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/flow.h>
76 #include <net/snmp.h>
77 #include <net/ip.h>
78 #include <net/route.h>
79 #include <net/protocol.h>
80 #include <net/icmp.h>
81 #include <net/tcp.h>
82 #include <net/udp.h>
83 #include <net/raw.h>
84 #include <net/ping.h>
85 #include <linux/skbuff.h>
86 #include <net/sock.h>
87 #include <linux/errno.h>
88 #include <linux/timer.h>
89 #include <linux/init.h>
90 #include <linux/uaccess.h>
91 #include <net/checksum.h>
92 #include <net/xfrm.h>
93 #include <net/inet_common.h>
94 #include <net/ip_fib.h>
95 #include <net/l3mdev.h>
96 #include <net/addrconf.h>
97 #include <net/inet_dscp.h>
98 #define CREATE_TRACE_POINTS
99 #include <trace/events/icmp.h>
100
101 /*
102 * Build xmit assembly blocks
103 */
104
105 struct icmp_bxm {
106 struct sk_buff *skb;
107 int offset;
108 int data_len;
109
110 struct {
111 struct icmphdr icmph;
112 __be32 times[3];
113 } data;
114 int head_len;
115 struct ip_options_data replyopts;
116 };
117
118 /* An array of errno for error messages from dest unreach. */
119 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
120
121 const struct icmp_err icmp_err_convert[] = {
122 {
123 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
124 .fatal = 0,
125 },
126 {
127 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
128 .fatal = 0,
129 },
130 {
131 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
132 .fatal = 1,
133 },
134 {
135 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
136 .fatal = 1,
137 },
138 {
139 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
140 .fatal = 0,
141 },
142 {
143 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
144 .fatal = 0,
145 },
146 {
147 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
148 .fatal = 1,
149 },
150 {
151 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
152 .fatal = 1,
153 },
154 {
155 .errno = ENONET, /* ICMP_HOST_ISOLATED */
156 .fatal = 1,
157 },
158 {
159 .errno = ENETUNREACH, /* ICMP_NET_ANO */
160 .fatal = 1,
161 },
162 {
163 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
164 .fatal = 1,
165 },
166 {
167 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
168 .fatal = 0,
169 },
170 {
171 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
172 .fatal = 0,
173 },
174 {
175 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
176 .fatal = 1,
177 },
178 {
179 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
180 .fatal = 1,
181 },
182 {
183 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
184 .fatal = 1,
185 },
186 };
187 EXPORT_SYMBOL(icmp_err_convert);
188
189 /*
190 * ICMP control array. This specifies what to do with each ICMP.
191 */
192
193 struct icmp_control {
194 enum skb_drop_reason (*handler)(struct sk_buff *skb);
195 short error; /* This ICMP is classed as an error message */
196 };
197
198 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
199
200 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
201
202 /* Called with BH disabled */
icmp_xmit_lock(struct net * net)203 static inline struct sock *icmp_xmit_lock(struct net *net)
204 {
205 struct sock *sk;
206
207 sk = this_cpu_read(ipv4_icmp_sk);
208
209 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
210 /* This can happen if the output path signals a
211 * dst_link_failure() for an outgoing ICMP packet.
212 */
213 return NULL;
214 }
215 sock_net_set(sk, net);
216 return sk;
217 }
218
icmp_xmit_unlock(struct sock * sk)219 static inline void icmp_xmit_unlock(struct sock *sk)
220 {
221 sock_net_set(sk, &init_net);
222 spin_unlock(&sk->sk_lock.slock);
223 }
224
225 /**
226 * icmp_global_allow - Are we allowed to send one more ICMP message ?
227 * @net: network namespace
228 *
229 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
230 * Returns false if we reached the limit and can not send another packet.
231 * Works in tandem with icmp_global_consume().
232 */
icmp_global_allow(struct net * net)233 bool icmp_global_allow(struct net *net)
234 {
235 u32 delta, now, oldstamp;
236 int incr, new, old;
237
238 /* Note: many cpus could find this condition true.
239 * Then later icmp_global_consume() could consume more credits,
240 * this is an acceptable race.
241 */
242 if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
243 return true;
244
245 now = jiffies;
246 oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
247 delta = min_t(u32, now - oldstamp, HZ);
248 if (delta < HZ / 50)
249 return false;
250
251 incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec) * delta / HZ;
252 if (!incr)
253 return false;
254
255 if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
256 old = atomic_read(&net->ipv4.icmp_global_credit);
257 do {
258 new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
259 } while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
260 }
261 return true;
262 }
263 EXPORT_SYMBOL(icmp_global_allow);
264
icmp_global_consume(struct net * net)265 void icmp_global_consume(struct net *net)
266 {
267 int credits = get_random_u32_below(3);
268
269 /* Note: this might make icmp_global.credit negative. */
270 if (credits)
271 atomic_sub(credits, &net->ipv4.icmp_global_credit);
272 }
273 EXPORT_SYMBOL(icmp_global_consume);
274
icmpv4_mask_allow(struct net * net,int type,int code)275 static 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
icmpv4_global_allow(struct net * net,int type,int code,bool * apply_ratelimit)291 static bool icmpv4_global_allow(struct net *net, int type, int code,
292 bool *apply_ratelimit)
293 {
294 if (icmpv4_mask_allow(net, type, code))
295 return true;
296
297 if (icmp_global_allow(net)) {
298 *apply_ratelimit = true;
299 return true;
300 }
301 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
302 return false;
303 }
304
305 /*
306 * Send an ICMP frame.
