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/nospec.h>
68 #include <linux/socket.h>
69 #include <linux/in.h>
70 #include <linux/inet.h>
71 #include <linux/inetdevice.h>
72 #include <linux/netdevice.h>
73 #include <linux/string.h>
74 #include <linux/netfilter_ipv4.h>
75 #include <linux/slab.h>
76 #include <net/flow.h>
77 #include <net/snmp.h>
78 #include <net/ip.h>
79 #include <net/route.h>
80 #include <net/protocol.h>
81 #include <net/icmp.h>
82 #include <net/tcp.h>
83 #include <net/udp.h>
84 #include <net/raw.h>
85 #include <net/ping.h>
86 #include <linux/skbuff.h>
87 #include <net/sock.h>
88 #include <linux/errno.h>
89 #include <linux/timer.h>
90 #include <linux/init.h>
91 #include <linux/uaccess.h>
92 #include <net/checksum.h>
93 #include <net/xfrm.h>
94 #include <net/inet_common.h>
95 #include <net/ip_fib.h>
96 #include <net/l3mdev.h>
97 #include <net/addrconf.h>
98 #include <net/inet_dscp.h>
99 #define CREATE_TRACE_POINTS
100 #include <trace/events/icmp.h>
101
102 /*
103 * Build xmit assembly blocks
104 */
105
106 struct icmp_bxm {
107 struct sk_buff *skb;
108 int offset;
109 int data_len;
110
111 struct {
112 struct icmphdr icmph;
113 __be32 times[3];
114 } data;
115 int head_len;
116
117 /* Must be last as it ends in a flexible-array member. */
118 struct ip_options_rcu replyopts;
119 };
120
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
123
124 const struct icmp_err icmp_err_convert[] = {
125 {
126 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
127 .fatal = 0,
128 },
129 {
130 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
131 .fatal = 0,
132 },
133 {
134 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
135 .fatal = 1,
136 },
137 {
138 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
139 .fatal = 1,
140 },
141 {
142 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
143 .fatal = 0,
144 },
145 {
146 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
147 .fatal = 0,
148 },
149 {
150 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
151 .fatal = 1,
152 },
153 {
154 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
155 .fatal = 1,
156 },
157 {
158 .errno = ENONET, /* ICMP_HOST_ISOLATED */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_ANO */
163 .fatal = 1,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
167 .fatal = 1,
168 },
169 {
170 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
171 .fatal = 0,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
175 .fatal = 0,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
179 .fatal = 1,
180 },
181 {
182 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
183 .fatal = 1,
184 },
185 {
186 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
187 .fatal = 1,
188 },
189 };
190 EXPORT_SYMBOL(icmp_err_convert);
191
192 /*
193 * ICMP control array. This specifies what to do with each ICMP.
194 */
195
196 struct icmp_control {
197 enum skb_drop_reason (*handler)(struct sk_buff *skb);
198 short error; /* This ICMP is classed as an error message */
199 };
200
201 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
202
203 static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
204
205 /* Called with BH disabled */
icmp_xmit_lock(struct net * net)206 static inline struct sock *icmp_xmit_lock(struct net *net)
207 {
208 struct sock *sk;
209
210 sk = this_cpu_read(ipv4_icmp_sk);
211
212 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
213 /* This can happen if the output path signals a
214 * dst_link_failure() for an outgoing ICMP packet.
215 */
216 return NULL;
217 }
218 sock_net_set(sk, net);
219 return sk;
220 }
221
icmp_xmit_unlock(struct sock * sk)222 static inline void icmp_xmit_unlock(struct sock *sk)
223 {
224 sock_net_set(sk, &init_net);
225 spin_unlock(&sk->sk_lock.slock);
226 }
227
228 /**
229 * icmp_global_allow - Are we allowed to send one more ICMP message ?
230 * @net: network namespace
231 *
232 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
233 * Returns false if we reached the limit and can not send another packet.
234 * Works in tandem with icmp_global_consume().
235 */
icmp_global_allow(struct net * net)236 bool icmp_global_allow(struct net *net)
237 {
238 u32 delta, now, oldstamp;
239 int incr, new, old;
240
241 /* Note: many cpus could find this condition true.
242 * Then later icmp_global_consume() could consume more credits,
243 * this is an acceptable race.
244 */
245 if (atomic_read(&net->ipv4.icmp_global_credit) > 0)
246 return true;
247
248 now = jiffies;
249 oldstamp = READ_ONCE(net->ipv4.icmp_global_stamp);
250 delta = min_t(u32, now - oldstamp, HZ);
251 if (delta < HZ / 50)
252 return false;
253
254 incr = READ_ONCE(net->ipv4.sysctl_icmp_msgs_per_sec);
255 incr = div_u64((u64)incr * delta, HZ);
256 if (!incr)
257 return false;
258
259 if (cmpxchg(&net->ipv4.icmp_global_stamp, oldstamp, now) == oldstamp) {
260 old = atomic_read(&net->ipv4.icmp_global_credit);
261 do {
262 new = min(old + incr, READ_ONCE(net->ipv4.sysctl_icmp_msgs_burst));
263 } while (!atomic_try_cmpxchg(&net->ipv4.icmp_global_credit, &old, new));
264 }
265 return true;
266 }
267
icmp_global_consume(struct net * net)268 void icmp_global_consume(struct net *net)
269 {
270 int credits = get_random_u32_below(3);
271
272 /* Note: this might make icmp_global.credit negative. */
273 if (credits)
274 atomic_sub(credits, &net->ipv4.icmp_global_credit);
275 }
276
icmpv4_mask_allow(struct net * net,int type,int code)277 static bool icmpv4_mask_allow(struct net *net, int type, int code)
278 {
279 if (type > NR_ICMP_TYPES)
280 return true;
281
282 /* Don't limit PMTU discovery. */
283 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
284 return true;
285
286 /* Limit if icmp type is enabled in ratemask. */
287 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
288 return true;
289
290 return false;
291 }
292
icmpv4_global_allow(struct net * net,int type,int code,bool * apply_ratelimit)293 static bool icmpv4_global_allow(struct net *net, int type, int code,
294 bool *apply_ratelimit)
295 {
296 if (icmpv4_mask_allow(net, type, code))
297 return true;
298
299 if (icmp_global_allow(net)) {
300 *apply_ratelimit = true;
301 return true;
302 }
303 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITGLOBAL);
304 return false;
305 }
306
307 /*
308 * Send an ICMP frame.
