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 = {};
552 unsigned long orefdst;
553
554 fl4_2.daddr = fl4_dec.saddr;
555 rt2 = ip_route_output_key(net, &fl4_2);
556 if (IS_ERR(rt2)) {
557 err = PTR_ERR(rt2);
558 goto relookup_failed;
559 }
560 /* Ugh! */
561 orefdst = skb_dstref_steal(skb_in);
562 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
563 dscp, rt2->dst.dev) ? -EINVAL : 0;
564
565 dst_release(&rt2->dst);
566 rt2 = skb_rtable(skb_in);
567 /* steal dst entry from skb_in, don't drop refcnt */
568 skb_dstref_steal(skb_in);
569 skb_dstref_restore(skb_in, orefdst);
570
571 /*
572 * At this point, fl4_dec.daddr should NOT be local (we
573 * checked fl4_dec.saddr above). However, a race condition
574 * may occur if the address is added to the interface
575 * concurrently. In that case, ip_route_input() returns a
576 * LOCAL route with dst.output=ip_rt_bug, which must not
577 * be used for output.
578 */
579 if (!err && rt2 && rt2->rt_type == RTN_LOCAL) {
580 net_warn_ratelimited("detected local route for %pI4 during ICMP sending, src %pI4\n",
581 &fl4_dec.daddr, &fl4_dec.saddr);
582 dst_release(&rt2->dst);
583 err = -EINVAL;
584 }
585 }
586
587 if (err)
588 goto relookup_failed;
589
590 dst2 = xfrm_lookup(net, &rt2->dst, flowi4_to_flowi(&fl4_dec), NULL,
591 XFRM_LOOKUP_ICMP);
592 rt2 = dst_rtable(dst2);
593 if (!IS_ERR(dst2)) {
594 dst_release(&rt->dst);
595 rt = rt2;
596 } else if (PTR_ERR(dst2) == -EPERM) {
597 if (rt)
598 dst_release(&rt->dst);
599 return rt2;
600 } else {
601 err = PTR_ERR(dst2);
602 goto relookup_failed;
603 }
604 return rt;
605
606 relookup_failed:
607 if (rt)
608 return rt;
609 return ERR_PTR(err);
610 }
611
612 struct icmp_ext_iio_addr4_subobj {
613 __be16 afi;
614 __be16 reserved;
615 __be32 addr4;
616 };
617
icmp_ext_iio_len(void)618 static unsigned int icmp_ext_iio_len(void)
619 {
620 return sizeof(struct icmp_extobj_hdr) +
621 /* ifIndex */
622 sizeof(__be32) +
623 /* Interface Address Sub-Object */
624 sizeof(struct icmp_ext_iio_addr4_subobj) +
625 /* Interface Name Sub-Object. Length must be a multiple of 4
626 * bytes.
627 */
628 ALIGN(sizeof(struct icmp_ext_iio_name_subobj), 4) +
629 /* MTU */
630 sizeof(__be32);
631 }
632
icmp_ext_max_len(u8 ext_objs)633 static unsigned int icmp_ext_max_len(u8 ext_objs)
634 {
635 unsigned int ext_max_len;
636
637 ext_max_len = sizeof(struct icmp_ext_hdr);
638
639 if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
640 ext_max_len += icmp_ext_iio_len();
641
642 return ext_max_len;
643 }
644
icmp_ext_iio_addr4_find(const struct net_device * dev)645 static __be32 icmp_ext_iio_addr4_find(const struct net_device *dev)
646 {
647 struct in_device *in_dev;
648 struct in_ifaddr *ifa;
649
650 in_dev = __in_dev_get_rcu(dev);
651 if (!in_dev)
652 return 0;
653
654 /* It is unclear from RFC 5837 which IP address should be chosen, but
655 * it makes sense to choose a global unicast address.