307 */
308
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code,bool apply_ratelimit)309 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
310 struct flowi4 *fl4, int type, int code,
311 bool apply_ratelimit)
312 {
313 struct dst_entry *dst = &rt->dst;
314 struct inet_peer *peer;
315 struct net_device *dev;
316 bool rc = true;
317
318 if (!apply_ratelimit)
319 return true;
320
321 /* No rate limit on loopback */
322 rcu_read_lock();
323 dev = dst_dev_rcu(dst);
324 if (dev && (dev->flags & IFF_LOOPBACK))
325 goto out;
326
327 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr,
328 l3mdev_master_ifindex_rcu(dev));
329 rc = inet_peer_xrlim_allow(peer,
330 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
331 out:
332 rcu_read_unlock();
333 if (!rc)
334 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
335 else
336 icmp_global_consume(net);
337 return rc;
338 }
339
340 /*
341 * Maintain the counters used in the SNMP statistics for outgoing ICMP
342 */
icmp_out_count(struct net * net,unsigned char type)343 void icmp_out_count(struct net *net, unsigned char type)
344 {
345 ICMPMSGOUT_INC_STATS(net, type);
346 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
347 }
348
349 /*
350 * Checksum each fragment, and on the first include the headers and final
351 * checksum.
352 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)353 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
354 struct sk_buff *skb)
355 {
356 struct icmp_bxm *icmp_param = from;
357 __wsum csum;
358
359 csum = skb_copy_and_csum_bits(icmp_param->skb,
360 icmp_param->offset + offset,
361 to, len);
362
363 skb->csum = csum_block_add(skb->csum, csum, odd);
364 if (icmp_pointers[icmp_param->data.icmph.type].error)
365 nf_ct_attach(skb, icmp_param->skb);
366 return 0;
367 }
368
icmp_push_reply(struct sock * sk,struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)369 static void icmp_push_reply(struct sock *sk,
370 struct icmp_bxm *icmp_param,
371 struct flowi4 *fl4,
372 struct ipcm_cookie *ipc, struct rtable **rt)
373 {
374 struct sk_buff *skb;
375
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 rtable *rt = skb_rtable(skb);
406 struct net *net = dev_net_rcu(rt->dst.dev);
407 bool apply_ratelimit = false;
408 struct ipcm_cookie ipc;
409 struct flowi4 fl4;
410 struct sock *sk;
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 icmpv4_global_allow and icmp_xmit_lock */
420 local_bh_disable();
421
422 /* is global icmp_msgs_per_sec exhausted ? */
423 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
424 goto out_bh_enable;
425
426 sk = icmp_xmit_lock(net);
427 if (!sk)
428 goto out_bh_enable;
429
430 icmp_param->data.icmph.checksum = 0;
431
432 ipcm_init(&ipc);
433 ipc.tos = ip_hdr(skb)->tos;
434 ipc.sockc.mark = mark;
435 daddr = ipc.addr = ip_hdr(skb)->saddr;
436 saddr = fib_compute_spec_dst(skb);
437
438 if (icmp_param->replyopts.opt.opt.optlen) {
439 ipc.opt = &icmp_param->replyopts.opt;
440 if (ipc.opt->opt.srr)
441 daddr = icmp_param->replyopts.opt.opt.faddr;
442 }
443 memset(&fl4, 0, sizeof(fl4));
444 fl4.daddr = daddr;
445 fl4.saddr = saddr;
446 fl4.flowi4_mark = mark;
447 fl4.flowi4_uid = sock_net_uid(net, NULL);
448 fl4.flowi4_dscp = ip4h_dscp(ip_hdr(skb));
449 fl4.flowi4_proto = IPPROTO_ICMP;
450 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
451 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
452 rt = ip_route_output_key(net, &fl4);
453 if (IS_ERR(rt))
454 goto out_unlock;
455 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
456 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
457 ip_rt_put(rt);
458 out_unlock:
459 icmp_xmit_unlock(sk);
460 out_bh_enable:
461 local_bh_enable();
462 }
463
464 /*
465 * The device used for looking up which routing table to use for sending an ICMP
466 * error is preferably the source whenever it is set, which should ensure the
467 * icmp error can be sent to the source host, else lookup using the routing
468 * table of the destination device, else use the main routing table (index 0).
