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
100 struct 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
116 const 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 };
182 EXPORT_SYMBOL(icmp_err_convert);
183
184 /*
185 * ICMP control array. This specifies what to do with each ICMP.
186 */
187
188 struct icmp_control {
189 bool (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191 };
192
193 static 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 */
icmp_sk(struct net * net)202 static struct sock *icmp_sk(struct net *net)
203 {
204 return this_cpu_read(*net->ipv4.icmp_sk);
205 }
206
207 /* Called with BH disabled */
icmp_xmit_lock(struct net * net)208 static 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
icmp_xmit_unlock(struct sock * sk)223 static inline void icmp_xmit_unlock(struct sock *sk)
224 {
225 spin_unlock(&sk->sk_lock.slock);
226 }
227
228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229 int sysctl_icmp_msgs_burst __read_mostly = 50;
230
231 static 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 */
icmp_global_allow(void)246 bool 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 = READ_ONCE(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,
269 READ_ONCE(sysctl_icmp_msgs_burst));
270 if (credit) {
271 /* We want to use a credit of one in average, but need to randomize
272 * it for security reasons.
273 */
274 credit = max_t(int, credit - prandom_u32_max(3), 0);
275 rc = true;
276 }
277 WRITE_ONCE(icmp_global.credit, credit);
278 spin_unlock(&icmp_global.lock);
279 return rc;
280 }
281 EXPORT_SYMBOL(icmp_global_allow);
282
icmpv4_mask_allow(struct net * net,int type,int code)283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
284 {
285 if (type > NR_ICMP_TYPES)
286 return true;
287
288 /* Don't limit PMTU discovery. */
289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
290 return true;
291
292 /* Limit if icmp type is enabled in ratemask. */
293 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
294 return true;
295
296 return false;
297 }
298
icmpv4_global_allow(struct net * net,int type,int code)299 static bool icmpv4_global_allow(struct net *net, int type, int code)
300 {
301 if (icmpv4_mask_allow(net, type, code))
302 return true;
303
304 if (icmp_global_allow())
305 return true;
306
307 return false;
308 }
309
310 /*
311 * Send an ICMP frame.
312 */
313
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code)314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 struct flowi4 *fl4, int type, int code)
316 {
317 struct dst_entry *dst = &rt->dst;
318 struct inet_peer *peer;
319 bool rc = true;
320 int vif;
321
322 if (icmpv4_mask_allow(net, type, code))
323 goto out;
324
325 /* No rate limit on loopback */
326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
327 goto out;
328
329 vif = l3mdev_master_ifindex(dst->dev);
330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 rc = inet_peer_xrlim_allow(peer,
332 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
333 if (peer)
334 inet_putpeer(peer);
335 out:
336 return rc;
337 }
338
339 /*
340 * Maintain the counters used in the SNMP statistics for outgoing ICMP
341 */
icmp_out_count(struct net * net,unsigned char type)342 void icmp_out_count(struct net *net, unsigned char type)
343 {
344 ICMPMSGOUT_INC_STATS(net, type);
345 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
346 }
347
348 /*
349 * Checksum each fragment, and on the first include the headers and final
350 * checksum.
351 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)352 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
353 struct sk_buff *skb)
354 {
355 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
356 __wsum csum;
357
358 csum = skb_copy_and_csum_bits(icmp_param->skb,
359 icmp_param->offset + offset,
360 to, len);
361
362 skb->csum = csum_block_add(skb->csum, csum, odd);
363 if (icmp_pointers[icmp_param->data.icmph.type].error)
364 nf_ct_attach(skb, icmp_param->skb);
365 return 0;
366 }
367
icmp_push_reply(struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)368 static void icmp_push_reply(struct icmp_bxm *icmp_param,
369 struct flowi4 *fl4,
370 struct ipcm_cookie *ipc, struct rtable **rt)
371 {
372 struct sock *sk;
373 struct sk_buff *skb;
374
375 sk = icmp_sk(dev_net((*rt)->dst.dev));
376 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
377 icmp_param->data_len+icmp_param->head_len,
378 icmp_param->head_len,
379 ipc, rt, MSG_DONTWAIT) < 0) {
380 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
381 ip_flush_pending_frames(sk);
382 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
383 struct icmphdr *icmph = icmp_hdr(skb);
384 __wsum csum;
385 struct sk_buff *skb1;
386
387 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
388 (char *)icmph,
389 icmp_param->head_len);
390 skb_queue_walk(&sk->sk_write_queue, skb1) {
391 csum = csum_add(csum, skb1->csum);
392 }
393 icmph->checksum = csum_fold(csum);
394 skb->ip_summed = CHECKSUM_NONE;
395 ip_push_pending_frames(sk, fl4);
396 }
397 }
398
399 /*
400 * Driving logic for building and sending ICMP messages.