309 */
310
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code,bool apply_ratelimit)311 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
312 struct flowi4 *fl4, int type, int code,
313 bool apply_ratelimit)
314 {
315 struct dst_entry *dst = &rt->dst;
316 struct inet_peer *peer;
317 struct net_device *dev;
318 int peer_timeout;
319 bool rc = true;
320
321 if (!apply_ratelimit)
322 return true;
323
324 peer_timeout = READ_ONCE(net->ipv4.sysctl_icmp_ratelimit);
325 if (!peer_timeout)
326 goto out;
327
328 /* No rate limit on loopback */
329 rcu_read_lock();
330 dev = dst_dev_rcu(dst);
331 if (dev && (dev->flags & IFF_LOOPBACK))
332 goto out_unlock;
333
334 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr,
335 l3mdev_master_ifindex_rcu(dev));
336 rc = inet_peer_xrlim_allow(peer, peer_timeout);
337
338 out_unlock:
339 rcu_read_unlock();
340 out:
341 if (!rc)
342 __ICMP_INC_STATS(net, ICMP_MIB_RATELIMITHOST);
343 else
344 icmp_global_consume(net);
345 return rc;
346 }
347
348 /*
349 * Maintain the counters used in the SNMP statistics for outgoing ICMP
350 */
icmp_out_count(struct net * net,unsigned char type)351 void icmp_out_count(struct net *net, unsigned char type)
352 {
353 ICMPMSGOUT_INC_STATS(net, type);
354 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
355 }
356
357 /*
358 * Checksum each fragment, and on the first include the headers and final
359 * checksum.
360 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)361 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
362 struct sk_buff *skb)
363 {
364 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
365 IP_OPTIONS_DATA_FIXED_SIZE);
366 __wsum csum;
367
368 icmp_param = from;
369
370 csum = skb_copy_and_csum_bits(icmp_param->skb,
371 icmp_param->offset + offset,
372 to, len);
373
374 skb->csum = csum_block_add(skb->csum, csum, odd);
375 if (icmp_param->data.icmph.type <= NR_ICMP_TYPES &&
376 icmp_pointers[array_index_nospec(icmp_param->data.icmph.type,
377 NR_ICMP_TYPES + 1)].error)
378 nf_ct_attach(skb, icmp_param->skb);
379 return 0;
380 }
381
icmp_push_reply(struct sock * sk,struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)382 static void icmp_push_reply(struct sock *sk,
383 struct icmp_bxm *icmp_param,
384 struct flowi4 *fl4,
385 struct ipcm_cookie *ipc, struct rtable **rt)
386 {
387 struct sk_buff *skb;
388
389 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
390 icmp_param->data_len+icmp_param->head_len,
391 icmp_param->head_len,
392 ipc, rt, MSG_DONTWAIT) < 0) {
393 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
394 ip_flush_pending_frames(sk);
395 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
396 struct icmphdr *icmph = icmp_hdr(skb);
397 __wsum csum;
398 struct sk_buff *skb1;
399
400 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
401 (char *)icmph,
402 icmp_param->head_len);
403 skb_queue_walk(&sk->sk_write_queue, skb1) {
404 csum = csum_add(csum, skb1->csum);
405 }
406 icmph->checksum = csum_fold(csum);
407 skb->ip_summed = CHECKSUM_NONE;
408 ip_push_pending_frames(sk, fl4);
409 }
410 }
411
412 /*
413 * Driving logic for building and sending ICMP messages.
414 */
415
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)416 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
417 {
418 struct rtable *rt = skb_rtable(skb);
419 struct net *net = dev_net_rcu(rt->dst.dev);
420 bool apply_ratelimit = false;
421 struct ipcm_cookie ipc;
422 struct flowi4 fl4;
423 struct sock *sk;
424 __be32 daddr, saddr;
425 u32 mark = IP4_REPLY_MARK(net, skb->mark);
426 int type = icmp_param->data.icmph.type;
427 int code = icmp_param->data.icmph.code;
428
429 if (ip_options_echo(net, &icmp_param->replyopts.opt, skb))
430 return;
431
432 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */
433 local_bh_disable();
434
435 /* is global icmp_msgs_per_sec exhausted ? */
436 if (!icmpv4_global_allow(net, type, code, &apply_ratelimit))
437 goto out_bh_enable;
438
439 sk = icmp_xmit_lock(net);
440 if (!sk)
441 goto out_bh_enable;
442
443 icmp_param->data.icmph.checksum = 0;
444
445 ipcm_init(&ipc);
446 ipc.tos = ip_hdr(skb)->tos;
447 ipc.sockc.mark = mark;
448 daddr = ipc.addr = ip_hdr(skb)->saddr;
449 saddr = fib_compute_spec_dst(skb);
450
451 if (icmp_param->replyopts.opt.optlen) {
452 ipc.opt = &icmp_param->replyopts;
453 if (ipc.opt->opt.srr)
454 daddr = icmp_param->replyopts.opt.faddr;
455 }
456 memset(&fl4, 0, sizeof(fl4));
457 fl4.daddr = daddr;
458 fl4.saddr = saddr;
459 fl4.flowi4_mark = mark;
460 fl4.flowi4_uid = sock_net_uid(net, NULL);
461 fl4.flowi4_dscp = ip4h_dscp(ip_hdr(skb));
462 fl4.flowi4_proto = IPPROTO_ICMP;
463 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
464 security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
465 rt = ip_route_output_key(net, &fl4);
466 if (IS_ERR(rt))
467 goto out_unlock;
468 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
469 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
470 ip_rt_put(rt);
471 out_unlock:
472 icmp_xmit_unlock(sk);
473 out_bh_enable:
474 local_bh_enable();
475 }
476
477 /*
478 * The device used for looking up which routing table to use for sending an ICMP
479 * error is preferably the source whenever it is set, which should ensure the
480 * icmp error can be sent to the source host, else lookup using the routing
481 * table of the destination device, else use the main routing table (index 0).
482 */
icmp_get_route_lookup_dev(struct sk_buff * skb)483 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
484 {
485 struct net_device *dev = skb->dev;
486 const struct dst_entry *dst;
487
488 if (dev)
489 return dev;
490 dst = skb_dst(skb);
491 return dst ? dst_dev(dst) : NULL;
492 }
493
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)494 static struct rtable *icmp_route_lookup(struct net *net, struct flowi4 *fl4,
495 struct sk_buff *skb_in,
496 const struct iphdr *iph, __be32 saddr,
497 dscp_t dscp, u32 mark, int type,
498 int code, struct icmp_bxm *param)
499 {
500 struct net_device *route_lookup_dev;
501 struct dst_entry *dst, *dst2;
502 struct rtable *rt, *rt2;
503 struct flowi4 fl4_dec;
504 int err;
505
506 memset(fl4, 0, sizeof(*fl4));
507 fl4->daddr = (param->replyopts.opt.srr ?