469 */
icmp_get_route_lookup_dev(struct sk_buff * skb)470 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
471 {
472 struct net_device *dev = skb->dev;
473 const struct dst_entry *dst;
474
475 if (dev)
476 return dev;
477 dst = skb_dst(skb);
478 return dst ? dst_dev(dst) : NULL;
479 }
480
icmp_route_lookup(struct net * net,struct flowi4 * fl4,struct sk_buff * skb_in,const struct iphdr * iph,__be32 saddr,dscp_t dscp,u32 mark,int type,int code,struct icmp_bxm * param)481 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4,
482 struct sk_buff *skb_in,
483 const struct iphdr *iph, __be32 saddr,
484 dscp_t dscp, u32 mark, int type,
485 int code, struct icmp_bxm *param)
486 {
487 struct net_device *route_lookup_dev;
488 struct dst_entry *dst, *dst2;
489 struct rtable *rt, *rt2;
490 struct flowi4 fl4_dec;
491 int err;
492
493 memset(fl4, 0, sizeof(*fl4));
494 fl4->daddr = (param->replyopts.opt.opt.srr ?
495 param->replyopts.opt.opt.faddr : iph->saddr);
496 fl4->saddr = saddr;
497 fl4->flowi4_mark = mark;
498 fl4->flowi4_uid = sock_net_uid(net, NULL);
499 fl4->flowi4_dscp = dscp;
500 fl4->flowi4_proto = IPPROTO_ICMP;
501 fl4->fl4_icmp_type = type;
502 fl4->fl4_icmp_code = code;
503 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
504 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
505
506 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
507 rt = ip_route_output_key_hash(net, fl4, skb_in);
508 if (IS_ERR(rt))
509 return rt;
510
511 /* No need to clone since we're just using its address. */
512 rt2 = rt;
513
514 dst = xfrm_lookup(net, &rt->dst,
515 flowi4_to_flowi(fl4), NULL, 0);
516 rt = dst_rtable(dst);
517 if (!IS_ERR(dst)) {
518 if (rt != rt2)
519 return rt;
520 if (inet_addr_type_dev_table(net, route_lookup_dev,
521 fl4->daddr) == RTN_LOCAL)
522 return rt;
523 } else if (PTR_ERR(dst) == -EPERM) {
524 rt = NULL;
525 } else {
526 return rt;
527 }
528 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
529 if (err)
530 goto relookup_failed;
531
532 if (inet_addr_type_dev_table(net, route_lookup_dev,
533 fl4_dec.saddr) == RTN_LOCAL) {
534 rt2 = __ip_route_output_key(net, &fl4_dec);
535 if (IS_ERR(rt2))
536 err = PTR_ERR(rt2);
537 } else {
538 struct flowi4 fl4_2 = {};
539 unsigned long orefdst;
540
541 fl4_2.daddr = fl4_dec.saddr;
542 rt2 = ip_route_output_key(net, &fl4_2);
543 if (IS_ERR(rt2)) {
544 err = PTR_ERR(rt2);
545 goto relookup_failed;
546 }
547 /* Ugh! */
548 orefdst = skb_dstref_steal(skb_in);
549 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
550 dscp, rt2->dst.dev) ? -EINVAL : 0;
551
552 dst_release(&rt2->dst);
553 rt2 = skb_rtable(skb_in);
554 /* steal dst entry from skb_in, don't drop refcnt */
555 skb_dstref_steal(skb_in);
556 skb_dstref_restore(skb_in, orefdst);
557 }
558
559 if (err)
560 goto relookup_failed;
561
562 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
563 XFRM_LOOKUP_ICMP);
564 rt2 = dst_rtable(dst2);
565 if (!IS_ERR(dst2)) {
566 dst_release(&rt->dst);
567 memcpy(fl4, &fl4_dec, sizeof(*fl4));
568 rt = rt2;
569 } else if (PTR_ERR(dst2) == -EPERM) {
570 if (rt)
571 dst_release(&rt->dst);
572 return rt2;
573 } else {
574 err = PTR_ERR(dst2);
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 inet_skb_parm * parm)596 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
597 const struct inet_skb_parm *parm)
598 {
599 struct iphdr *iph;
600 int room;
601 struct icmp_bxm icmp_param;
602 struct rtable *rt = skb_rtable(skb_in);
603 bool apply_ratelimit = false;
604 struct ipcm_cookie ipc;
605 struct flowi4 fl4;
606 __be32 saddr;
607 u8 tos;
608 u32 mark;
609 struct net *net;
610 struct sock *sk;
611
612 if (!rt)
613 return;
614
615 rcu_read_lock();
616
617 if (rt->dst.dev)
618 net = dev_net_rcu(rt->dst.dev);
619 else if (skb_in->dev)
620 net = dev_net_rcu(skb_in->dev);
621 else
622 goto out;
623
624 /*
625 * Find the original header. It is expected to be valid, of course.
626 * Check this, icmp_send is called from the most obscure devices
627 * sometimes.
628 */
629 iph = ip_hdr(skb_in);
630
631 if ((u8 *)iph < skb_in->head ||
632 (skb_network_header(skb_in) + sizeof(*iph)) >
633 skb_tail_pointer(skb_in))
634 goto out;
635
636 /*
637 * No replies to physical multicast/broadcast
638 */
639 if (skb_in->pkt_type != PACKET_HOST)
640 goto out;
641
642 /*
643 * Now check at the protocol level
644 */
645 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
646 goto out;
647
648 /*
649 * Only reply to fragment 0. We byte re-order the constant
650 * mask for efficiency.
651 */
652 if (iph->frag_off & htons(IP_OFFSET))
653 goto out;
654
655 /*
656 * If we send an ICMP error to an ICMP error a mess would result..