401 */
402
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
404 {
405 struct ipcm_cookie ipc;
406 struct rtable *rt = skb_rtable(skb);
407 struct net *net = dev_net(rt->dst.dev);
408 struct flowi4 fl4;
409 struct sock *sk;
410 struct inet_sock *inet;
411 __be32 daddr, saddr;
412 u32 mark = IP4_REPLY_MARK(net, skb->mark);
413 int type = icmp_param->data.icmph.type;
414 int code = icmp_param->data.icmph.code;
415
416 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
417 return;
418
419 /* Needed by both icmp_global_allow and icmp_xmit_lock */
420 local_bh_disable();
421
422 /* global icmp_msgs_per_sec */
423 if (!icmpv4_global_allow(net, type, code))
424 goto out_bh_enable;
425
426 sk = icmp_xmit_lock(net);
427 if (!sk)
428 goto out_bh_enable;
429 inet = inet_sk(sk);
430
431 icmp_param->data.icmph.checksum = 0;
432
433 ipcm_init(&ipc);
434 inet->tos = ip_hdr(skb)->tos;
435 ipc.sockc.mark = mark;
436 daddr = ipc.addr = ip_hdr(skb)->saddr;
437 saddr = fib_compute_spec_dst(skb);
438
439 if (icmp_param->replyopts.opt.opt.optlen) {
440 ipc.opt = &icmp_param->replyopts.opt;
441 if (ipc.opt->opt.srr)
442 daddr = icmp_param->replyopts.opt.opt.faddr;
443 }
444 memset(&fl4, 0, sizeof(fl4));
445 fl4.daddr = daddr;
446 fl4.saddr = saddr;
447 fl4.flowi4_mark = mark;
448 fl4.flowi4_uid = sock_net_uid(net, NULL);
449 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
450 fl4.flowi4_proto = IPPROTO_ICMP;
451 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
452 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
453 rt = ip_route_output_key(net, &fl4);
454 if (IS_ERR(rt))
455 goto out_unlock;
456 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
457 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
458 ip_rt_put(rt);
459 out_unlock:
460 icmp_xmit_unlock(sk);
461 out_bh_enable:
462 local_bh_enable();
463 }
464
465 /*
466 * The device used for looking up which routing table to use for sending an ICMP
467 * error is preferably the source whenever it is set, which should ensure the
468 * icmp error can be sent to the source host, else lookup using the routing
469 * table of the destination device, else use the main routing table (index 0).
470 */
icmp_get_route_lookup_dev(struct sk_buff * skb)471 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
472 {
473 struct net_device *route_lookup_dev = NULL;
474
475 if (skb->dev)
476 route_lookup_dev = skb->dev;
477 else if (skb_dst(skb))
478 route_lookup_dev = skb_dst(skb)->dev;
479 return route_lookup_dev;
480 }
481
icmp_route_lookup(struct net * net,struct flowi4 * fl4,struct sk_buff * skb_in,const struct iphdr * iph,__be32 saddr,u8 tos,u32 mark,int type,int code,struct icmp_bxm * param)482 static struct rtable *icmp_route_lookup(struct net *net,
483 struct flowi4 *fl4,
484 struct sk_buff *skb_in,
485 const struct iphdr *iph,
486 __be32 saddr, u8 tos, u32 mark,
487 int type, int code,
488 struct icmp_bxm *param)
489 {
490 struct net_device *route_lookup_dev;
491 struct rtable *rt, *rt2;
492 struct flowi4 fl4_dec;
493 int err;
494
495 memset(fl4, 0, sizeof(*fl4));
496 fl4->daddr = (param->replyopts.opt.opt.srr ?