656 */
657 in_dev_for_each_ifa_rcu(ifa, in_dev) {
658 if (READ_ONCE(ifa->ifa_flags) & IFA_F_SECONDARY)
659 continue;
660 if (ifa->ifa_scope != RT_SCOPE_UNIVERSE ||
661 ipv4_is_multicast(ifa->ifa_address))
662 continue;
663 return ifa->ifa_address;
664 }
665
666 return 0;
667 }
668
icmp_ext_iio_iif_append(struct net * net,struct sk_buff * skb,int iif)669 static void icmp_ext_iio_iif_append(struct net *net, struct sk_buff *skb,
670 int iif)
671 {
672 struct icmp_ext_iio_name_subobj *name_subobj;
673 struct icmp_extobj_hdr *objh;
674 struct net_device *dev;
675 __be32 data;
676
677 if (!iif)
678 return;
679
680 /* Add the fields in the order specified by RFC 5837. */
681 objh = skb_put(skb, sizeof(*objh));
682 objh->class_num = ICMP_EXT_OBJ_CLASS_IIO;
683 objh->class_type = ICMP_EXT_CTYPE_IIO_ROLE(ICMP_EXT_CTYPE_IIO_ROLE_IIF);
684
685 data = htonl(iif);
686 skb_put_data(skb, &data, sizeof(__be32));
687 objh->class_type |= ICMP_EXT_CTYPE_IIO_IFINDEX;
688
689 rcu_read_lock();
690
691 dev = dev_get_by_index_rcu(net, iif);
692 if (!dev)
693 goto out;
694
695 data = icmp_ext_iio_addr4_find(dev);
696 if (data) {
697 struct icmp_ext_iio_addr4_subobj *addr4_subobj;
698
699 addr4_subobj = skb_put_zero(skb, sizeof(*addr4_subobj));
700 addr4_subobj->afi = htons(ICMP_AFI_IP);
701 addr4_subobj->addr4 = data;
702 objh->class_type |= ICMP_EXT_CTYPE_IIO_IPADDR;
703 }
704
705 name_subobj = skb_put_zero(skb, ALIGN(sizeof(*name_subobj), 4));
706 name_subobj->len = ALIGN(sizeof(*name_subobj), 4);
707 netdev_copy_name(dev, name_subobj->name);
708 objh->class_type |= ICMP_EXT_CTYPE_IIO_NAME;
709
710 data = htonl(READ_ONCE(dev->mtu));
711 skb_put_data(skb, &data, sizeof(__be32));
712 objh->class_type |= ICMP_EXT_CTYPE_IIO_MTU;
713
714 out:
715 rcu_read_unlock();
716 objh->length = htons(skb_tail_pointer(skb) - (unsigned char *)objh);
717 }
718
icmp_ext_objs_append(struct net * net,struct sk_buff * skb,u8 ext_objs,int iif)719 static void icmp_ext_objs_append(struct net *net, struct sk_buff *skb,
720 u8 ext_objs, int iif)
721 {
722 if (ext_objs & BIT(ICMP_ERR_EXT_IIO_IIF))
723 icmp_ext_iio_iif_append(net, skb, iif);
724 }
725
726 static struct sk_buff *
icmp_ext_append(struct net * net,struct sk_buff * skb_in,struct icmphdr * icmph,unsigned int room,int iif)727 icmp_ext_append(struct net *net, struct sk_buff *skb_in, struct icmphdr *icmph,
728 unsigned int room, int iif)
729 {
730 unsigned int payload_len, ext_max_len, ext_len;
731 struct icmp_ext_hdr *ext_hdr;
732 struct sk_buff *skb;
733 u8 ext_objs;
734 int nhoff;
735
736 switch (icmph->type) {
737 case ICMP_DEST_UNREACH:
738 case ICMP_TIME_EXCEEDED:
739 case ICMP_PARAMETERPROB:
740 break;
741 default:
742 return NULL;
743 }
744
745 ext_objs = READ_ONCE(net->ipv4.sysctl_icmp_errors_extension_mask);
746 if (!ext_objs)
747 return NULL;
748
749 ext_max_len = icmp_ext_max_len(ext_objs);
750 if (ICMP_EXT_ORIG_DGRAM_MIN_LEN + ext_max_len > room)
751 return NULL;
752
753 skb = skb_clone(skb_in, GFP_ATOMIC);
754 if (!skb)
755 return NULL;
756
757 nhoff = skb_network_offset(skb);
758 payload_len = min(skb->len - nhoff, ICMP_EXT_ORIG_DGRAM_MIN_LEN);
759
760 if (!pskb_network_may_pull(skb, payload_len))
761 goto free_skb;
762
763 if (pskb_trim(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN) ||
764 __skb_put_padto(skb, nhoff + ICMP_EXT_ORIG_DGRAM_MIN_LEN, false))
765 goto free_skb;
766
767 if (pskb_expand_head(skb, 0, ext_max_len, GFP_ATOMIC))
768 goto free_skb;
769
770 ext_hdr = skb_put_zero(skb, sizeof(*ext_hdr));
771 ext_hdr->version = ICMP_EXT_VERSION_2;
772
773 icmp_ext_objs_append(net, skb, ext_objs, iif);
774
775 /* Do not send an empty extension structure. */
776 ext_len = skb_tail_pointer(skb) - (unsigned char *)ext_hdr;
777 if (ext_len == sizeof(*ext_hdr))
778 goto free_skb;
779
780 ext_hdr->checksum = ip_compute_csum(ext_hdr, ext_len);
781 /* The length of the original datagram in 32-bit words (RFC 4884). */
782 icmph->un.reserved[1] = ICMP_EXT_ORIG_DGRAM_MIN_LEN / sizeof(u32);
783
784 return skb;
785
786 free_skb:
787 consume_skb(skb);
788 return NULL;
789 }
790
791 /*
792 * Send an ICMP message in response to a situation
793 *
794 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
795 * MAY send more (we do).
796 * MUST NOT change this header information.
797 * MUST NOT reply to a multicast/broadcast IP address.
798 * MUST NOT reply to a multicast/broadcast MAC address.