657 */
658 if (icmp_pointers[type].error) {
659 /*
660 * We are an error, check if we are replying to an
661 * ICMP error
662 */
663 if (iph->protocol == IPPROTO_ICMP) {
664 u8 _inner_type, *itp;
665
666 itp = skb_header_pointer(skb_in,
667 skb_network_header(skb_in) +
668 (iph->ihl << 2) +
669 offsetof(struct icmphdr,
670 type) -
671 skb_in->data,
672 sizeof(_inner_type),
673 &_inner_type);
674 if (!itp)
675 goto out;
676
677 /*
678 * Assume any unknown ICMP type is an error. This
679 * isn't specified by the RFC, but think about it..
680 */
681 if (*itp > NR_ICMP_TYPES ||
682 icmp_pointers[*itp].error)
683 goto out;
684 }
685 }
686
687 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */
688 local_bh_disable();
689
690 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
691 * incoming dev is loopback. If outgoing dev change to not be
692 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
693 */
694 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
695 !icmpv4_global_allow(net, type, code, &apply_ratelimit))
696 goto out_bh_enable;
697
698 sk = icmp_xmit_lock(net);
699 if (!sk)
700 goto out_bh_enable;
701
702 /*
703 * Construct source address and options.
704 */
705
706 saddr = iph->daddr;
707 if (!(rt->rt_flags & RTCF_LOCAL)) {
708 struct net_device *dev = NULL;
709
710 rcu_read_lock();
711 if (rt_is_input_route(rt) &&
712 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
713 dev = dev_get_by_index_rcu(net, parm->iif ? parm->iif :
714 inet_iif(skb_in));
715
716 if (dev)
717 saddr = inet_select_addr(dev, iph->saddr,
718 RT_SCOPE_LINK);
719 else
720 saddr = 0;
721 rcu_read_unlock();
722 }
723
724 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
725 IPTOS_PREC_INTERNETCONTROL) :
726 iph->tos;
727 mark = IP4_REPLY_MARK(net, skb_in->mark);
728
729 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in,
730 &parm->opt))
731 goto out_unlock;
732
733
734 /*
735 * Prepare data for ICMP header.
736 */
737
738 icmp_param.data.icmph.type = type;
739 icmp_param.data.icmph.code = code;
740 icmp_param.data.icmph.un.gateway = info;
741 icmp_param.data.icmph.checksum = 0;
742 icmp_param.skb = skb_in;
743 icmp_param.offset = skb_network_offset(skb_in);
744 ipcm_init(&ipc);
745 ipc.tos = tos;
746 ipc.addr = iph->saddr;
747 ipc.opt = &icmp_param.replyopts.opt;
748 ipc.sockc.mark = mark;
749
750 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
751 inet_dsfield_to_dscp(tos), mark, type, code,
752 &icmp_param);
753 if (IS_ERR(rt))
754 goto out_unlock;
755
756 /* peer icmp_ratelimit */
757 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
758 goto ende;
759
760 /* RFC says return as much as we can without exceeding 576 bytes. */
761
762 room = dst_mtu(&rt->dst);
763 if (room > 576)
764 room = 576;
765 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
766 room -= sizeof(struct icmphdr);
767 /* Guard against tiny mtu. We need to include at least one
768 * IP network header for this message to make any sense.
769 */
770 if (room <= (int)sizeof(struct iphdr))
771 goto ende;
772
773 icmp_param.data_len = skb_in->len - icmp_param.offset;
774 if (icmp_param.data_len > room)
775 icmp_param.data_len = room;
776 icmp_param.head_len = sizeof(struct icmphdr);
777
778 /* if we don't have a source address at this point, fall back to the
779 * dummy address instead of sending out a packet with a source address
780 * of 0.0.0.0
781 */
782 if (!fl4.saddr)
783 fl4.saddr = htonl(INADDR_DUMMY);
784
785 trace_icmp_send(skb_in, type, code);
786
787 icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
788 ende:
789 ip_rt_put(rt);
790 out_unlock:
791 icmp_xmit_unlock(sk);
792 out_bh_enable:
793 local_bh_enable();
794 out:
795 rcu_read_unlock();
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 enum ip_conntrack_info ctinfo;
805 enum ip_conntrack_dir dir;
806 struct inet_skb_parm parm;
807 struct nf_conn *ct;
808 __be32 orig_ip;
809
810 memset(&parm, 0, sizeof(parm));
811 ct = nf_ct_get(skb_in, &ctinfo);
812 if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
813 __icmp_send(skb_in, type, code, info, &parm);
814 return;
815 }
816
817 if (skb_shared(skb_in))
818 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
819
820 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
821 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
822 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
823 skb_network_offset(skb_in) + sizeof(struct iphdr))))
824 goto out;
825
826 orig_ip = ip_hdr(skb_in)->saddr;
827 dir = CTINFO2DIR(ctinfo);
828 ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
829 __icmp_send(skb_in, type, code, info, &parm);
830 ip_hdr(skb_in)->saddr = orig_ip;
831 out:
832 consume_skb(cloned_skb);
833 }
834 EXPORT_SYMBOL(icmp_ndo_send);
835 #endif
836
icmp_socket_deliver(struct sk_buff * skb,u32 info)837 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
838 {
839 const struct iphdr *iph = (const struct iphdr *)skb->data;
840 const struct net_protocol *ipprot;
841 int protocol = iph->protocol;
842
843 /* Checkin full IP header plus 8 bytes of protocol to
844 * avoid additional coding at protocol handlers.