508 param->replyopts.opt.faddr : iph->saddr);
509 fl4->saddr = saddr;
510 fl4->flowi4_mark = mark;
511 fl4->flowi4_uid = sock_net_uid(net, NULL);
512 fl4->flowi4_dscp = dscp;
513 fl4->flowi4_proto = IPPROTO_ICMP;
514 fl4->fl4_icmp_type = type;
515 fl4->fl4_icmp_code = code;
516 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
517 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
518
519 security_skb_classify_flow(skb_in, flowi4_to_flowi_common(fl4));
520 rt = ip_route_output_key_hash(net, fl4, skb_in);
521 if (IS_ERR(rt))
522 return rt;
523
524 /* No need to clone since we're just using its address. */
525 rt2 = rt;
526
527 dst = xfrm_lookup(net, &rt->dst,
528 flowi4_to_flowi(fl4), NULL, 0);
529 rt = dst_rtable(dst);
530 if (!IS_ERR(dst)) {
531 if (rt != rt2)
532 return rt;
533 if (inet_addr_type_dev_table(net, route_lookup_dev,
534 fl4->daddr) == RTN_LOCAL)
535 return rt;
536 } else if (PTR_ERR(dst) == -EPERM) {
537 rt = NULL;
538 } else {
539 return rt;
540 }
541 err = xfrm_decode_session_reverse(net, skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
542 if (err)
543 goto relookup_failed;
544
545 if (inet_addr_type_dev_table(net, route_lookup_dev,
546 fl4_dec.saddr) == RTN_LOCAL) {
547 rt2 = __ip_route_output_key(net, &fl4_dec);
548 if (IS_ERR(rt2))
549 err = PTR_ERR(rt2);
550 } else {
551 struct flowi4 fl4_2 = fl4_dec;
552 unsigned long orefdst;
553
554 swap(fl4_2.daddr, fl4_2.saddr);
555 switch (fl4_2.flowi4_proto) {
556 case IPPROTO_TCP:
557 case IPPROTO_UDP:
558 case IPPROTO_SCTP:
559 case IPPROTO_DCCP:
560 swap(fl4_2.fl4_sport, fl4_2.fl4_dport);
561 break;
562 }
563
564 fl4_2.flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
565 fl4_2.flowi4_flags |= FLOWI_FLAG_ANYSRC;
566
567 rt2 = __ip_route_output_key(net, &fl4_2);
568 if (IS_ERR(rt2)) {
569 err = PTR_ERR(rt2);
570 goto relookup_failed;
571 }
572 /* Ugh! */
573 orefdst = skb_dstref_steal(skb_in);
574 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
575 dscp, rt2->dst.dev) ? -EINVAL : 0;
576
577 dst_release(&rt2->dst);
578 rt2 = skb_rtable(skb_in);
579 /* steal dst entry from skb_in, don't drop refcnt */
580 skb_dstref_steal(skb_in);
581 skb_dstref_restore(skb_in, orefdst);
582
583 /*
584 * fl4_dec.daddr is not expected to be local here, but it can be
585 * added to an interface concurrently, in which case
586 * ip_route_input() returns a LOCAL route. It can also fail to
587 * build a forwarding route towards fl4_dec.daddr, for example,
588 * when forwarding is disabled, and return an UNREACHABLE route.
589 * Both cases will result in a route with dst.output=ip_rt_bug,
590 * which must not be used for output.
591 */
592 if (!err && rt2 && rt2->rt_type == RTN_LOCAL)
593 net_warn_ratelimited("detected local route for %pI4 during ICMP sending, src %pI4\n",
594 &fl4_dec.daddr, &fl4_dec.saddr);
595 if (!err && rt2 &&
596 (rt2->rt_type == RTN_LOCAL || rt2->rt_type == RTN_UNREACHABLE)) {
597 dst_release(&rt2->dst);
598 err = -EINVAL;
599 }
600 }
601
602 if (err)
603 goto relookup_failed;
604
605 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
606 XFRM_LOOKUP_ICMP);
607 rt2 = dst_rtable(dst2);
608 if (!IS_ERR(dst2)) {
609 dst_release(&rt->dst);
610 rt = rt2;
611 } else if (PTR_ERR(dst2) == -EPERM) {
612 if (rt)
613 dst_release(&rt->dst);
614 return rt2;
615 } else {
616 err = PTR_ERR(dst2);
617 goto relookup_failed;
618 }
619 return rt;
620
621 relookup_failed:
622 if (rt)
623 return rt;
624 return ERR_PTR(err);
625 }
626
627 struct icmp_ext_iio_addr4_subobj {
628 __be16 afi;
629 __be16 reserved;
630 __be32 addr4;
631 };
632
icmp_ext_iio_len(void)633 static unsigned int icmp_ext_iio_len(void)
634 {
635 return sizeof(struct icmp_extobj_hdr) +
636 /* ifIndex */
637 sizeof(__be32) +
638 /* Interface Address Sub-Object */
639 sizeof(struct icmp_ext_iio_addr4_subobj) +
640 /* Interface Name Sub-Object. Length must be a multiple of 4
641 * bytes.
642 */
643 ALIGN(sizeof(struct icmp_ext_iio_name_subobj), 4) +
644 /* MTU */
645 sizeof(__be32);
646 }
647
icmp_ext_max_len(u8 ext_objs)648 static unsigned int icmp_ext_max_len(u8 ext_objs)
649 {
650 unsigned int ext_max_len;
651
652 ext_max_len = sizeof(struct icmp_ext_hdr);
653
654 if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
655 ext_max_len += icmp_ext_iio_len();
656
657 return ext_max_len;
658 }
659
icmp_ext_iio_addr4_find(const struct net_device * dev)660 static __be32 icmp_ext_iio_addr4_find(const struct net_device *dev)
661 {
662 struct in_device *in_dev;
663 struct in_ifaddr *ifa;
664
665 in_dev = __in_dev_get_rcu(dev);
666 if (!in_dev)
667 return 0;
668
669 /* It is unclear from RFC 5837 which IP address should be chosen, but
670 * it makes sense to choose a global unicast address.