799 * MUST reply to only the first fragment.
800 */
801
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct inet_skb_parm * parm)802 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
803 const struct inet_skb_parm *parm)
804 {
805 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
806 IP_OPTIONS_DATA_FIXED_SIZE);
807 struct iphdr *iph;
808 int room;
809 struct rtable *rt = skb_rtable(skb_in);
810 bool apply_ratelimit = false;
811 struct sk_buff *ext_skb;
812 struct ipcm_cookie ipc;
813 struct flowi4 fl4;
814 __be32 saddr;
815 u8 tos;
816 u32 mark;
817 struct net *net;
818 struct sock *sk;
819
820 if (!rt)
821 return;
822
823 rcu_read_lock();
824
825 if (rt->dst.dev)
826 net = dev_net_rcu(rt->dst.dev);
827 else if (skb_in->dev)
828 net = dev_net_rcu(skb_in->dev);
829 else
830 goto out;
831
832 /*
833 * Find the original header. It is expected to be valid, of course.
834 * Check this, icmp_send is called from the most obscure devices
835 * sometimes.
836 */
837 iph = ip_hdr(skb_in);
838
839 if ((u8 *)iph < skb_in->head ||
840 (skb_network_header(skb_in) + sizeof(*iph)) >
841 skb_tail_pointer(skb_in))
842 goto out;
843
844 /*
845 * No replies to physical multicast/broadcast
846 */
847 if (skb_in->pkt_type != PACKET_HOST)
848 goto out;
849
850 /*
851 * Now check at the protocol level
852 */
853 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
854 goto out;
855
856 /*
857 * Only reply to fragment 0. We byte re-order the constant
858 * mask for efficiency.
859 */
860 if (iph->frag_off & htons(IP_OFFSET))
861 goto out;
862
863 /*
864 * If we send an ICMP error to an ICMP error a mess would result..
865 */
866 if (icmp_pointers[type].error) {
867 /*
868 * We are an error, check if we are replying to an
869 * ICMP error
870 */
871 if (iph->protocol == IPPROTO_ICMP) {
872 u8 _inner_type, *itp;
873
874 itp = skb_header_pointer(skb_in,
875 skb_network_header(skb_in) +
876 (iph->ihl << 2) +
877 offsetof(struct icmphdr,
878 type) -
879 skb_in->data,
880 sizeof(_inner_type),
881 &_inner_type);
882 if (!itp)
883 goto out;
884
885 /*
886 * Assume any unknown ICMP type is an error. This
887 * isn't specified by the RFC, but think about it..
888 */
889 if (*itp > NR_ICMP_TYPES ||
890 icmp_pointers[*itp].error)
891 goto out;
892 }
893 }
894
895 /* Needed by both icmpv4_global_allow and icmp_xmit_lock */
896 local_bh_disable();
897
898 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
899 * incoming dev is loopback. If outgoing dev change to not be
900 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
901 */
902 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
903 !icmpv4_global_allow(net, type, code, &apply_ratelimit))
904 goto out_bh_enable;
905
906 sk = icmp_xmit_lock(net);
907 if (!sk)
908 goto out_bh_enable;
909
910 /*
911 * Construct source address and options.
912 */
913
914 saddr = iph->daddr;
915 if (!(rt->rt_flags & RTCF_LOCAL)) {
916 struct net_device *dev = NULL;
917
918 rcu_read_lock();
919 if (rt_is_input_route(rt) &&
920 READ_ONCE(net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr))
921 dev = dev_get_by_index_rcu(net, parm->iif ? parm->iif :
922 inet_iif(skb_in));
923
924 if (dev)
925 saddr = inet_select_addr(dev, iph->saddr,
926 RT_SCOPE_LINK);
927 else
928 saddr = 0;
929 rcu_read_unlock();
930 }
931
932 tos = icmp_pointers[type].error ? (RT_TOS(iph->tos) |
933 IPTOS_PREC_INTERNETCONTROL) :
934 iph->tos;
935 mark = IP4_REPLY_MARK(net, skb_in->mark);
936
937 if (__ip_options_echo(net, &icmp_param->replyopts.opt, skb_in,
938 &parm->opt))
939 goto out_unlock;
940
941
942 /*
943 * Prepare data for ICMP header.
944 */
945
946 icmp_param->data.icmph.type = type;
947 icmp_param->data.icmph.code = code;
948 icmp_param->data.icmph.un.gateway = info;
949 icmp_param->data.icmph.checksum = 0;
950 icmp_param->skb = skb_in;
951 icmp_param->offset = skb_network_offset(skb_in);
952 ipcm_init(&ipc);
953 ipc.tos = tos;
954 ipc.addr = iph->saddr;
955 ipc.opt = &icmp_param->replyopts;
956 ipc.sockc.mark = mark;
957
958 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr,
959 inet_dsfield_to_dscp(tos), mark, type, code,
960 icmp_param);
961 if (IS_ERR(rt))
962 goto out_unlock;
963
964 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
965 goto ende;
966
967 /* peer icmp_ratelimit */
968 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
969 goto ende;
970
971 /* RFC says return as much as we can without exceeding 576 bytes. */
972
973 room = dst4_mtu(&rt->dst);
974 if (room > 576)
975 room = 576;
976 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.optlen;
977 room -= sizeof(struct icmphdr);
978 /* Guard against tiny mtu. We need to include at least one
979 * IP network header for this message to make any sense.