845 */
846 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
847 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
848 return;
849 }
850
851 raw_icmp_error(skb, protocol, info);
852
853 ipprot = rcu_dereference(inet_protos[protocol]);
854 if (ipprot && ipprot->err_handler)
855 ipprot->err_handler(skb, info);
856 }
857
icmp_tag_validation(int proto)858 static bool icmp_tag_validation(int proto)
859 {
860 bool ok;
861
862 rcu_read_lock();
863 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
864 rcu_read_unlock();
865 return ok;
866 }
867
868 /*
869 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
870 * ICMP_PARAMETERPROB.
871 */
872
icmp_unreach(struct sk_buff * skb)873 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
874 {
875 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
876 const struct iphdr *iph;
877 struct icmphdr *icmph;
878 struct net *net;
879 u32 info = 0;
880
881 net = skb_dst_dev_net_rcu(skb);
882
883 /*
884 * Incomplete header ?
885 * Only checks for the IP header, there should be an
886 * additional check for longer headers in upper levels.
887 */
888
889 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
890 goto out_err;
891
892 icmph = icmp_hdr(skb);
893 iph = (const struct iphdr *)skb->data;
894
895 if (iph->ihl < 5) { /* Mangled header, drop. */
896 reason = SKB_DROP_REASON_IP_INHDR;
897 goto out_err;
898 }
899
900 switch (icmph->type) {
901 case ICMP_DEST_UNREACH:
902 switch (icmph->code & 15) {
903 case ICMP_NET_UNREACH:
904 case ICMP_HOST_UNREACH:
905 case ICMP_PROT_UNREACH:
906 case ICMP_PORT_UNREACH:
907 break;
908 case ICMP_FRAG_NEEDED:
909 /* for documentation of the ip_no_pmtu_disc
910 * values please see
911 * Documentation/networking/ip-sysctl.rst
912 */
913 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
914 default:
915 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
916 &iph->daddr);
917 break;
918 case 2:
919 goto out;
920 case 3:
921 if (!icmp_tag_validation(iph->protocol))
922 goto out;
923 fallthrough;
924 case 0:
925 info = ntohs(icmph->un.frag.mtu);
926 }
927 break;
928 case ICMP_SR_FAILED:
929 net_dbg_ratelimited("%pI4: Source Route Failed\n",
930 &iph->daddr);
931 break;
932 default:
933 break;
934 }
935 if (icmph->code > NR_ICMP_UNREACH)
936 goto out;
937 break;
938 case ICMP_PARAMETERPROB:
939 info = ntohl(icmph->un.gateway) >> 24;
940 break;
941 case ICMP_TIME_EXCEEDED:
942 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
943 if (icmph->code == ICMP_EXC_FRAGTIME)
944 goto out;
945 break;
946 }
947
948 /*
949 * Throw it at our lower layers
950 *
951 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
952 * header.
953 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
954 * transport layer.
955 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
956 * transport layer.
957 */
958
959 /*
960 * Check the other end isn't violating RFC 1122. Some routers send
961 * bogus responses to broadcast frames. If you see this message
962 * first check your netmask matches at both ends, if it does then
963 * get the other vendor to fix their kit.
964 */
965
966 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
967 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
968 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
969 &ip_hdr(skb)->saddr,
970 icmph->type, icmph->code,
971 &iph->daddr, skb->dev->name);
972 goto out;
973 }
974
975 icmp_socket_deliver(skb, info);
976
977 out:
978 return reason;
979 out_err:
980 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
981 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
982 }
983
984
985 /*
986 * Handle ICMP_REDIRECT.
987 */
988
icmp_redirect(struct sk_buff * skb)989 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
990 {
991 if (skb->len < sizeof(struct iphdr)) {
992 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
993 return SKB_DROP_REASON_PKT_TOO_SMALL;
994 }
995
996 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
997 /* there aught to be a stat */
998 return SKB_DROP_REASON_NOMEM;
999 }
1000
1001 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1002 return SKB_NOT_DROPPED_YET;
1003 }
1004
1005 /*
1006 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1007 *
1008 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1009 * requests.
1010 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1011 * included in the reply.
1012 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1013 * echo requests, MUST have default=NOT.
1014 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
1015 * Extended Echo Functionality, MUST be disabled by default
1016 * See also WRT handling of options once they are done and working.