671 */
672 in_dev_for_each_ifa_rcu(ifa, in_dev) {
673 if (READ_ONCE(ifa->ifa_flags) & IFA_F_SECONDARY)
674 continue;
675 if (ifa->ifa_scope != RT_SCOPE_UNIVERSE ||
676 ipv4_is_multicast(ifa->ifa_address))
677 continue;
678 return ifa->ifa_address;
679 }
680
681 return 0;
682 }
683
icmp_ext_iio_iif_append(struct net * net,struct sk_buff * skb,int iif)684 static void icmp_ext_iio_iif_append(struct net *net, struct sk_buff *skb,
685 int iif)
686 {
687 struct icmp_ext_iio_name_subobj *name_subobj;
688 struct icmp_extobj_hdr *objh;
689 struct net_device *dev;
690 __be32 data;
691
692 if (!iif)
693 return;
694
695 /* Add the fields in the order specified by RFC 5837. */
696 objh = skb_put(skb, sizeof(*objh));
697 objh->class_num = ICMP_EXT_OBJ_CLASS_IIO;
698 objh->class_type = ICMP_EXT_CTYPE_IIO_ROLE(ICMP_EXT_CTYPE_IIO_ROLE_IIF);
699
700 data = htonl(iif);
701 skb_put_data(skb, &data, sizeof(__be32));
702 objh->class_type |= ICMP_EXT_CTYPE_IIO_IFINDEX;
703
704 rcu_read_lock();
705
706 dev = dev_get_by_index_rcu(net, iif);
707 if (!dev)
708 goto out;
709
710 data = icmp_ext_iio_addr4_find(dev);
711 if (data) {
712 struct icmp_ext_iio_addr4_subobj *addr4_subobj;
713
714 addr4_subobj = skb_put_zero(skb, sizeof(*addr4_subobj));
715 addr4_subobj->afi = htons(ICMP_AFI_IP);
716 addr4_subobj->addr4 = data;
717 objh->class_type |= ICMP_EXT_CTYPE_IIO_IPADDR;
718 }
719
720 name_subobj = skb_put_zero(skb, ALIGN(sizeof(*name_subobj), 4));
721 name_subobj->len = ALIGN(sizeof(*name_subobj), 4);
722 netdev_copy_name(dev, name_subobj->name);
723 objh->class_type |= ICMP_EXT_CTYPE_IIO_NAME;
724
725 data = htonl(READ_ONCE(dev->mtu));
726 skb_put_data(skb, &data, sizeof(__be32));
727 objh->class_type |= ICMP_EXT_CTYPE_IIO_MTU;
728
729 out:
730 rcu_read_unlock();
731 objh->length = htons(skb_tail_pointer(skb) - (unsigned char *)objh);
732 }
733
icmp_ext_objs_append(struct net * net,struct sk_buff * skb,u8 ext_objs,int iif)734 static void icmp_ext_objs_append(struct net *net, struct sk_buff *skb,
735 u8 ext_objs, int iif)
736 {
737 if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
738 icmp_ext_iio_iif_append(net, skb, iif);
739 }
740
741 static struct sk_buff *
icmp_ext_append(struct net * net,struct sk_buff * skb_in,struct icmphdr * icmph,unsigned int room,int iif)742 icmp_ext_append(struct net *net, struct sk_buff *skb_in, struct icmphdr *icmph,
743 unsigned int room, int iif)
744 {
745 unsigned int payload_len, ext_max_len, ext_len;
746 struct icmp_ext_hdr *ext_hdr;
747 struct sk_buff *skb;
748 u8 ext_objs;
749 int nhoff;
750
751 switch (icmph->type) {
752 case ICMP_DEST_UNREACH:
753 case ICMP_TIME_EXCEEDED:
754 case ICMP_PARAMETERPROB:
755 break;
756 default:
757 return NULL;
758 }
759
760 ext_objs = READ_ONCE(net->ipv4.sysctl_icmp_errors_extension_mask);
761 if (!ext_objs)
762 return NULL;
763
764 ext_max_len = icmp_ext_max_len(ext_objs);
765 if (ICMP_EXT_ORIG_DGRAM_MIN_LEN + ext_max_len > room)
766 return NULL;
767
768 skb = skb_clone(skb_in, GFP_ATOMIC);
769 if (!skb)
770 return NULL;
771
772 nhoff = skb_network_offset(skb);
773 payload_len = min(skb->len - nhoff, ICMP_EXT_ORIG_DGRAM_MIN_LEN);
774
775 if (!pskb_network_may_pull(skb, payload_len))
776 goto free_skb;
777
778 if (pskb_trim(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN) ||
779 __skb_put_padto(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN, false))
780 goto free_skb;
781
782 if (pskb_expand_head(skb, 0, ext_max_len, GFP_ATOMIC))
783 goto free_skb;
784
785 ext_hdr = skb_put_zero(skb, sizeof(*ext_hdr));
786 ext_hdr->version = ICMP_EXT_VERSION_2;
787
788 icmp_ext_objs_append(net, skb, ext_objs, iif);
789
790 /* Do not send an empty extension structure. */
791 ext_len = skb_tail_pointer(skb) - (unsigned char *)ext_hdr;
792 if (ext_len == sizeof(*ext_hdr))
793 goto free_skb;
794
795 ext_hdr->checksum = ip_compute_csum(ext_hdr, ext_len);
796 /* The length of the original datagram in 32-bit words (RFC 4884). */
797 icmph->un.reserved[1] = ICMP_EXT_ORIG_DGRAM_MIN_LEN / sizeof(u32);
798
799 return skb;
800
801 free_skb:
802 consume_skb(skb);
803 return NULL;
804 }
805
806 /*
807 * Send an ICMP message in response to a situation
808 *
809 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
810 * MAY send more (we do).
811 * MUST NOT change this header information.
812 * MUST NOT reply to a multicast/broadcast IP address.
813 * MUST NOT reply to a multicast/broadcast MAC address.
814 * MUST reply to only the first fragment.
815 */
816
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct inet_skb_parm * parm)817 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
818 const struct inet_skb_parm *parm)
819 {
820 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
821 IP_OPTIONS_DATA_FIXED_SIZE);
822 struct iphdr *iph;
823 int room;
824 struct rtable *rt = skb_rtable(skb_in);
825 bool apply_ratelimit = false;
826 struct sk_buff *ext_skb;
827 struct ipcm_cookie ipc;
828 struct flowi4 fl4;
829 __be32 saddr;
830 u8 tos;
831 u32 mark;
832 struct net *net;
833 struct sock *sk;
834
835 if (!rt)
836 return;
837
838 rcu_read_lock();
839
840 if (rt->dst.dev)
841 net = dev_net_rcu(rt->dst.dev);
842 else if (skb_in->dev)
843 net = dev_net_rcu(skb_in->dev);
844 else
845 goto out;
846
847 /*
848 * Find the original header. It is expected to be valid, of course.
849 * Check this, icmp_send is called from the most obscure devices
850 * sometimes.
851 */
852 iph = ip_hdr(skb_in);
853
854 if ((u8 *)iph < skb_in->head ||
855 (skb_network_header(skb_in) + sizeof(*iph)) >
856 skb_tail_pointer(skb_in))
857 goto out;
858
859 /*
860 * No replies to physical multicast/broadcast
861 */
862 if (skb_in->pkt_type != PACKET_HOST)
863 goto out;
864
865 /*
866 * Now check at the protocol level
867 */
868 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
869 goto out;
870
871 /*
872 * Only reply to fragment 0. We byte re-order the constant
873 * mask for efficiency.
874 */
875 if (iph->frag_off & htons(IP_OFFSET))
876 goto out;
877
878 /*
879 * If we send an ICMP error to an ICMP error a mess would result..