980 */
981 if (room <= (int)sizeof(struct iphdr))
982 goto ende;
983
984 ext_skb = icmp_ext_append(net, skb_in, &icmp_param->data.icmph, room,
985 parm->iif);
986 if (ext_skb)
987 icmp_param->skb = ext_skb;
988
989 icmp_param->data_len = icmp_param->skb->len - icmp_param->offset;
990 if (icmp_param->data_len > room)
991 icmp_param->data_len = room;
992 icmp_param->head_len = sizeof(struct icmphdr);
993
994 /* if we don't have a source address at this point, fall back to the
995 * dummy address instead of sending out a packet with a source address
996 * of 0.0.0.0
997 */
998 if (!fl4.saddr)
999 fl4.saddr = htonl(INADDR_DUMMY);
1000
1001 trace_icmp_send(skb_in, type, code);
1002
1003 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
1004
1005 if (ext_skb)
1006 consume_skb(ext_skb);
1007 ende:
1008 ip_rt_put(rt);
1009 out_unlock:
1010 icmp_xmit_unlock(sk);
1011 out_bh_enable:
1012 local_bh_enable();
1013 out:
1014 rcu_read_unlock();
1015 }
1016 EXPORT_SYMBOL(__icmp_send);
1017
1018 #if IS_ENABLED(CONFIG_NF_NAT)
1019 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)1020 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
1021 {
1022 struct sk_buff *cloned_skb = NULL;
1023 enum ip_conntrack_info ctinfo;
1024 enum ip_conntrack_dir dir;
1025 struct inet_skb_parm parm;
1026 struct nf_conn *ct;
1027 __be32 orig_ip;
1028
1029 memset(&parm, 0, sizeof(parm));
1030 ct = nf_ct_get(skb_in, &ctinfo);
1031 if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
1032 __icmp_send(skb_in, type, code, info, &parm);
1033 return;
1034 }
1035
1036 if (skb_shared(skb_in))
1037 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
1038
1039 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
1040 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
1041 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
1042 skb_network_offset(skb_in) + sizeof(struct iphdr))))
1043 goto out;
1044
1045 orig_ip = ip_hdr(skb_in)->saddr;
1046 dir = CTINFO2DIR(ctinfo);
1047 ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
1048 __icmp_send(skb_in, type, code, info, &parm);
1049 ip_hdr(skb_in)->saddr = orig_ip;
1050 out:
1051 consume_skb(cloned_skb);
1052 }
1053 EXPORT_SYMBOL(icmp_ndo_send);
1054 #endif
1055
icmp_socket_deliver(struct sk_buff * skb,u32 info)1056 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
1057 {
1058 const struct iphdr *iph = (const struct iphdr *)skb->data;
1059 const struct net_protocol *ipprot;
1060 int protocol = iph->protocol;
1061
1062 /* Checkin full IP header plus 8 bytes of protocol to
1063 * avoid additional coding at protocol handlers.
1064 */
1065 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
1066 goto out;
1067
1068 /* IPPROTO_RAW sockets are not supposed to receive anything. */
1069 if (protocol == IPPROTO_RAW)
1070 goto out;
1071
1072 raw_icmp_error(skb, protocol, info);
1073
1074 ipprot = rcu_dereference(inet_protos[protocol]);
1075 if (ipprot && ipprot->err_handler)
1076 ipprot->err_handler(skb, info);
1077 return;
1078
1079 out:
1080 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1081 }
1082
icmp_tag_validation(int proto)1083 static bool icmp_tag_validation(int proto)
1084 {
1085 const struct net_protocol *ipprot;
1086 bool ok;
1087
1088 rcu_read_lock();
1089 ipprot = rcu_dereference(inet_protos[proto]);
1090 ok = ipprot ? ipprot->icmp_strict_tag_validation : false;
1091 rcu_read_unlock();
1092 return ok;
1093 }
1094
1095 /*
1096 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
1097 * ICMP_PARAMETERPROB.
1098 */
1099
icmp_unreach(struct sk_buff * skb)1100 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
1101 {
1102 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
1103 const struct iphdr *iph;
1104 struct icmphdr *icmph;
1105 struct net *net;
1106 u32 info = 0;
1107
1108 net = skb_dst_dev_net_rcu(skb);
1109
1110 /*
1111 * Incomplete header ?
1112 * Only checks for the IP header, there should be an
1113 * additional check for longer headers in upper levels.