1017 */
1018
icmp_echo(struct sk_buff * skb)1019 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1020 {
1021 struct icmp_bxm icmp_param;
1022 struct net *net;
1023
1024 net = skb_dst_dev_net_rcu(skb);
1025 /* should there be an ICMP stat for ignored echos? */
1026 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1027 return SKB_NOT_DROPPED_YET;
1028
1029 icmp_param.data.icmph = *icmp_hdr(skb);
1030 icmp_param.skb = skb;
1031 icmp_param.offset = 0;
1032 icmp_param.data_len = skb->len;
1033 icmp_param.head_len = sizeof(struct icmphdr);
1034
1035 if (icmp_param.data.icmph.type == ICMP_ECHO)
1036 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1037 else if (!icmp_build_probe(skb, &icmp_param.data.icmph))
1038 return SKB_NOT_DROPPED_YET;
1039
1040 icmp_reply(&icmp_param, skb);
1041 return SKB_NOT_DROPPED_YET;
1042 }
1043
1044 /* Helper for icmp_echo and icmpv6_echo_reply.
1045 * Searches for net_device that matches PROBE interface identifier
1046 * and builds PROBE reply message in icmphdr.
1047 *
1048 * Returns false if PROBE responses are disabled via sysctl
1049 */
1050
icmp_build_probe(struct sk_buff * skb,struct icmphdr * icmphdr)1051 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1052 {
1053 struct net *net = dev_net_rcu(skb->dev);
1054 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1055 struct icmp_ext_echo_iio *iio, _iio;
1056 struct inet6_dev *in6_dev;
1057 struct in_device *in_dev;
1058 struct net_device *dev;
1059 char buff[IFNAMSIZ];
1060 u16 ident_len;
1061 u8 status;
1062
1063 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1064 return false;
1065
1066 /* We currently only support probing interfaces on the proxy node
1067 * Check to ensure L-bit is set
1068 */
1069 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1070 return false;
1071 /* Clear status bits in reply message */
1072 icmphdr->un.echo.sequence &= htons(0xFF00);
1073 if (icmphdr->type == ICMP_EXT_ECHO)
1074 icmphdr->type = ICMP_EXT_ECHOREPLY;
1075 else
1076 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1077 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1078 /* Size of iio is class_type dependent.
1079 * Only check header here and assign length based on ctype in the switch statement
1080 */
1081 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1082 if (!ext_hdr || !iio)
1083 goto send_mal_query;
1084 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1085 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1086 goto send_mal_query;
1087 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1088 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1089 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1090 if (!iio)
1091 goto send_mal_query;
1092
1093 status = 0;
1094 dev = NULL;
1095 switch (iio->extobj_hdr.class_type) {
1096 case ICMP_EXT_ECHO_CTYPE_NAME:
1097 if (ident_len >= IFNAMSIZ)
1098 goto send_mal_query;
1099 memset(buff, 0, sizeof(buff));
1100 memcpy(buff, &iio->ident.name, ident_len);
1101 dev = dev_get_by_name(net, buff);
1102 break;
1103 case ICMP_EXT_ECHO_CTYPE_INDEX:
1104 if (ident_len != sizeof(iio->ident.ifindex))
1105 goto send_mal_query;
1106 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1107 break;
1108 case ICMP_EXT_ECHO_CTYPE_ADDR:
1109 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1110 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1111 iio->ident.addr.ctype3_hdr.addrlen)
1112 goto send_mal_query;
1113 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1114 case ICMP_AFI_IP:
1115 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1116 goto send_mal_query;
1117 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1118 break;
1119 #if IS_ENABLED(CONFIG_IPV6)
1120 case ICMP_AFI_IP6:
1121 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1122 goto send_mal_query;
1123 dev = ipv6_stub->ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1124 dev_hold(dev);
1125 break;
1126 #endif
1127 default:
1128 goto send_mal_query;
1129 }
1130 break;
1131 default:
1132 goto send_mal_query;
1133 }
1134 if (!dev) {
1135 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1136 return true;
1137 }
1138 /* Fill bits in reply message */
1139 if (dev->flags & IFF_UP)
1140 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1141
1142 in_dev = __in_dev_get_rcu(dev);
1143 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1144 status |= ICMP_EXT_ECHOREPLY_IPV4;
1145
1146 in6_dev = __in6_dev_get(dev);
1147 if (in6_dev && !list_empty(&in6_dev->addr_list))
1148 status |= ICMP_EXT_ECHOREPLY_IPV6;
1149
1150 dev_put(dev);
1151 icmphdr->un.echo.sequence |= htons(status);
1152 return true;
1153 send_mal_query:
1154 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1155 return true;
1156 }
1157 EXPORT_SYMBOL_GPL(icmp_build_probe);
1158
1159 /*
1160 * Handle ICMP Timestamp requests.
1161 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1162 * SHOULD be in the kernel for minimum random latency.
1163 * MUST be accurate to a few minutes.
1164 * MUST be updated at least at 15Hz.
1165 */
icmp_timestamp(struct sk_buff * skb)1166 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1167 {
1168 struct icmp_bxm icmp_param;
1169 /*
1170 * Too short.
1171 */
1172 if (skb->len < 4)
1173 goto out_err;
1174
1175 /*
1176 * Fill in the current time as ms since midnight UT:
1177 */
1178 icmp_param.data.times[1] = inet_current_timestamp();
1179 icmp_param.data.times[2] = icmp_param.data.times[1];
1180
1181 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1182
1183 icmp_param.data.icmph = *icmp_hdr(skb);
1184 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1185 icmp_param.data.icmph.code = 0;
1186 icmp_param.skb = skb;
1187 icmp_param.offset = 0;
1188 icmp_param.data_len = 0;
1189 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1190 icmp_reply(&icmp_param, skb);
1191 return SKB_NOT_DROPPED_YET;
1192
1193 out_err:
1194 __ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1195 return SKB_DROP_REASON_PKT_TOO_SMALL;
1196 }
1197
icmp_discard(struct sk_buff * skb)1198 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1199 {
1200 /* pretend it was a success */
1201 return SKB_NOT_DROPPED_YET;
1202 }
1203
1204 /*
1205 * Deal with incoming ICMP packets.