497 param->replyopts.opt.opt.faddr : iph->saddr);
498 fl4->saddr = saddr;
499 fl4->flowi4_mark = mark;
500 fl4->flowi4_uid = sock_net_uid(net, NULL);
501 fl4->flowi4_tos = RT_TOS(tos);
502 fl4->flowi4_proto = IPPROTO_ICMP;
503 fl4->fl4_icmp_type = type;
504 fl4->fl4_icmp_code = code;
505 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
506 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
507
508 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
509 rt = ip_route_output_key_hash(net, fl4, skb_in);
510 if (IS_ERR(rt))
511 return rt;
512
513 /* No need to clone since we're just using its address. */
514 rt2 = rt;
515
516 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
517 flowi4_to_flowi(fl4), NULL, 0);
518 if (!IS_ERR(rt)) {
519 if (rt != rt2)
520 return rt;
521 if (inet_addr_type_dev_table(net, route_lookup_dev,
522 fl4->daddr) == RTN_LOCAL)
523 return rt;
524 } else if (PTR_ERR(rt) == -EPERM) {
525 rt = NULL;
526 } else
527 return rt;
528
529 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
530 if (err)
531 goto relookup_failed;
532
533 if (inet_addr_type_dev_table(net, route_lookup_dev,
534 fl4_dec.saddr) == RTN_LOCAL) {
535 rt2 = __ip_route_output_key(net, &fl4_dec);
536 if (IS_ERR(rt2))
537 err = PTR_ERR(rt2);
538 } else {
539 struct flowi4 fl4_2 = {};
540 unsigned long orefdst;
541
542 fl4_2.daddr = fl4_dec.saddr;
543 rt2 = ip_route_output_key(net, &fl4_2);
544 if (IS_ERR(rt2)) {
545 err = PTR_ERR(rt2);
546 goto relookup_failed;
547 }
548 /* Ugh! */
549 orefdst = skb_in->_skb_refdst; /* save old refdst */
550 skb_dst_set(skb_in, NULL);
551 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
552 RT_TOS(tos), rt2->dst.dev);
553
554 dst_release(&rt2->dst);
555 rt2 = skb_rtable(skb_in);
556 skb_in->_skb_refdst = orefdst; /* restore old refdst */
557 }
558
559 if (err)
560 goto relookup_failed;
561
562 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
563 flowi4_to_flowi(&fl4_dec), NULL,
564 XFRM_LOOKUP_ICMP);
565 if (!IS_ERR(rt2)) {
566 dst_release(&rt->dst);
567 memcpy(fl4, &fl4_dec, sizeof(*fl4));
568 rt = rt2;
569 } else if (PTR_ERR(rt2) == -EPERM) {
570 if (rt)
571 dst_release(&rt->dst);
572 return rt2;
573 } else {
574 err = PTR_ERR(rt2);
575 goto relookup_failed;
576 }
577 return rt;
578
579 relookup_failed:
580 if (rt)
581 return rt;
582 return ERR_PTR(err);
583 }
584
585 /*
586 * Send an ICMP message in response to a situation
587 *
588 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
589 * MAY send more (we do).
590 * MUST NOT change this header information.
591 * MUST NOT reply to a multicast/broadcast IP address.
592 * MUST NOT reply to a multicast/broadcast MAC address.
593 * MUST reply to only the first fragment.
594 */
595
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct ip_options * opt)596 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
597 const struct ip_options *opt)
598 {
599 struct iphdr *iph;
600 int room;
601 struct icmp_bxm icmp_param;
602 struct rtable *rt = skb_rtable(skb_in);
603 struct ipcm_cookie ipc;
604 struct flowi4 fl4;
605 __be32 saddr;
606 u8 tos;
607 u32 mark;
608 struct net *net;
609 struct sock *sk;
610
611 if (!rt)
612 goto out;
613
614 if (rt->dst.dev)
615 net = dev_net(rt->dst.dev);
616 else if (skb_in->dev)
617 net = dev_net(skb_in->dev);
618 else
619 goto out;
620
621 /*
622 * Find the original header. It is expected to be valid, of course.
623 * Check this, icmp_send is called from the most obscure devices
624 * sometimes.
