmp_error_rfc4884_validate(skb, off))
1347 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1348}
1349EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1350
1351int icmp_err(struct sk_buff *skb, u32 info)
1352{
1353 struct iphdr *iph = (struct iphdr *)skb->data;
1354 int offset = iph->ihl<<2;
1355 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1356 int type = icmp_hdr(skb)->type;
1357 int code = icmp_hdr(skb)->code;
1358 struct net *net = dev_net(skb->dev);
1359
1360 /*
1361 * Use ping_err to handle all icmp errors except those
1362 * triggered by ICMP_ECHOREPLY which sent from kernel.
1363 */
1364 if (icmph->type != ICMP_ECHOREPLY) {
1365 ping_err(skb, offset, info);
1366 return 0;
1367 }
1368
1369 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1370 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1371 else if (type == ICMP_REDIRECT)
1372 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1373
1374 return 0;
1375}
1376
1377/*
1378 * This table is the definition of how we handle ICMP.
1379 */
1380static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1381 [ICMP_ECHOREPLY] = {
1382 .handler = ping_rcv,
1383 },
1384 [1] = {
1385 .handler = icmp_discard,
1386 .error = 1,
1387 },
1388 [2] = {
1389 .handler = icmp_discard,
1390 .error = 1,
1391 },
1392 [ICMP_DEST_UNREACH] = {
1393 .handler = icmp_unreach,
1394 .error = 1,
1395 },
1396 [ICMP_SOURCE_QUENCH] = {
1397 .handler = icmp_unreach,
1398 .error = 1,
1399 },
1400 [ICMP_REDIRECT] = {
1401 .handler = icmp_redirect,
1402 .error = 1,
1403 },
1404 [6] = {
1405 .handler = icmp_discard,
1406 .error = 1,
1407 },
1408 [7] = {
1409 .handler = icmp_discard,
1410 .error = 1,
1411 },
1412 [ICMP_ECHO] = {
1413 .handler = icmp_echo,
1414 },
1415 [9] = {
1416 .handler = icmp_discard,
1417 .error = 1,
1418 },
1419 [10] = {
1420 .handler = icmp_discard,
1421 .error = 1,
1422 },
1423 [ICMP_TIME_EXCEEDED] = {
1424 .handler = icmp_unreach,
1425 .error = 1,
1426 },
1427 [ICMP_PARAMETERPROB] = {
1428 .handler = icmp_unreach,
1429 .error = 1,
1430 },
1431 [ICMP_TIMESTAMP] = {
1432 .handler = icmp_timestamp,
1433 },
1434 [ICMP_TIMESTAMPREPLY] = {
1435 .handler = icmp_discard,
1436 },
1437 [ICMP_INFO_REQUEST] = {
1438 .handler = icmp_discard,
1439 },
1440 [ICMP_INFO_REPLY] = {
1441 .handler = icmp_discard,
1442 },
1443 [ICMP_ADDRESS] = {
1444 .handler = icmp_discard,
1445 },
1446 [ICMP_ADDRESSREPLY] = {
1447 .handler = icmp_discard,
1448 },
1449};
1450
1451static int __net_init icmp_sk_init(struct net *net)
1452{
1453 /* Control parameters for ECHO replies. */
1454 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1455 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1456 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1457
1458 /* Control parameter - ignore bogus broadcast responses? */
1459 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1460
1461 /*
1462 * Configurable global rate limit.
1463 *
1464 * ratelimit defines tokens/packet consumed for dst->rate_token
1465 * bucket ratemask defines which icmp types are ratelimited by
1466 * setting it's bit position.
1467 *
1468 * default:
1469 * dest unreachable (3), source quench (4),
1470 * time exceeded (11), parameter problem (12)
1471 */
1472
1473 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1474 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1475 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1476
1477 return 0;
1478}
1479
1480static struct pernet_operations __net_initdata icmp_sk_ops = {
1481 .init = icmp_sk_init,
1482};
1483
1484int __init icmp_init(void)
1485{
1486 int err, i;
1487
1488 for_each_possible_cpu(i) {
1489 struct sock *sk;
1490
1491 err = inet_ctl_sock_create(&sk, PF_INET,
1492 SOCK_RAW, IPPROTO_ICMP, &init_net);
1493 if (err < 0)
1494 return err;
1495
1496 per_cpu(ipv4_icmp_sk, i) = sk;
1497
1498 /* Enough space for 2 64K ICMP packets, including
1499 * sk_buff/skb_shared_info struct overhead.
1500 */
1501 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1502
1503 /*
1504 * Speedup sock_wfree()
1505 */
1506 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1507 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1508 }
1509 return register_pernet_subsys(&icmp_sk_ops);
1510}