1206 */
icmp_rcv(struct sk_buff * skb)1207 int icmp_rcv(struct sk_buff *skb)
1208 {
1209 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1210 struct rtable *rt = skb_rtable(skb);
1211 struct net *net = dev_net_rcu(rt->dst.dev);
1212 struct icmphdr *icmph;
1213
1214 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1215 struct sec_path *sp = skb_sec_path(skb);
1216 int nh;
1217
1218 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1219 XFRM_STATE_ICMP)) {
1220 reason = SKB_DROP_REASON_XFRM_POLICY;
1221 goto drop;
1222 }
1223
1224 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1225 goto drop;
1226
1227 nh = skb_network_offset(skb);
1228 skb_set_network_header(skb, sizeof(*icmph));
1229
1230 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1231 skb)) {
1232 reason = SKB_DROP_REASON_XFRM_POLICY;
1233 goto drop;
1234 }
1235
1236 skb_set_network_header(skb, nh);
1237 }
1238
1239 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1240
1241 if (skb_checksum_simple_validate(skb))
1242 goto csum_error;
1243
1244 if (!pskb_pull(skb, sizeof(*icmph)))
1245 goto error;
1246
1247 icmph = icmp_hdr(skb);
1248
1249 ICMPMSGIN_INC_STATS(net, icmph->type);
1250
1251 /* Check for ICMP Extended Echo (PROBE) messages */
1252 if (icmph->type == ICMP_EXT_ECHO) {
1253 /* We can't use icmp_pointers[].handler() because it is an array of
1254 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1255 */
1256 reason = icmp_echo(skb);
1257 goto reason_check;
1258 }
1259
1260 /*
1261 * Parse the ICMP message
1262 */
1263
1264 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1265 /*
1266 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1267 * silently ignored (we let user decide with a sysctl).
1268 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1269 * discarded if to broadcast/multicast.
1270 */
1271 if ((icmph->type == ICMP_ECHO ||
1272 icmph->type == ICMP_TIMESTAMP) &&
1273 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1274 reason = SKB_DROP_REASON_INVALID_PROTO;
1275 goto error;
1276 }
1277 if (icmph->type != ICMP_ECHO &&
1278 icmph->type != ICMP_TIMESTAMP &&
1279 icmph->type != ICMP_ADDRESS &&
1280 icmph->type != ICMP_ADDRESSREPLY) {
1281 reason = SKB_DROP_REASON_INVALID_PROTO;
1282 goto error;
1283 }
1284 }
1285
1286 if (icmph->type == ICMP_EXT_ECHOREPLY ||
1287 icmph->type == ICMP_ECHOREPLY) {
1288 reason = ping_rcv(skb);
1289 return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1290 }
1291
1292 /*
1293 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1294 *
1295 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1296 * discarded.
1297 */
1298 if (icmph->type > NR_ICMP_TYPES) {
1299 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1300 goto error;
1301 }
1302
1303 reason = icmp_pointers[icmph->type].handler(skb);
1304 reason_check:
1305 if (!reason) {
1306 consume_skb(skb);
1307 return NET_RX_SUCCESS;
1308 }
1309
1310 drop:
1311 kfree_skb_reason(skb, reason);
1312 return NET_RX_DROP;
1313 csum_error:
1314 reason = SKB_DROP_REASON_ICMP_CSUM;
1315 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1316 error:
1317 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1318 goto drop;
1319 }
1320
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1321 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1322 {
1323 struct icmp_extobj_hdr *objh, _objh;
1324 struct icmp_ext_hdr *exth, _exth;
1325 u16 olen;
1326
1327 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1328 if (!exth)
1329 return false;
1330 if (exth->version != 2)
1331 return true;
1332
1333 if (exth->checksum &&
1334 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1335 return false;
1336
1337 off += sizeof(_exth);
1338 while (off < skb->len) {
1339 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1340 if (!objh)
1341 return false;
1342
1343 olen = ntohs(objh->length);
1344 if (olen < sizeof(_objh))
1345 return false;
1346
1347 off += olen;
1348 if (off > skb->len)
1349 return false;
1350 }
1351
1352 return true;
1353 }
1354
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1355 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1356 struct sock_ee_data_rfc4884 *out,
1357 int thlen, int off)
1358 {
1359 int hlen;
1360
1361 /* original datagram headers: end of icmph to payload (skb->data) */
1362 hlen = -skb_transport_offset(skb) - thlen;
1363
1364 /* per rfc 4884: minimal datagram length of 128 bytes */
1365 if (off < 128 || off < hlen)
1366 return;
1367
1368 /* kernel has stripped headers: return payload offset in bytes */
1369 off -= hlen;
1370 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1371 return;
1372
1373 out->len = off;
1374
1375 if (!ip_icmp_error_rfc4884_validate(skb, off))
1376 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1377 }
1378 EXPORT_SYMBOL_GPL(ip_icmp_error_rfc4884);
1379
icmp_err(struct sk_buff * skb,u32 info)1380 int icmp_err(struct sk_buff *skb, u32 info)
1381 {
1382 struct iphdr *iph = (struct iphdr *)skb->data;
1383 int offset = iph->ihl<<2;
1384 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1385 struct net *net = dev_net_rcu(skb->dev);
1386 int type = icmp_hdr(skb)->type;
1387 int code = icmp_hdr(skb)->code;
1388
1389 /*
1390 * Use ping_err to handle all icmp errors except those
1391 * triggered by ICMP_ECHOREPLY which sent from kernel.