625 */
626 iph = ip_hdr(skb_in);
627
628 if ((u8 *)iph < skb_in->head ||
629 (skb_network_header(skb_in) + sizeof(*iph)) >
630 skb_tail_pointer(skb_in))
631 goto out;
632
633 /*
634 * No replies to physical multicast/broadcast
635 */
636 if (skb_in->pkt_type != PACKET_HOST)
637 goto out;
638
639 /*
640 * Now check at the protocol level
641 */
642 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
643 goto out;
644
645 /*
646 * Only reply to fragment 0. We byte re-order the constant
647 * mask for efficiency.
648 */
649 if (iph->frag_off & htons(IP_OFFSET))
650 goto out;
651
652 /*
653 * If we send an ICMP error to an ICMP error a mess would result..
654 */
655 if (icmp_pointers[type].error) {
656 /*
657 * We are an error, check if we are replying to an
658 * ICMP error
659 */
660 if (iph->protocol == IPPROTO_ICMP) {
661 u8 _inner_type, *itp;
662
663 itp = skb_header_pointer(skb_in,
664 skb_network_header(skb_in) +
665 (iph->ihl << 2) +
666 offsetof(struct icmphdr,
667 type) -
668 skb_in->data,
669 sizeof(_inner_type),
670 &_inner_type);
671 if (!itp)
672 goto out;
673
674 /*
675 * Assume any unknown ICMP type is an error. This
676 * isn't specified by the RFC, but think about it..
677 */
678 if (*itp > NR_ICMP_TYPES ||
679 icmp_pointers[*itp].error)
680 goto out;
681 }
682 }
683
684 /* Needed by both icmp_global_allow and icmp_xmit_lock */
685 local_bh_disable();
686
687 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
688 * incoming dev is loopback. If outgoing dev change to not be
689 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
690 */
691 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
692 !icmpv4_global_allow(net, type, code))
693 goto out_bh_enable;
694
695 sk = icmp_xmit_lock(net);
696 if (!sk)
697 goto out_bh_enable;
698
699 /*
700 * Construct source address and options.
701 */
702
703 saddr = iph->daddr;
704 if (!(rt->rt_flags & RTCF_LOCAL)) {
705 struct net_device *dev = NULL;
706
707 rcu_read_lock();
708 if (rt_is_input_route(rt) &&
709 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
710 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
711
712 if (dev)
713 saddr = inet_select_addr(dev, iph->saddr,
714 RT_SCOPE_LINK);
715 else
716 saddr = 0;
717 rcu_read_unlock();
718 }
719
720 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
721 IPTOS_PREC_INTERNETCONTROL) :
722 iph->tos;
723 mark = IP4_REPLY_MARK(net, skb_in->mark);
724
725 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
726 goto out_unlock;
727
728
729 /*
730 * Prepare data for ICMP header.
731 */
732
733 icmp_param.data.icmph.type = type;
734 icmp_param.data.icmph.code = code;
735 icmp_param.data.icmph.un.gateway = info;
736 icmp_param.data.icmph.checksum = 0;
737 icmp_param.skb = skb_in;
738 icmp_param.offset = skb_network_offset(skb_in);
739 inet_sk(sk)->tos = tos;
740 ipcm_init(&ipc);
741 ipc.addr = iph->saddr;
742 ipc.opt = &icmp_param.replyopts.opt;
743 ipc.sockc.mark = mark;
744
745 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
746 type, code, &icmp_param);
747 if (IS_ERR(rt))
748 goto out_unlock;
749
750 /* peer icmp_ratelimit */
751 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
752 goto ende;
753
754 /* RFC says return as much as we can without exceeding 576 bytes. */
755
756 room = dst_mtu(&rt->dst);
757 if (room > 576)
758 room = 576;
759 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
760 room -= sizeof(struct icmphdr);
761 /* Guard against tiny mtu. We need to include at least one
762 * IP network header for this message to make any sense.
