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