1392 */
1393 if (icmph->type != ICMP_ECHOREPLY) {
1394 ping_err(skb, offset, info);
1395 return 0;
1396 }
1397
1398 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1399 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1400 else if (type == ICMP_REDIRECT)
1401 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1402
1403 return 0;
1404 }
1405
1406 /*
1407 * This table is the definition of how we handle ICMP.
1408 */
1409 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1410 [ICMP_ECHOREPLY] = {
1411 .handler = ping_rcv,
1412 },
1413 [1] = {
1414 .handler = icmp_discard,
1415 .error = 1,
1416 },
1417 [2] = {
1418 .handler = icmp_discard,
1419 .error = 1,
1420 },
1421 [ICMP_DEST_UNREACH] = {
1422 .handler = icmp_unreach,
1423 .error = 1,
1424 },
1425 [ICMP_SOURCE_QUENCH] = {
1426 .handler = icmp_unreach,
1427 .error = 1,
1428 },
1429 [ICMP_REDIRECT] = {
1430 .handler = icmp_redirect,
1431 .error = 1,
1432 },
1433 [6] = {
1434 .handler = icmp_discard,
1435 .error = 1,
1436 },
1437 [7] = {
1438 .handler = icmp_discard,
1439 .error = 1,
1440 },
1441 [ICMP_ECHO] = {
1442 .handler = icmp_echo,
1443 },
1444 [9] = {
1445 .handler = icmp_discard,
1446 .error = 1,
1447 },
1448 [10] = {
1449 .handler = icmp_discard,
1450 .error = 1,
1451 },
1452 [ICMP_TIME_EXCEEDED] = {
1453 .handler = icmp_unreach,
1454 .error = 1,
1455 },
1456 [ICMP_PARAMETERPROB] = {
1457 .handler = icmp_unreach,
1458 .error = 1,
1459 },
1460 [ICMP_TIMESTAMP] = {
1461 .handler = icmp_timestamp,
1462 },
1463 [ICMP_TIMESTAMPREPLY] = {
1464 .handler = icmp_discard,
1465 },
1466 [ICMP_INFO_REQUEST] = {
1467 .handler = icmp_discard,
1468 },
1469 [ICMP_INFO_REPLY] = {
1470 .handler = icmp_discard,
1471 },
1472 [ICMP_ADDRESS] = {
1473 .handler = icmp_discard,
1474 },
1475 [ICMP_ADDRESSREPLY] = {
1476 .handler = icmp_discard,
1477 },
1478 };
1479
icmp_sk_init(struct net * net)1480 static int __net_init icmp_sk_init(struct net *net)
1481 {
1482 /* Control parameters for ECHO replies. */
1483 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1484 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1485 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1486
1487 /* Control parameter - ignore bogus broadcast responses? */
1488 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1489
1490 /*
1491 * Configurable global rate limit.
1492 *
1493 * ratelimit defines tokens/packet consumed for dst->rate_token
1494 * bucket ratemask defines which icmp types are ratelimited by
1495 * setting it's bit position.
1496 *
1497 * default:
1498 * dest unreachable (3), source quench (4),
1499 * time exceeded (11), parameter problem (12)
1500 */
1501
1502 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1503 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1504 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1505 net->ipv4.sysctl_icmp_msgs_per_sec = 1000;
1506 net->ipv4.sysctl_icmp_msgs_burst = 50;
1507
1508 return 0;
1509 }
1510
1511 static struct pernet_operations __net_initdata icmp_sk_ops = {
1512 .init = icmp_sk_init,
1513 };
1514
icmp_init(void)1515 int __init icmp_init(void)
1516 {
1517 int err, i;
1518
1519 for_each_possible_cpu(i) {
1520 struct sock *sk;
1521
1522 err = inet_ctl_sock_create(&sk, PF_INET,
1523 SOCK_RAW, IPPROTO_ICMP, &init_net);
1524 if (err < 0)
1525 return err;
1526
1527 per_cpu(ipv4_icmp_sk, i) = sk;
1528
1529 /* Enough space for 2 64K ICMP packets, including
1530 * sk_buff/skb_shared_info struct overhead.
1531 */
1532 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1533
1534 /*
1535 * Speedup sock_wfree()
1536 */
1537 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1538 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1539 }
1540 return register_pernet_subsys(&icmp_sk_ops);
1541 }
1542