763 */
764 if (room <= (int)sizeof(struct iphdr))
765 goto ende;
766
767 icmp_param.data_len = skb_in->len - icmp_param.offset;
768 if (icmp_param.data_len > room)
769 icmp_param.data_len = room;
770 icmp_param.head_len = sizeof(struct icmphdr);
771
772 /* if we don't have a source address at this point, fall back to the
773 * dummy address instead of sending out a packet with a source address
774 * of 0.0.0.0
775 */
776 if (!fl4.saddr)
777 fl4.saddr = htonl(INADDR_DUMMY);
778
779 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
780 ende:
781 ip_rt_put(rt);
782 out_unlock:
783 icmp_xmit_unlock(sk);
784 out_bh_enable:
785 local_bh_enable();
786 out:;
787 }
788 EXPORT_SYMBOL(__icmp_send);
789
790 #if IS_ENABLED(CONFIG_NF_NAT)
791 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)792 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
793 {
794 struct sk_buff *cloned_skb = NULL;
795 struct ip_options opts = { 0 };
796 enum ip_conntrack_info ctinfo;
797 struct nf_conn *ct;
798 __be32 orig_ip;
799
800 ct = nf_ct_get(skb_in, &ctinfo);
801 if (!ct || !(ct->status & IPS_SRC_NAT)) {
802 __icmp_send(skb_in, type, code, info, &opts);
803 return;
804 }
805
806 if (skb_shared(skb_in))
807 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
808
809 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
810 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
811 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
812 skb_network_offset(skb_in) + sizeof(struct iphdr))))
813 goto out;
814
815 orig_ip = ip_hdr(skb_in)->saddr;
816 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
817 __icmp_send(skb_in, type, code, info, &opts);
818 ip_hdr(skb_in)->saddr = orig_ip;
819 out:
820 consume_skb(cloned_skb);
821 }
822 EXPORT_SYMBOL(icmp_ndo_send);
823 #endif
824
icmp_socket_deliver(struct sk_buff * skb,u32 info)825 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
826 {
827 const struct iphdr *iph = (const struct iphdr *)skb->data;
828 const struct net_protocol *ipprot;
829 int protocol = iph->protocol;
830
831 /* Checkin full IP header plus 8 bytes of protocol to
832 * avoid additional coding at protocol handlers.
833 */
834 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
835 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
836 return;
837 }
838
839 raw_icmp_error(skb, protocol, info);
840
841 ipprot = rcu_dereference(inet_protos[protocol]);
842 if (ipprot && ipprot->err_handler)
843 ipprot->err_handler(skb, info);
844 }
845
icmp_tag_validation(int proto)846 static bool icmp_tag_validation(int proto)
847 {
848 bool ok;
849
850 rcu_read_lock();
851 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
852 rcu_read_unlock();
853 return ok;
854 }
855
856 /*
857 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
858 * ICMP_PARAMETERPROB.
859 */
860
icmp_unreach(struct sk_buff * skb)861 static bool icmp_unreach(struct sk_buff *skb)
862 {
863 const struct iphdr *iph;
864 struct icmphdr *icmph;
865 struct net *net;
866 u32 info = 0;
867
868 net = dev_net(skb_dst(skb)->dev);
869
870 /*
871 * Incomplete header ?
872 * Only checks for the IP header, there should be an
873 * additional check for longer headers in upper levels.
874 */
875
876 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
877 goto out_err;
878
879 icmph = icmp_hdr(skb);
880 iph = (const struct iphdr *)skb->data;
881
882 if (iph->ihl < 5) /* Mangled header, drop. */
883 goto out_err;
884
885 switch (icmph->type) {
886 case ICMP_DEST_UNREACH:
887 switch (icmph->code & 15) {
888 case ICMP_NET_UNREACH:
889 case ICMP_HOST_UNREACH:
890 case ICMP_PROT_UNREACH:
891 case ICMP_PORT_UNREACH:
892 break;
893 case ICMP_FRAG_NEEDED:
894 /* for documentation of the ip_no_pmtu_disc
895 * values please see
896 * Documentation/networking/ip-sysctl.rst
897 */
898 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
899 default:
900 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
901 &iph->daddr);
902 break;
903 case 2:
904 goto out;
905 case 3:
906 if (!icmp_tag_validation(iph->protocol))
907 goto out;
908 fallthrough;
909 case 0:
910 info = ntohs(icmph->un.frag.mtu);
911 }
912 break;
913 case ICMP_SR_FAILED:
914 net_dbg_ratelimited("%pI4: Source Route Failed\n",
915 &iph->daddr);
916 break;
917 default:
918 break;
919 }
920 if (icmph->code > NR_ICMP_UNREACH)
921 goto out;
922 break;
923 case ICMP_PARAMETERPROB:
924 info = ntohl(icmph->un.gateway) >> 24;
925 break;
926 case ICMP_TIME_EXCEEDED:
927 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
928 if (icmph->code == ICMP_EXC_FRAGTIME)
929 goto out;
930 break;
931 }
932
933 /*
934 * Throw it at our lower layers
935 *
936 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
937 * header.