880 */
881 if (icmp_pointers[type].error) {
882 /*
883 * We are an error, check if we are replying to an
884 * ICMP error
885 */
886 if (iph->protocol == IPPROTO_ICMP) {
887 u8 _inner_type, *itp;
888
889 itp = skb_header_pointer(skb_in,
890 skb_network_header(skb_in) +
891 (iph->ihl << 2) +
892 offsetof(struct icmphdr,
893 type) -
894 skb_in->data,
895 sizeof(_inner_type),
896 &_inner_type);
897 if (!itp)
898 goto out;
899
900 /*
901 * Assume any unknown ICMP type is an error. This
902 * isn't specified by the RFC, but think about it..
903 */
904 if (*itp > NR_ICMP_TYPES ||
905 icmp_pointers[*itp].error)
906 goto out;
907 }
908 }
909
910 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */
911 local_bh_disable();
912
913 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
914 * incoming dev is loopback. If outgoing dev change to not be
915 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
916 */
917 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
918 !icmpv4_global_allow(net, type, code, &apply_ratelimit))
919 goto out_bh_enable;
920
921 sk = icmp_xmit_lock(net);
922 if (!sk)
923 goto out_bh_enable;
924
925 /*
926 * Construct source address and options.
927 */
928
929 saddr = iph->daddr;
930 if (!(rt->rt_flags & RTCF_LOCAL)) {
931 struct net_device *dev = NULL;
932
933 rcu_read_lock();
934 if (rt_is_input_route(rt) &&
935 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
936 dev = dev_get_by_index_rcu(net, parm->iif ? parm->iif :
937 inet_iif(skb_in));
938
939 if (dev)
940 saddr = inet_select_addr(dev, iph->saddr,
941 RT_SCOPE_LINK);
942 else
943 saddr = 0;
944 rcu_read_unlock();
945 }
946
947 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
948 IPTOS_PREC_INTERNETCONTROL) :
949 iph->tos;
950 mark = IP4_REPLY_MARK(net, skb_in->mark);
951
952 if (__ip_options_echo(net, &icmp_param->replyopts.opt, skb_in,
953 &parm->opt))
954 goto out_unlock;
955
956
957 /*
958 * Prepare data for ICMP header.
959 */
960
961 icmp_param->data.icmph.type = type;
962 icmp_param->data.icmph.code = code;
963 icmp_param->data.icmph.un.gateway = info;
964 icmp_param->data.icmph.checksum = 0;
965 icmp_param->skb = skb_in;
966 icmp_param->offset = skb_network_offset(skb_in);
967 ipcm_init(&ipc);
968 ipc.tos = tos;
969 ipc.addr = iph->saddr;
970 ipc.opt = &icmp_param->replyopts;
971 ipc.sockc.mark = mark;
972
973 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
974 inet_dsfield_to_dscp(tos), mark, type, code,
975 icmp_param);
976 if (IS_ERR(rt))
977 goto out_unlock;
978
979 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
980 goto ende;
981
982 /* peer icmp_ratelimit */
983 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
984 goto ende;
985
986 /* RFC says return as much as we can without exceeding 576 bytes. */
987
988 room = dst4_mtu(&rt->dst);
989 if (room > 576)
990 room = 576;
991 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.optlen;
992 room -= sizeof(struct icmphdr);
993 /* Guard against tiny mtu. We need to include at least one
994 * IP network header for this message to make any sense.
995 */
996 if (room <= (int)sizeof(struct iphdr))
997 goto ende;
998
999 ext_skb = icmp_ext_append(net, skb_in, &icmp_param->data.icmph, room,
1000 parm->iif);
1001 if (ext_skb)
1002 icmp_param->skb = ext_skb;
1003
1004 icmp_param->data_len = icmp_param->skb->len - icmp_param->offset;
1005 if (icmp_param->data_len > room)
1006 icmp_param->data_len = room;
1007 icmp_param->head_len = sizeof(struct icmphdr);
1008
1009 /* if we don't have a source address at this point, fall back to the
1010 * dummy address instead of sending out a packet with a source address
1011 * of 0.0.0.0
1012 */
1013 if (!fl4.saddr)
1014 fl4.saddr = htonl(INADDR_DUMMY);
1015
1016 trace_icmp_send(skb_in, type, code);
1017
1018 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
1019
1020 if (ext_skb)
1021 consume_skb(ext_skb);
1022 ende:
1023 ip_rt_put(rt);
1024 out_unlock:
1025 icmp_xmit_unlock(sk);
1026 out_bh_enable:
1027 local_bh_enable();
1028 out:
1029 rcu_read_unlock();
1030 }
1031 EXPORT_SYMBOL(__icmp_send);
1032
1033 #if IS_ENABLED(CONFIG_NF_NAT)
1034 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)1035 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
1036 {
1037 struct sk_buff *cloned_skb = NULL;
1038 enum ip_conntrack_info ctinfo;
1039 enum ip_conntrack_dir dir;
1040 struct inet_skb_parm parm;
1041 struct nf_conn *ct;
1042 __be32 orig_ip;
1043
1044 memset(&parm, 0, sizeof(parm));
1045 ct = nf_ct_get(skb_in, &ctinfo);
1046 if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
1047 __icmp_send(skb_in, type, code, info, &parm);
1048 return;
1049 }
1050
1051 if (skb_shared(skb_in))
1052 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
1053
1054 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
1055 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
1056 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
1057 skb_network_offset(skb_in) + sizeof(struct iphdr))))
1058 goto out;
1059
1060 orig_ip = ip_hdr(skb_in)->saddr;
1061 dir = CTINFO2DIR(ctinfo);
1062 ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
1063 __icmp_send(skb_in, type, code, info, &parm);
1064 ip_hdr(skb_in)->saddr = orig_ip;
1065 out:
1066 consume_skb(cloned_skb);
1067 }
1068 EXPORT_SYMBOL(icmp_ndo_send);
1069 #endif
1070
icmp_socket_deliver(struct sk_buff * skb,u32 info)1071 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
1072 {
1073 const struct iphdr *iph = (const struct iphdr *)skb->data;
1074 const struct net_protocol *ipprot;
1075 int protocol = iph->protocol;
1076
1077 /* Checkin full IP header plus 8 bytes of protocol to
1078 * avoid additional coding at protocol handlers.
1079 */
1080 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
1081 goto out;
1082
1083 /* IPPROTO_RAW sockets are not supposed to receive anything. */
1084 if (protocol == IPPROTO_RAW)
1085 goto out;
1086
1087 raw_icmp_error(skb, protocol, info);
1088
1089 ipprot = rcu_dereference(inet_protos[protocol]);
1090 if (ipprot && ipprot->err_handler)
1091 ipprot->err_handler(skb, info);
1092 return;
1093
1094 out:
1095 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1096 }
1097
icmp_tag_validation(int proto)1098 static bool icmp_tag_validation(int proto)
1099 {
1100 const struct net_protocol *ipprot;
1101 bool ok;
1102
1103 rcu_read_lock();
1104 ipprot = rcu_dereference(inet_protos[proto]);
1105 ok = ipprot ? ipprot->icmp_strict_tag_validation : false;
1106 rcu_read_unlock();
1107 return ok;
1108 }
1109
1110 /*
1111 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
1112 * ICMP_PARAMETERPROB.