1114 */
1115
1116 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1117 goto out_err;
1118
1119 icmph = icmp_hdr(skb);
1120 iph = (const struct iphdr *)skb->data;
1121
1122 if (iph->ihl < 5) { /* Mangled header, drop. */
1123 reason = SKB_DROP_REASON_IP_INHDR;
1124 goto out_err;
1125 }
1126
1127 switch (icmph->type) {
1128 case ICMP_DEST_UNREACH:
1129 switch (icmph->code & 15) {
1130 case ICMP_NET_UNREACH:
1131 case ICMP_HOST_UNREACH:
1132 case ICMP_PROT_UNREACH:
1133 case ICMP_PORT_UNREACH:
1134 break;
1135 case ICMP_FRAG_NEEDED:
1136 /* for documentation of the ip_no_pmtu_disc
1137 * values please see
1138 * Documentation/networking/ip-sysctl.rst
1139 */
1140 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
1141 default:
1142 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
1143 &iph->daddr);
1144 break;
1145 case 2:
1146 goto out;
1147 case 3:
1148 if (!icmp_tag_validation(iph->protocol))
1149 goto out;
1150 fallthrough;
1151 case 0:
1152 info = ntohs(icmph->un.frag.mtu);
1153 }
1154 break;
1155 case ICMP_SR_FAILED:
1156 net_dbg_ratelimited("%pI4: Source Route Failed\n",
1157 &iph->daddr);
1158 break;
1159 default:
1160 break;
1161 }
1162 if (icmph->code > NR_ICMP_UNREACH)
1163 goto out;
1164 break;
1165 case ICMP_PARAMETERPROB:
1166 info = ntohl(icmph->un.gateway) >> 24;
1167 break;
1168 case ICMP_TIME_EXCEEDED:
1169 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
1170 if (icmph->code == ICMP_EXC_FRAGTIME)
1171 goto out;
1172 break;
1173 }
1174
1175 /*
1176 * Throw it at our lower layers
1177 *
1178 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
1179 * header.
1180 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
1181 * transport layer.
1182 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
1183 * transport layer.
1184 */
1185
1186 /*
1187 * Check the other end isn't violating RFC 1122. Some routers send
1188 * bogus responses to broadcast frames. If you see this message
1189 * first check your netmask matches at both ends, if it does then
1190 * get the other vendor to fix their kit.
1191 */
1192
1193 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
1194 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
1195 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
1196 &ip_hdr(skb)->saddr,
1197 icmph->type, icmph->code,
1198 &iph->daddr, skb->dev->name);
1199 goto out;
1200 }
1201
1202 icmp_socket_deliver(skb, info);
1203
1204 out:
1205 return reason;
1206 out_err:
1207 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1208 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
1209 }
1210
1211
1212 /*
1213 * Handle ICMP_REDIRECT.
1214 */
1215
icmp_redirect(struct sk_buff * skb)1216 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
1217 {
1218 if (skb->len < sizeof(struct iphdr)) {
1219 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1220 return SKB_DROP_REASON_PKT_TOO_SMALL;
1221 }
1222
1223 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
1224 /* there aught to be a stat */
1225 return SKB_DROP_REASON_NOMEM;
1226 }
1227
1228 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1229 return SKB_NOT_DROPPED_YET;
1230 }
1231
1232 /*
1233 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1234 *
1235 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1236 * requests.
1237 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1238 * included in the reply.
1239 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1240 * echo requests, MUST have default=NOT.
1241 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
1242 * Extended Echo Functionality, MUST be disabled by default
1243 * See also WRT handling of options once they are done and working.
1244 */
1245
icmp_echo(struct sk_buff * skb)1246 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1247 {
1248 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1249 IP_OPTIONS_DATA_FIXED_SIZE);
1250 struct net *net;
1251
1252 net = skb_dst_dev_net_rcu(skb);
1253 /* should there be an ICMP stat for ignored echos? */
1254 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1255 return SKB_NOT_DROPPED_YET;
1256
1257 icmp_param->data.icmph = *icmp_hdr(skb);
1258 icmp_param->skb = skb;
1259 icmp_param->offset = 0;
1260 icmp_param->data_len = skb->len;
1261 icmp_param->head_len = sizeof(struct icmphdr);
1262
1263 if (icmp_param->data.icmph.type == ICMP_ECHO)
1264 icmp_param->data.icmph.type = ICMP_ECHOREPLY;
1265 else if (!icmp_build_probe(skb, &icmp_param->data.icmph))
1266 return SKB_NOT_DROPPED_YET;
1267
1268 icmp_reply(icmp_param, skb);
1269 return SKB_NOT_DROPPED_YET;
1270 }
1271
1272 /* Helper for icmp_echo and icmpv6_echo_reply.
1273 * Searches for net_device that matches PROBE interface identifier
1274 * and builds PROBE reply message in icmphdr.