938 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
939 * transport layer.
940 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
941 * transport layer.
942 */
943
944 /*
945 * Check the other end isn't violating RFC 1122. Some routers send
946 * bogus responses to broadcast frames. If you see this message
947 * first check your netmask matches at both ends, if it does then
948 * get the other vendor to fix their kit.
949 */
950
951 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
952 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
953 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
954 &ip_hdr(skb)->saddr,
955 icmph->type, icmph->code,
956 &iph->daddr, skb->dev->name);
957 goto out;
958 }
959
960 icmp_socket_deliver(skb, info);
961
962 out:
963 return true;
964 out_err:
965 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
966 return false;
967 }
968
969
970 /*
971 * Handle ICMP_REDIRECT.
972 */
973
icmp_redirect(struct sk_buff * skb)974 static bool icmp_redirect(struct sk_buff *skb)
975 {
976 if (skb->len < sizeof(struct iphdr)) {
977 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
978 return false;
979 }
980
981 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
982 /* there aught to be a stat */
983 return false;
984 }
985
986 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
987 return true;
988 }
989
990 /*
991 * Handle ICMP_ECHO ("ping") requests.
992 *
993 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
994 * requests.
995 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
996 * included in the reply.
997 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
998 * echo requests, MUST have default=NOT.
999 * See also WRT handling of options once they are done and working.
1000 */
1001
icmp_echo(struct sk_buff * skb)1002 static bool icmp_echo(struct sk_buff *skb)
1003 {
1004 struct net *net;
1005
1006 net = dev_net(skb_dst(skb)->dev);
1007 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
1008 struct icmp_bxm icmp_param;
1009
1010 icmp_param.data.icmph = *icmp_hdr(skb);
1011 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1012 icmp_param.skb = skb;
1013 icmp_param.offset = 0;
1014 icmp_param.data_len = skb->len;
1015 icmp_param.head_len = sizeof(struct icmphdr);
1016 icmp_reply(&icmp_param, skb);
1017 }
1018 /* should there be an ICMP stat for ignored echos? */
1019 return true;
1020 }
1021
1022 /*
1023 * Handle ICMP Timestamp requests.
1024 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1025 * SHOULD be in the kernel for minimum random latency.
1026 * MUST be accurate to a few minutes.
1027 * MUST be updated at least at 15Hz.
1028 */
icmp_timestamp(struct sk_buff * skb)1029 static bool icmp_timestamp(struct sk_buff *skb)
1030 {
1031 struct icmp_bxm icmp_param;
1032 /*
1033 * Too short.
1034 */
1035 if (skb->len < 4)
1036 goto out_err;
1037
1038 /*
1039 * Fill in the current time as ms since midnight UT:
1040 */
1041 icmp_param.data.times[1] = inet_current_timestamp();
1042 icmp_param.data.times[2] = icmp_param.data.times[1];
1043
1044 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1045
1046 icmp_param.data.icmph = *icmp_hdr(skb);
1047 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1048 icmp_param.data.icmph.code = 0;
1049 icmp_param.skb = skb;
1050 icmp_param.offset = 0;
1051 icmp_param.data_len = 0;
1052 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1053 icmp_reply(&icmp_param, skb);
1054 return true;
1055
1056 out_err:
1057 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1058 return false;
1059 }
1060
icmp_discard(struct sk_buff * skb)1061 static bool icmp_discard(struct sk_buff *skb)
1062 {
1063 /* pretend it was a success */
1064 return true;
1065 }
1066
1067 /*
1068 * Deal with incoming ICMP packets.