1113 */
1114
icmp_unreach(struct sk_buff * skb)1115 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
1116 {
1117 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
1118 const struct iphdr *iph;
1119 struct icmphdr *icmph;
1120 struct net *net;
1121 u32 info = 0;
1122
1123 net = skb_dst_dev_net_rcu(skb);
1124
1125 /*
1126 * Incomplete header ?
1127 * Only checks for the IP header, there should be an
1128 * additional check for longer headers in upper levels.
1129 */
1130
1131 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1132 goto out_err;
1133
1134 icmph = icmp_hdr(skb);
1135 iph = (const struct iphdr *)skb->data;
1136
1137 if (iph->ihl < 5) { /* Mangled header, drop. */
1138 reason = SKB_DROP_REASON_IP_INHDR;
1139 goto out_err;
1140 }
1141
1142 switch (icmph->type) {
1143 case ICMP_DEST_UNREACH:
1144 switch (icmph->code & 15) {
1145 case ICMP_NET_UNREACH:
1146 case ICMP_HOST_UNREACH:
1147 case ICMP_PROT_UNREACH:
1148 case ICMP_PORT_UNREACH:
1149 break;
1150 case ICMP_FRAG_NEEDED:
1151 /* for documentation of the ip_no_pmtu_disc
1152 * values please see
1153 * Documentation/networking/ip-sysctl.rst
1154 */
1155 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
1156 default:
1157 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
1158 &iph->daddr);
1159 break;
1160 case 2:
1161 goto out;
1162 case 3:
1163 if (!icmp_tag_validation(iph->protocol))
1164 goto out;
1165 fallthrough;
1166 case 0:
1167 info = ntohs(icmph->un.frag.mtu);
1168 }
1169 break;
1170 case ICMP_SR_FAILED:
1171 net_dbg_ratelimited("%pI4: Source Route Failed\n",
1172 &iph->daddr);
1173 break;
1174 default:
1175 break;
1176 }
1177 if (icmph->code > NR_ICMP_UNREACH)
1178 goto out;
1179 break;
1180 case ICMP_PARAMETERPROB:
1181 info = ntohl(icmph->un.gateway) >> 24;
1182 break;
1183 case ICMP_TIME_EXCEEDED:
1184 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
1185 if (icmph->code == ICMP_EXC_FRAGTIME)
1186 goto out;
1187 break;
1188 }
1189
1190 /*
1191 * Throw it at our lower layers
1192 *
1193 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
1194 * header.
1195 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
1196 * transport layer.
1197 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
1198 * transport layer.
1199 */
1200
1201 /*
1202 * Check the other end isn't violating RFC 1122. Some routers send
1203 * bogus responses to broadcast frames. If you see this message
1204 * first check your netmask matches at both ends, if it does then
1205 * get the other vendor to fix their kit.
1206 */
1207
1208 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
1209 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
1210 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
1211 &ip_hdr(skb)->saddr,
1212 icmph->type, icmph->code,
1213 &iph->daddr, skb->dev->name);
1214 goto out;
1215 }
1216
1217 icmp_socket_deliver(skb, info);
1218
1219 out:
1220 return reason;
1221 out_err:
1222 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1223 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
1224 }
1225
1226
1227 /*
1228 * Handle ICMP_REDIRECT.
1229 */
1230
icmp_redirect(struct sk_buff * skb)1231 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
1232 {
1233 if (skb->len < sizeof(struct iphdr)) {
1234 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1235 return SKB_DROP_REASON_PKT_TOO_SMALL;
1236 }
1237
1238 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
1239 /* there aught to be a stat */
1240 return SKB_DROP_REASON_NOMEM;
1241 }
1242
1243 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1244 return SKB_NOT_DROPPED_YET;
1245 }
1246
1247 /*
1248 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1249 *
1250 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1251 * requests.
1252 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1253 * included in the reply.
1254 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1255 * echo requests, MUST have default=NOT.
1256 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
1257 * Extended Echo Functionality, MUST be disabled by default
1258 * See also WRT handling of options once they are done and working.
1259 */
1260
icmp_echo(struct sk_buff * skb)1261 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1262 {
1263 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1264 IP_OPTIONS_DATA_FIXED_SIZE);
1265 struct net *net;
1266
1267 net = skb_dst_dev_net_rcu(skb);
1268 /* should there be an ICMP stat for ignored echos? */
1269 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1270 return SKB_NOT_DROPPED_YET;
1271
1272 icmp_param->data.icmph = *icmp_hdr(skb);
1273 icmp_param->skb = skb;
1274 icmp_param->offset = 0;
1275 icmp_param->data_len = skb->len;
1276 icmp_param->head_len = sizeof(struct icmphdr);
1277
1278 if (icmp_param->data.icmph.type == ICMP_ECHO)
1279 icmp_param->data.icmph.type = ICMP_ECHOREPLY;
1280 else if (!icmp_build_probe(skb, &icmp_param->data.icmph))
1281 return SKB_NOT_DROPPED_YET;
1282
1283 icmp_reply(icmp_param, skb);
1284 return SKB_NOT_DROPPED_YET;
1285 }
1286
1287 /* Helper for icmp_echo and icmpv6_echo_reply.
1288 * Searches for net_device that matches PROBE interface identifier
1289 * and builds PROBE reply message in icmphdr.
1290 *
1291 * Returns false if PROBE responses are disabled via sysctl
1292 */
1293
icmp_build_probe(struct sk_buff * skb,struct icmphdr * icmphdr)1294 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1295 {
1296 struct net *net = dev_net_rcu(skb->dev);
1297 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1298 struct icmp_ext_echo_iio *iio, _iio;
1299 struct inet6_dev *in6_dev;
1300 struct in_device *in_dev;
1301 struct net_device *dev;
1302 char buff[IFNAMSIZ];
1303 u16 ident_len;
1304 u8 status;
1305
1306 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1307 return false;
1308
1309 /* We currently only support probing interfaces on the proxy node
1310 * Check to ensure L-bit is set
1311 */
1312 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1313 return false;
1314 /* Clear status bits in reply message */
1315 icmphdr->un.echo.sequence &= htons(0xFF00);
1316 if (icmphdr->type == ICMP_EXT_ECHO)
1317 icmphdr->type = ICMP_EXT_ECHOREPLY;
1318 else
1319 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1320 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1321 /* Size of iio is class_type dependent.