1275 *
1276 * Returns false if PROBE responses are disabled via sysctl
1277 */
1278
icmp_build_probe(struct sk_buff * skb,struct icmphdr * icmphdr)1279 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1280 {
1281 struct net *net = dev_net_rcu(skb->dev);
1282 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1283 struct icmp_ext_echo_iio *iio, _iio;
1284 struct inet6_dev *in6_dev;
1285 struct in_device *in_dev;
1286 struct net_device *dev;
1287 char buff[IFNAMSIZ];
1288 u16 ident_len;
1289 u8 status;
1290
1291 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1292 return false;
1293
1294 /* We currently only support probing interfaces on the proxy node
1295 * Check to ensure L-bit is set
1296 */
1297 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1298 return false;
1299 /* Clear status bits in reply message */
1300 icmphdr->un.echo.sequence &= htons(0xFF00);
1301 if (icmphdr->type == ICMP_EXT_ECHO)
1302 icmphdr->type = ICMP_EXT_ECHOREPLY;
1303 else
1304 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1305 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1306 /* Size of iio is class_type dependent.
1307 * Only check header here and assign length based on ctype in the switch statement
1308 */
1309 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1310 if (!ext_hdr || !iio)
1311 goto send_mal_query;
1312 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1313 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1314 goto send_mal_query;
1315 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1316 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1317 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1318 if (!iio)
1319 goto send_mal_query;
1320
1321 status = 0;
1322 dev = NULL;
1323 switch (iio->extobj_hdr.class_type) {
1324 case ICMP_EXT_ECHO_CTYPE_NAME:
1325 if (ident_len >= IFNAMSIZ)
1326 goto send_mal_query;
1327 memset(buff, 0, sizeof(buff));
1328 memcpy(buff, &iio->ident.name, ident_len);
1329 dev = dev_get_by_name(net, buff);
1330 break;
1331 case ICMP_EXT_ECHO_CTYPE_INDEX:
1332 if (ident_len != sizeof(iio->ident.ifindex))
1333 goto send_mal_query;
1334 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1335 break;
1336 case ICMP_EXT_ECHO_CTYPE_ADDR:
1337 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1338 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1339 iio->ident.addr.ctype3_hdr.addrlen)
1340 goto send_mal_query;
1341 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1342 case ICMP_AFI_IP:
1343 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1344 goto send_mal_query;
1345 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1346 break;
1347 #if IS_ENABLED(CONFIG_IPV6)
1348 case ICMP_AFI_IP6:
1349 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1350 goto send_mal_query;
1351 dev = ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1352 dev_hold(dev);
1353 break;
1354 #endif
1355 default:
1356 goto send_mal_query;
1357 }
1358 break;
1359 default:
1360 goto send_mal_query;
1361 }
1362 if (!dev) {
1363 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1364 return true;
1365 }
1366 /* Fill bits in reply message */
1367 if (dev->flags & IFF_UP)
1368 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1369
1370 in_dev = __in_dev_get_rcu(dev);
1371 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1372 status |= ICMP_EXT_ECHOREPLY_IPV4;
1373
1374 in6_dev = __in6_dev_get(dev);
1375 if (in6_dev && !list_empty(&in6_dev->addr_list))
1376 status |= ICMP_EXT_ECHOREPLY_IPV6;
1377
1378 dev_put(dev);
1379 icmphdr->un.echo.sequence |= htons(status);
1380 return true;
1381 send_mal_query:
1382 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1383 return true;
1384 }
1385
1386 /*
1387 * Handle ICMP Timestamp requests.
1388 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1389 * SHOULD be in the kernel for minimum random latency.
1390 * MUST be accurate to a few minutes.
1391 * MUST be updated at least at 15Hz.
1392 */
icmp_timestamp(struct sk_buff * skb)1393 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1394 {
1395 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1396 IP_OPTIONS_DATA_FIXED_SIZE);
1397 /*
1398 * Too short.
1399 */
1400 if (skb->len < 4)
1401 goto out_err;
1402
1403 /*
1404 * Fill in the current time as ms since midnight UT:
1405 */
1406 icmp_param->data.times[1] = inet_current_timestamp();
1407 icmp_param->data.times[2] = icmp_param->data.times[1];
1408
1409 BUG_ON(skb_copy_bits(skb, 0, &icmp_param->data.times[0], 4));
1410
1411 icmp_param->data.icmph = *icmp_hdr(skb);
1412 icmp_param->data.icmph.type = ICMP_TIMESTAMPREPLY;
1413 icmp_param->data.icmph.code = 0;
1414 icmp_param->skb = skb;
1415 icmp_param->offset = 0;
1416 icmp_param->data_len = 0;
1417 icmp_param->head_len = sizeof(struct icmphdr) + 12;
1418 icmp_reply(icmp_param, skb);
1419 return SKB_NOT_DROPPED_YET;
1420
1421 out_err:
1422 __ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1423 return SKB_DROP_REASON_PKT_TOO_SMALL;
1424 }
1425
icmp_discard(struct sk_buff * skb)1426 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1427 {
1428 /* pretend it was a success */
1429 return SKB_NOT_DROPPED_YET;
1430 }
1431
1432 /*
1433 * Deal with incoming ICMP packets.