1069 */
icmp_rcv(struct sk_buff * skb)1070 int icmp_rcv(struct sk_buff *skb)
1071 {
1072 struct icmphdr *icmph;
1073 struct rtable *rt = skb_rtable(skb);
1074 struct net *net = dev_net(rt->dst.dev);
1075 bool success;
1076
1077 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1078 struct sec_path *sp = skb_sec_path(skb);
1079 int nh;
1080
1081 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1082 XFRM_STATE_ICMP))
1083 goto drop;
1084
1085 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1086 goto drop;
1087
1088 nh = skb_network_offset(skb);
1089 skb_set_network_header(skb, sizeof(*icmph));
1090
1091 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1092 goto drop;
1093
1094 skb_set_network_header(skb, nh);
1095 }
1096
1097 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1098
1099 if (skb_checksum_simple_validate(skb))
1100 goto csum_error;
1101
1102 if (!pskb_pull(skb, sizeof(*icmph)))
1103 goto error;
1104
1105 icmph = icmp_hdr(skb);
1106
1107 ICMPMSGIN_INC_STATS(net, icmph->type);
1108 /*
1109 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1110 *
1111 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1112 * discarded.
1113 */
1114 if (icmph->type > NR_ICMP_TYPES)
1115 goto error;
1116
1117
1118 /*
1119 * Parse the ICMP message
1120 */
1121
1122 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1123 /*
1124 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1125 * silently ignored (we let user decide with a sysctl).
1126 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1127 * discarded if to broadcast/multicast.
1128 */
1129 if ((icmph->type == ICMP_ECHO ||
1130 icmph->type == ICMP_TIMESTAMP) &&
1131 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1132 goto error;
1133 }
1134 if (icmph->type != ICMP_ECHO &&
1135 icmph->type != ICMP_TIMESTAMP &&
1136 icmph->type != ICMP_ADDRESS &&
1137 icmph->type != ICMP_ADDRESSREPLY) {
1138 goto error;
1139 }
1140 }
1141
1142 success = icmp_pointers[icmph->type].handler(skb);
1143
1144 if (success) {
1145 consume_skb(skb);
1146 return NET_RX_SUCCESS;
1147 }
1148
1149 drop:
1150 kfree_skb(skb);
1151 return NET_RX_DROP;
1152 csum_error:
1153 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1154 error:
1155 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1156 goto drop;
1157 }
1158
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1159 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1160 {
1161 struct icmp_extobj_hdr *objh, _objh;
1162 struct icmp_ext_hdr *exth, _exth;
1163 u16 olen;
1164
1165 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1166 if (!exth)
1167 return false;
1168 if (exth->version != 2)
1169 return true;
1170
1171 if (exth->checksum &&
1172 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1173 return false;
1174
1175 off += sizeof(_exth);
1176 while (off < skb->len) {
1177 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1178 if (!objh)
1179 return false;
1180
1181 olen = ntohs(objh->length);
1182 if (olen < sizeof(_objh))
1183 return false;
1184
1185 off += olen;
1186 if (off > skb->len)
1187 return false;
1188 }
1189
1190 return true;
1191 }
1192
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1193 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1194 struct sock_ee_data_rfc4884 *out,
1195 int thlen, int off)
1196 {
1197 int hlen;
1198
1199 /* original datagram headers: end of icmph to payload (skb->data) */
1200 hlen = -skb_transport_offset(skb) - thlen;
1201
1202 /* per rfc 4884: minimal datagram length of 128 bytes */
1203 if (off < 128 || off < hlen)
1204 return;
1205
1206 /* kernel has stripped headers: return payload offset in bytes */
1207 off -= hlen;
1208 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1209 return;
1210
1211 out->len = off;
1212
1213 if (!ip_icmp_error_rfc4884_validate(skb, off))
1214 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1215 }
1216 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1217
icmp_err(struct sk_buff * skb,u32 info)1218 int icmp_err(struct sk_buff *skb, u32 info)
1219 {
1220 struct iphdr *iph = (struct iphdr *)skb->data;
1221 int offset = iph->ihl<<2;
1222 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1223 int type = icmp_hdr(skb)->type;
1224 int code = icmp_hdr(skb)->code;
1225 struct net *net = dev_net(skb->dev);
1226
1227 /*
1228 * Use ping_err to handle all icmp errors except those
1229 * triggered by ICMP_ECHOREPLY which sent from kernel.
1230 */
1231 if (icmph->type != ICMP_ECHOREPLY) {
1232 ping_err(skb, offset, info);
1233 return 0;
1234 }
1235
1236 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1237 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1238 else if (type == ICMP_REDIRECT)
1239 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1240
1241 return 0;
1242 }
1243
1244 /*
1245 * This table is the definition of how we handle ICMP.