1322 * Only check header here and assign length based on ctype in the switch statement
1323 */
1324 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1325 if (!ext_hdr || !iio)
1326 goto send_mal_query;
1327 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1328 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1329 goto send_mal_query;
1330 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1331 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1332 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1333 if (!iio)
1334 goto send_mal_query;
1335
1336 status = 0;
1337 dev = NULL;
1338 switch (iio->extobj_hdr.class_type) {
1339 case ICMP_EXT_ECHO_CTYPE_NAME:
1340 if (ident_len >= IFNAMSIZ)
1341 goto send_mal_query;
1342 memset(buff, 0, sizeof(buff));
1343 memcpy(buff, &iio->ident.name, ident_len);
1344 dev = dev_get_by_name(net, buff);
1345 break;
1346 case ICMP_EXT_ECHO_CTYPE_INDEX:
1347 if (ident_len != sizeof(iio->ident.ifindex))
1348 goto send_mal_query;
1349 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1350 break;
1351 case ICMP_EXT_ECHO_CTYPE_ADDR:
1352 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1353 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1354 iio->ident.addr.ctype3_hdr.addrlen)
1355 goto send_mal_query;
1356 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1357 case ICMP_AFI_IP:
1358 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1359 goto send_mal_query;
1360 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1361 break;
1362 #if IS_ENABLED(CONFIG_IPV6)
1363 case ICMP_AFI_IP6:
1364 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1365 goto send_mal_query;
1366 dev = ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1367 dev_hold(dev);
1368 break;
1369 #endif
1370 default:
1371 goto send_mal_query;
1372 }
1373 break;
1374 default:
1375 goto send_mal_query;
1376 }
1377 if (!dev) {
1378 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1379 return true;
1380 }
1381 /* Fill bits in reply message */
1382 if (dev->flags & IFF_UP)
1383 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1384
1385 in_dev = __in_dev_get_rcu(dev);
1386 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1387 status |= ICMP_EXT_ECHOREPLY_IPV4;
1388
1389 in6_dev = __in6_dev_get(dev);
1390 if (in6_dev && !list_empty(&in6_dev->addr_list))
1391 status |= ICMP_EXT_ECHOREPLY_IPV6;
1392
1393 dev_put(dev);
1394 icmphdr->un.echo.sequence |= htons(status);
1395 return true;
1396 send_mal_query:
1397 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1398 return true;
1399 }
1400
1401 /*
1402 * Handle ICMP Timestamp requests.
1403 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1404 * SHOULD be in the kernel for minimum random latency.
1405 * MUST be accurate to a few minutes.
1406 * MUST be updated at least at 15Hz.
1407 */
icmp_timestamp(struct sk_buff * skb)1408 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1409 {
1410 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1411 IP_OPTIONS_DATA_FIXED_SIZE);
1412 /*
1413 * Too short.
1414 */
1415 if (skb->len < 4)
1416 goto out_err;
1417
1418 /*
1419 * Fill in the current time as ms since midnight UT:
1420 */
1421 icmp_param->data.times[1] = inet_current_timestamp();
1422 icmp_param->data.times[2] = icmp_param->data.times[1];
1423
1424 BUG_ON(skb_copy_bits(skb, 0, &icmp_param->data.times[0], 4));
1425
1426 icmp_param->data.icmph = *icmp_hdr(skb);
1427 icmp_param->data.icmph.type = ICMP_TIMESTAMPREPLY;
1428 icmp_param->data.icmph.code = 0;
1429 icmp_param->skb = skb;
1430 icmp_param->offset = 0;
1431 icmp_param->data_len = 0;
1432 icmp_param->head_len = sizeof(struct icmphdr) + 12;
1433 icmp_reply(icmp_param, skb);
1434 return SKB_NOT_DROPPED_YET;
1435
1436 out_err:
1437 __ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1438 return SKB_DROP_REASON_PKT_TOO_SMALL;
1439 }
1440
icmp_discard(struct sk_buff * skb)1441 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1442 {
1443 /* pretend it was a success */
1444 return SKB_NOT_DROPPED_YET;
1445 }
1446
1447 /*
1448 * Deal with incoming ICMP packets.
1449 */
icmp_rcv(struct sk_buff * skb)1450 int icmp_rcv(struct sk_buff *skb)
1451 {
1452 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1453 struct rtable *rt = skb_rtable(skb);
1454 struct net *net = dev_net_rcu(rt->dst.dev);
1455 struct icmphdr *icmph;
1456
1457 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1458 struct sec_path *sp = skb_sec_path(skb);
1459 int nh;
1460
1461 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1462 XFRM_STATE_ICMP)) {
1463 reason = SKB_DROP_REASON_XFRM_POLICY;
1464 goto drop;
1465 }
1466
1467 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1468 goto drop;
1469
1470 nh = skb_network_offset(skb);
1471 skb_set_network_header(skb, sizeof(*icmph));
1472
1473 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1474 skb)) {
1475 reason = SKB_DROP_REASON_XFRM_POLICY;
1476 goto drop;
1477 }
1478
1479 skb_set_network_header(skb, nh);
1480 }
1481
1482 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1483
1484 if (skb_checksum_simple_validate(skb))
1485 goto csum_error;
1486
1487 if (!pskb_pull(skb, sizeof(*icmph)))
1488 goto error;
1489
1490 icmph = icmp_hdr(skb);
1491
1492 ICMPMSGIN_INC_STATS(net, icmph->type);
1493
1494 /* Check for ICMP Extended Echo (PROBE) messages */
1495 if (icmph->type == ICMP_EXT_ECHO) {
1496 /* We can't use icmp_pointers[].handler() because it is an array of
1497 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1498 */
1499 reason = icmp_echo(skb);
1500 goto reason_check;
1501 }
1502
1503 /*
1504 * Parse the ICMP message
1505 */
1506
1507 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1508 /*
1509 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1510 * silently ignored (we let user decide with a sysctl).
1511 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1512 * discarded if to broadcast/multicast.
1513 */
1514 if ((icmph->type == ICMP_ECHO ||
1515 icmph->type == ICMP_TIMESTAMP) &&
1516 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1517 reason = SKB_DROP_REASON_INVALID_PROTO;
1518 goto error;
1519 }
1520 if (icmph->type != ICMP_ECHO &&
1521 icmph->type != ICMP_TIMESTAMP &&
1522 icmph->type != ICMP_ADDRESS &&
1523 icmph->type != ICMP_ADDRESSREPLY) {
1524 reason = SKB_DROP_REASON_INVALID_PROTO;
1525 goto error;
1526 }
1527 }
1528
1529 if (icmph->type == ICMP_EXT_ECHOREPLY ||
1530 icmph->type == ICMP_ECHOREPLY) {
1531 reason = ping_rcv(skb);
1532 return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1533 }
1534
1535 /*
1536 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1537 *
1538 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1539 * discarded.