1434 */
icmp_rcv(struct sk_buff * skb)1435 int icmp_rcv(struct sk_buff *skb)
1436 {
1437 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1438 struct rtable *rt = skb_rtable(skb);
1439 struct net *net = dev_net_rcu(rt->dst.dev);
1440 struct icmphdr *icmph;
1441
1442 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1443 struct sec_path *sp = skb_sec_path(skb);
1444 int nh;
1445
1446 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1447 XFRM_STATE_ICMP)) {
1448 reason = SKB_DROP_REASON_XFRM_POLICY;
1449 goto drop;
1450 }
1451
1452 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1453 goto drop;
1454
1455 nh = skb_network_offset(skb);
1456 skb_set_network_header(skb, sizeof(*icmph));
1457
1458 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1459 skb)) {
1460 reason = SKB_DROP_REASON_XFRM_POLICY;
1461 goto drop;
1462 }
1463
1464 skb_set_network_header(skb, nh);
1465 }
1466
1467 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1468
1469 if (skb_checksum_simple_validate(skb))
1470 goto csum_error;
1471
1472 if (!pskb_pull(skb, sizeof(*icmph)))
1473 goto error;
1474
1475 icmph = icmp_hdr(skb);
1476
1477 ICMPMSGIN_INC_STATS(net, icmph->type);
1478
1479 /* Check for ICMP Extended Echo (PROBE) messages */
1480 if (icmph->type == ICMP_EXT_ECHO) {
1481 /* We can't use icmp_pointers[].handler() because it is an array of
1482 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1483 */
1484 reason = icmp_echo(skb);
1485 goto reason_check;
1486 }
1487
1488 /*
1489 * Parse the ICMP message
1490 */
1491
1492 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1493 /*
1494 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1495 * silently ignored (we let user decide with a sysctl).
1496 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1497 * discarded if to broadcast/multicast.
1498 */
1499 if ((icmph->type == ICMP_ECHO ||
1500 icmph->type == ICMP_TIMESTAMP) &&
1501 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1502 reason = SKB_DROP_REASON_INVALID_PROTO;
1503 goto error;
1504 }
1505 if (icmph->type != ICMP_ECHO &&
1506 icmph->type != ICMP_TIMESTAMP &&
1507 icmph->type != ICMP_ADDRESS &&
1508 icmph->type != ICMP_ADDRESSREPLY) {
1509 reason = SKB_DROP_REASON_INVALID_PROTO;
1510 goto error;
1511 }
1512 }
1513
1514 if (icmph->type == ICMP_EXT_ECHOREPLY ||
1515 icmph->type == ICMP_ECHOREPLY) {
1516 reason = ping_rcv(skb);
1517 return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1518 }
1519
1520 /*
1521 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1522 *
1523 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1524 * discarded.
1525 */
1526 if (icmph->type > NR_ICMP_TYPES) {
1527 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1528 goto error;
1529 }
1530
1531 reason = icmp_pointers[icmph->type].handler(skb);
1532 reason_check:
1533 if (!reason) {
1534 consume_skb(skb);
1535 return NET_RX_SUCCESS;
1536 }
1537
1538 drop:
1539 kfree_skb_reason(skb, reason);
1540 return NET_RX_DROP;
1541 csum_error:
1542 reason = SKB_DROP_REASON_ICMP_CSUM;
1543 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1544 error:
1545 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1546 goto drop;
1547 }
1548
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1549 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1550 {
1551 struct icmp_extobj_hdr *objh, _objh;
1552 struct icmp_ext_hdr *exth, _exth;
1553 u16 olen;
1554
1555 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1556 if (!exth)
1557 return false;
1558 if (exth->version != 2)
1559 return true;
1560
1561 if (exth->checksum &&
1562 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1563 return false;
1564
1565 off += sizeof(_exth);
1566 while (off < skb->len) {
1567 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1568 if (!objh)
1569 return false;
1570
1571 olen = ntohs(objh->length);
1572 if (olen < sizeof(_objh))
1573 return false;
1574
1575 off += olen;
1576 if (off > skb->len)
1577 return false;
1578 }
1579
1580 return true;
1581 }
1582
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1583 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1584 struct sock_ee_data_rfc4884 *out,
1585 int thlen, int off)
1586 {
1587 int hlen;
1588
1589 /* original datagram headers: end of icmph to payload (skb->data) */
1590 hlen = -skb_transport_offset(skb) - thlen;
1591
1592 /* per rfc 4884: minimal datagram length of 128 bytes */
1593 if (off < 128 || off < hlen)
1594 return;
1595
1596 /* kernel has stripped headers: return payload offset in bytes */
1597 off -= hlen;
1598 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1599 return;
1600
1601 out->len = off;
1602
1603 if (!ip_icmp_error_rfc4884_validate(skb, off))
1604 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1605 }
1606
icmp_err(struct sk_buff * skb,u32 info)1607 int icmp_err(struct sk_buff *skb, u32 info)
1608 {
1609 struct iphdr *iph = (struct iphdr *)skb->data;
1610 int offset = iph->ihl<<2;
1611 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1612 struct net *net = dev_net_rcu(skb->dev);
1613 int type = icmp_hdr(skb)->type;
1614 int code = icmp_hdr(skb)->code;
1615
1616 /*
1617 * Use ping_err to handle all icmp errors except those
1618 * triggered by ICMP_ECHOREPLY which sent from kernel.