1246 */
1247 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1248 [ICMP_ECHOREPLY] = {
1249 .handler = ping_rcv,
1250 },
1251 [1] = {
1252 .handler = icmp_discard,
1253 .error = 1,
1254 },
1255 [2] = {
1256 .handler = icmp_discard,
1257 .error = 1,
1258 },
1259 [ICMP_DEST_UNREACH] = {
1260 .handler = icmp_unreach,
1261 .error = 1,
1262 },
1263 [ICMP_SOURCE_QUENCH] = {
1264 .handler = icmp_unreach,
1265 .error = 1,
1266 },
1267 [ICMP_REDIRECT] = {
1268 .handler = icmp_redirect,
1269 .error = 1,
1270 },
1271 [6] = {
1272 .handler = icmp_discard,
1273 .error = 1,
1274 },
1275 [7] = {
1276 .handler = icmp_discard,
1277 .error = 1,
1278 },
1279 [ICMP_ECHO] = {
1280 .handler = icmp_echo,
1281 },
1282 [9] = {
1283 .handler = icmp_discard,
1284 .error = 1,
1285 },
1286 [10] = {
1287 .handler = icmp_discard,
1288 .error = 1,
1289 },
1290 [ICMP_TIME_EXCEEDED] = {
1291 .handler = icmp_unreach,
1292 .error = 1,
1293 },
1294 [ICMP_PARAMETERPROB] = {
1295 .handler = icmp_unreach,
1296 .error = 1,
1297 },
1298 [ICMP_TIMESTAMP] = {
1299 .handler = icmp_timestamp,
1300 },
1301 [ICMP_TIMESTAMPREPLY] = {
1302 .handler = icmp_discard,
1303 },
1304 [ICMP_INFO_REQUEST] = {
1305 .handler = icmp_discard,
1306 },
1307 [ICMP_INFO_REPLY] = {
1308 .handler = icmp_discard,
1309 },
1310 [ICMP_ADDRESS] = {
1311 .handler = icmp_discard,
1312 },
1313 [ICMP_ADDRESSREPLY] = {
1314 .handler = icmp_discard,
1315 },
1316 };
1317
icmp_sk_exit(struct net * net)1318 static void __net_exit icmp_sk_exit(struct net *net)
1319 {
1320 int i;
1321
1322 for_each_possible_cpu(i)
1323 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1324 free_percpu(net->ipv4.icmp_sk);
1325 net->ipv4.icmp_sk = NULL;
1326 }
1327
icmp_sk_init(struct net * net)1328 static int __net_init icmp_sk_init(struct net *net)
1329 {
1330 int i, err;
1331
1332 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1333 if (!net->ipv4.icmp_sk)
1334 return -ENOMEM;
1335
1336 for_each_possible_cpu(i) {
1337 struct sock *sk;
1338
1339 err = inet_ctl_sock_create(&sk, PF_INET,
1340 SOCK_RAW, IPPROTO_ICMP, net);
1341 if (err < 0)
1342 goto fail;
1343
1344 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1345
1346 /* Enough space for 2 64K ICMP packets, including
1347 * sk_buff/skb_shared_info struct overhead.
1348 */
1349 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1350
1351 /*
1352 * Speedup sock_wfree()
1353 */
1354 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1355 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1356 }
1357
1358 /* Control parameters for ECHO replies. */
1359 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1360 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1361
1362 /* Control parameter - ignore bogus broadcast responses? */
1363 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1364
1365 /*
1366 * Configurable global rate limit.
1367 *
1368 * ratelimit defines tokens/packet consumed for dst->rate_token
1369 * bucket ratemask defines which icmp types are ratelimited by
1370 * setting it's bit position.
1371 *
1372 * default:
1373 * dest unreachable (3), source quench (4),
1374 * time exceeded (11), parameter problem (12)
1375 */
1376
1377 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1378 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1379 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1380
1381 return 0;
1382
1383 fail:
1384 icmp_sk_exit(net);
1385 return err;
1386 }
1387
1388 static struct pernet_operations __net_initdata icmp_sk_ops = {
1389 .init = icmp_sk_init,
1390 .exit = icmp_sk_exit,
1391 };
1392
icmp_init(void)1393 int __init icmp_init(void)
1394 {
1395 return register_pernet_subsys(&icmp_sk_ops);
1396 }
1397