1540 */
1541 if (icmph->type > NR_ICMP_TYPES) {
1542 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1543 goto error;
1544 }
1545
1546 reason = icmp_pointers[icmph->type].handler(skb);
1547 reason_check:
1548 if (!reason) {
1549 consume_skb(skb);
1550 return NET_RX_SUCCESS;
1551 }
1552
1553 drop:
1554 kfree_skb_reason(skb, reason);
1555 return NET_RX_DROP;
1556 csum_error:
1557 reason = SKB_DROP_REASON_ICMP_CSUM;
1558 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1559 error:
1560 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1561 goto drop;
1562 }
1563
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1564 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1565 {
1566 struct icmp_extobj_hdr *objh, _objh;
1567 struct icmp_ext_hdr *exth, _exth;
1568 u16 olen;
1569
1570 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1571 if (!exth)
1572 return false;
1573 if (exth->version != 2)
1574 return true;
1575
1576 if (exth->checksum &&
1577 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1578 return false;
1579
1580 off += sizeof(_exth);
1581 while (off < skb->len) {
1582 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1583 if (!objh)
1584 return false;
1585
1586 olen = ntohs(objh->length);
1587 if (olen < sizeof(_objh))
1588 return false;
1589
1590 off += olen;
1591 if (off > skb->len)
1592 return false;
1593 }
1594
1595 return true;
1596 }
1597
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1598 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1599 struct sock_ee_data_rfc4884 *out,
1600 int thlen, int off)
1601 {
1602 int hlen;
1603
1604 /* original datagram headers: end of icmph to payload (skb->data) */
1605 hlen = -skb_transport_offset(skb) - thlen;
1606
1607 /* per rfc 4884: minimal datagram length of 128 bytes */
1608 if (off < 128 || off < hlen)
1609 return;
1610
1611 /* kernel has stripped headers: return payload offset in bytes */
1612 off -= hlen;
1613 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1614 return;
1615
1616 out->len = off;
1617
1618 if (!ip_icmp_error_rfc4884_validate(skb, off))
1619 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1620 }
1621
icmp_err(struct sk_buff * skb,u32 info)1622 int icmp_err(struct sk_buff *skb, u32 info)
1623 {
1624 struct iphdr *iph = (struct iphdr *)skb->data;
1625 int offset = iph->ihl<<2;
1626 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1627 struct net *net = dev_net_rcu(skb->dev);
1628 int type = icmp_hdr(skb)->type;
1629 int code = icmp_hdr(skb)->code;
1630
1631 /*
1632 * Use ping_err to handle all icmp errors except those
1633 * triggered by ICMP_ECHOREPLY which sent from kernel.
1634 */
1635 if (icmph->type != ICMP_ECHOREPLY) {
1636 ping_err(skb, offset, info);
1637 return 0;
1638 }
1639
1640 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1641 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1642 else if (type == ICMP_REDIRECT)
1643 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1644
1645 return 0;
1646 }
1647
1648 /*
1649 * This table is the definition of how we handle ICMP.
1650 */
1651 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1652 [ICMP_ECHOREPLY] = {
1653 .handler = ping_rcv,
1654 },
1655 [1] = {
1656 .handler = icmp_discard,
1657 .error = 1,
1658 },
1659 [2] = {
1660 .handler = icmp_discard,
1661 .error = 1,
1662 },
1663 [ICMP_DEST_UNREACH] = {
1664 .handler = icmp_unreach,
1665 .error = 1,
1666 },
1667 [ICMP_SOURCE_QUENCH] = {
1668 .handler = icmp_unreach,
1669 .error = 1,
1670 },
1671 [ICMP_REDIRECT] = {
1672 .handler = icmp_redirect,
1673 .error = 1,
1674 },
1675 [6] = {
1676 .handler = icmp_discard,
1677 .error = 1,
1678 },
1679 [7] = {
1680 .handler = icmp_discard,
1681 .error = 1,
1682 },
1683 [ICMP_ECHO] = {
1684 .handler = icmp_echo,
1685 },
1686 [9] = {
1687 .handler = icmp_discard,
1688 .error = 1,
1689 },
1690 [10] = {
1691 .handler = icmp_discard,
1692 .error = 1,
1693 },
1694 [ICMP_TIME_EXCEEDED] = {
1695 .handler = icmp_unreach,
1696 .error = 1,
1697 },
1698 [ICMP_PARAMETERPROB] = {
1699 .handler = icmp_unreach,
1700 .error = 1,
1701 },
1702 [ICMP_TIMESTAMP] = {
1703 .handler = icmp_timestamp,
1704 },
1705 [ICMP_TIMESTAMPREPLY] = {
1706 .handler = icmp_discard,
1707 },
1708 [ICMP_INFO_REQUEST] = {
1709 .handler = icmp_discard,
1710 },
1711 [ICMP_INFO_REPLY] = {
1712 .handler = icmp_discard,
1713 },
1714 [ICMP_ADDRESS] = {
1715 .handler = icmp_discard,
1716 },
1717 [ICMP_ADDRESSREPLY] = {
1718 .handler = icmp_discard,
1719 },
1720 };
1721
icmp_sk_init(struct net * net)1722 static int __net_init icmp_sk_init(struct net *net)
1723 {
1724 /* Control parameters for ECHO replies. */
1725 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1726 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1727 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1728
1729 /* Control parameter - ignore bogus broadcast responses? */
1730 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1731
1732 /*
1733 * Configurable global rate limit.
1734 *
1735 * ratelimit defines tokens/packet consumed for dst->rate_token
1736 * bucket ratemask defines which icmp types are ratelimited by
1737 * setting it's bit position.
1738 *
1739 * default:
1740 * dest unreachable (3), source quench (4),
1741 * time exceeded (11), parameter problem (12)
1742 */
1743
1744 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1745 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1746 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1747 net->ipv4.sysctl_icmp_errors_extension_mask = 0;
1748 net->ipv4.sysctl_icmp_msgs_per_sec = 10000;
1749 net->ipv4.sysctl_icmp_msgs_burst = 10000;
1750
1751 return 0;
1752 }
1753
1754 static struct pernet_operations __net_initdata icmp_sk_ops = {
1755 .init = icmp_sk_init,
1756 };
1757
icmp_init(void)1758 int __init icmp_init(void)
1759 {
1760 int err, i;
1761
1762 for_each_possible_cpu(i) {
1763 struct sock *sk;
1764
1765 err = inet_ctl_sock_create(&sk, PF_INET,
1766 SOCK_RAW, IPPROTO_ICMP, &init_net);
1767 if (err < 0)
1768 return err;
1769
1770 per_cpu(ipv4_icmp_sk, i) = sk;
1771
1772 /* Enough space for 2 64K ICMP packets, including
1773 * sk_buff/skb_shared_info struct overhead.
1774 */
1775 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1776
1777 /*
1778 * Speedup sock_wfree()
1779 */
1780 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1781 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1782 }
1783 return register_pernet_subsys(&icmp_sk_ops);
1784 }
1785