1619 */
1620 if (icmph->type != ICMP_ECHOREPLY) {
1621 ping_err(skb, offset, info);
1622 return 0;
1623 }
1624
1625 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1626 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1627 else if (type == ICMP_REDIRECT)
1628 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1629
1630 return 0;
1631 }
1632
1633 /*
1634 * This table is the definition of how we handle ICMP.
1635 */
1636 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1637 [ICMP_ECHOREPLY] = {
1638 .handler = ping_rcv,
1639 },
1640 [1] = {
1641 .handler = icmp_discard,
1642 .error = 1,
1643 },
1644 [2] = {
1645 .handler = icmp_discard,
1646 .error = 1,
1647 },
1648 [ICMP_DEST_UNREACH] = {
1649 .handler = icmp_unreach,
1650 .error = 1,
1651 },
1652 [ICMP_SOURCE_QUENCH] = {
1653 .handler = icmp_unreach,
1654 .error = 1,
1655 },
1656 [ICMP_REDIRECT] = {
1657 .handler = icmp_redirect,
1658 .error = 1,
1659 },
1660 [6] = {
1661 .handler = icmp_discard,
1662 .error = 1,
1663 },
1664 [7] = {
1665 .handler = icmp_discard,
1666 .error = 1,
1667 },
1668 [ICMP_ECHO] = {
1669 .handler = icmp_echo,
1670 },
1671 [9] = {
1672 .handler = icmp_discard,
1673 .error = 1,
1674 },
1675 [10] = {
1676 .handler = icmp_discard,
1677 .error = 1,
1678 },
1679 [ICMP_TIME_EXCEEDED] = {
1680 .handler = icmp_unreach,
1681 .error = 1,
1682 },
1683 [ICMP_PARAMETERPROB] = {
1684 .handler = icmp_unreach,
1685 .error = 1,
1686 },
1687 [ICMP_TIMESTAMP] = {
1688 .handler = icmp_timestamp,
1689 },
1690 [ICMP_TIMESTAMPREPLY] = {
1691 .handler = icmp_discard,
1692 },
1693 [ICMP_INFO_REQUEST] = {
1694 .handler = icmp_discard,
1695 },
1696 [ICMP_INFO_REPLY] = {
1697 .handler = icmp_discard,
1698 },
1699 [ICMP_ADDRESS] = {
1700 .handler = icmp_discard,
1701 },
1702 [ICMP_ADDRESSREPLY] = {
1703 .handler = icmp_discard,
1704 },
1705 };
1706
icmp_sk_init(struct net * net)1707 static int __net_init icmp_sk_init(struct net *net)
1708 {
1709 /* Control parameters for ECHO replies. */
1710 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1711 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1712 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1713
1714 /* Control parameter - ignore bogus broadcast responses? */
1715 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1716
1717 /*
1718 * Configurable global rate limit.
1719 *
1720 * ratelimit defines tokens/packet consumed for dst->rate_token
1721 * bucket ratemask defines which icmp types are ratelimited by
1722 * setting it's bit position.
1723 *
1724 * default:
1725 * dest unreachable (3), source quench (4),
1726 * time exceeded (11), parameter problem (12)
1727 */
1728
1729 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1730 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1731 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1732 net->ipv4.sysctl_icmp_errors_extension_mask = 0;
1733 net->ipv4.sysctl_icmp_msgs_per_sec = 10000;
1734 net->ipv4.sysctl_icmp_msgs_burst = 10000;
1735
1736 return 0;
1737 }
1738
1739 static struct pernet_operations __net_initdata icmp_sk_ops = {
1740 .init = icmp_sk_init,
1741 };
1742
icmp_init(void)1743 int __init icmp_init(void)
1744 {
1745 int err, i;
1746
1747 for_each_possible_cpu(i) {
1748 struct sock *sk;
1749
1750 err = inet_ctl_sock_create(&sk, PF_INET,
1751 SOCK_RAW, IPPROTO_ICMP, &init_net);
1752 if (err < 0)
1753 return err;
1754
1755 per_cpu(ipv4_icmp_sk, i) = sk;
1756
1757 /* Enough space for 2 64K ICMP packets, including
1758 * sk_buff/skb_shared_info struct overhead.
1759 */
1760 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1761
1762 /*
1763 * Speedup sock_wfree()
1764 */
1765 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1766 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1767 }
1768 return register_pernet_subsys(&icmp_sk_ops);
1769 }
1770