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 /* peer icmp_ratelimit */
965 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code, apply_ratelimit))
966 goto ende;
967
968 /* RFC says return as much as we can without exceeding 576 bytes. */
969
970 room = dst4_mtu(&rt->dst);
971 if (room > 576)
972 room = 576;
973 room -= sizeof(struct iphdr) + icmp_param->replyopts.opt.optlen;
974 room -= sizeof(struct icmphdr);
975 /* Guard against tiny mtu. We need to include at least one
976 * IP network header for this message to make any sense.
977 */
978 if (room <= (int)sizeof(struct iphdr))
979 goto ende;
980
981 ext_skb = icmp_ext_append(net, skb_in, &icmp_param->data.icmph, room,
982 parm->iif);
983 if (ext_skb)
984 icmp_param->skb = ext_skb;
985
986 icmp_param->data_len = icmp_param->skb->len - icmp_param->offset;
987 if (icmp_param->data_len > room)
988 icmp_param->data_len = room;
989 icmp_param->head_len = sizeof(struct icmphdr);
990
991 /* if we don't have a source address at this point, fall back to the
992 * dummy address instead of sending out a packet with a source address
993 * of 0.0.0.0
994 */
995 if (!fl4.saddr)
996 fl4.saddr = htonl(INADDR_DUMMY);
997
998 trace_icmp_send(skb_in, type, code);
999
1000 icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
1001
1002 if (ext_skb)
1003 consume_skb(ext_skb);
1004 ende:
1005 ip_rt_put(rt);
1006 out_unlock:
1007 icmp_xmit_unlock(sk);
1008 out_bh_enable:
1009 local_bh_enable();
1010 out:
1011 rcu_read_unlock();
1012 }
1013 EXPORT_SYMBOL(__icmp_send);
1014
1015 #if IS_ENABLED(CONFIG_NF_NAT)
1016 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)1017 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
1018 {
1019 struct sk_buff *cloned_skb = NULL;
1020 enum ip_conntrack_info ctinfo;
1021 enum ip_conntrack_dir dir;
1022 struct inet_skb_parm parm;
1023 struct nf_conn *ct;
1024 __be32 orig_ip;
1025
1026 memset(&parm, 0, sizeof(parm));
1027 ct = nf_ct_get(skb_in, &ctinfo);
1028 if (!ct || !(READ_ONCE(ct->status) & IPS_NAT_MASK)) {
1029 __icmp_send(skb_in, type, code, info, &parm);
1030 return;
1031 }
1032
1033 if (skb_shared(skb_in))
1034 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
1035
1036 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
1037 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
1038 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
1039 skb_network_offset(skb_in) + sizeof(struct iphdr))))
1040 goto out;
1041
1042 orig_ip = ip_hdr(skb_in)->saddr;
1043 dir = CTINFO2DIR(ctinfo);
1044 ip_hdr(skb_in)->saddr = ct->tuplehash[dir].tuple.src.u3.ip;
1045 __icmp_send(skb_in, type, code, info, &parm);
1046 ip_hdr(skb_in)->saddr = orig_ip;
1047 out:
1048 consume_skb(cloned_skb);
1049 }
1050 EXPORT_SYMBOL(icmp_ndo_send);
1051 #endif
1052
icmp_socket_deliver(struct sk_buff * skb,u32 info)1053 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
1054 {
1055 const struct iphdr *iph = (const struct iphdr *)skb->data;
1056 const struct net_protocol *ipprot;
1057 int protocol = iph->protocol;
1058
1059 /* Checkin full IP header plus 8 bytes of protocol to
1060 * avoid additional coding at protocol handlers.
1061 */
1062 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
1063 goto out;
1064
1065 /* IPPROTO_RAW sockets are not supposed to receive anything. */
1066 if (protocol == IPPROTO_RAW)
1067 goto out;
1068
1069 raw_icmp_error(skb, protocol, info);
1070
1071 ipprot = rcu_dereference(inet_protos[protocol]);
1072 if (ipprot && ipprot->err_handler)
1073 ipprot->err_handler(skb, info);
1074 return;
1075
1076 out:
1077 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1078 }
1079
icmp_tag_validation(int proto)1080 static bool icmp_tag_validation(int proto)
1081 {
1082 const struct net_protocol *ipprot;
1083 bool ok;
1084
1085 rcu_read_lock();
1086 ipprot = rcu_dereference(inet_protos[proto]);
1087 ok = ipprot ? ipprot->icmp_strict_tag_validation : false;
1088 rcu_read_unlock();
1089 return ok;
1090 }
1091
1092 /*
1093 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
1094 * ICMP_PARAMETERPROB.
1095 */
1096
icmp_unreach(struct sk_buff * skb)1097 static enum skb_drop_reason icmp_unreach(struct sk_buff *skb)
1098 {
1099 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
1100 const struct iphdr *iph;
1101 struct icmphdr *icmph;
1102 struct net *net;
1103 u32 info = 0;
1104
1105 net = skb_dst_dev_net_rcu(skb);
1106
1107 /*
1108 * Incomplete header ?
1109 * Only checks for the IP header, there should be an
1110 * additional check for longer headers in upper levels.
1111 */
1112
1113 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1114 goto out_err;
1115
1116 icmph = icmp_hdr(skb);
1117 iph = (const struct iphdr *)skb->data;
1118
1119 if (iph->ihl < 5) { /* Mangled header, drop. */
1120 reason = SKB_DROP_REASON_IP_INHDR;
1121 goto out_err;
1122 }
1123
1124 switch (icmph->type) {
1125 case ICMP_DEST_UNREACH:
1126 switch (icmph->code & 15) {
1127 case ICMP_NET_UNREACH:
1128 case ICMP_HOST_UNREACH:
1129 case ICMP_PROT_UNREACH:
1130 case ICMP_PORT_UNREACH:
1131 break;
1132 case ICMP_FRAG_NEEDED:
1133 /* for documentation of the ip_no_pmtu_disc
1134 * values please see
1135 * Documentation/networking/ip-sysctl.rst
1136 */
1137 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
1138 default:
1139 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
1140 &iph->daddr);
1141 break;
1142 case 2:
1143 goto out;
1144 case 3:
1145 if (!icmp_tag_validation(iph->protocol))
1146 goto out;
1147 fallthrough;
1148 case 0:
1149 info = ntohs(icmph->un.frag.mtu);
1150 }
1151 break;
1152 case ICMP_SR_FAILED:
1153 net_dbg_ratelimited("%pI4: Source Route Failed\n",
1154 &iph->daddr);
1155 break;
1156 default:
1157 break;
1158 }
1159 if (icmph->code > NR_ICMP_UNREACH)
1160 goto out;
1161 break;
1162 case ICMP_PARAMETERPROB:
1163 info = ntohl(icmph->un.gateway) >> 24;
1164 break;
1165 case ICMP_TIME_EXCEEDED:
1166 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
1167 if (icmph->code == ICMP_EXC_FRAGTIME)
1168 goto out;
1169 break;
1170 }
1171
1172 /*
1173 * Throw it at our lower layers
1174 *
1175 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
1176 * header.
1177 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
1178 * transport layer.
1179 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
1180 * transport layer.
1181 */
1182
1183 /*
1184 * Check the other end isn't violating RFC 1122. Some routers send
1185 * bogus responses to broadcast frames. If you see this message
1186 * first check your netmask matches at both ends, if it does then
1187 * get the other vendor to fix their kit.
1188 */
1189
1190 if (!READ_ONCE(net->ipv4.sysctl_icmp_ignore_bogus_error_responses) &&
1191 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
1192 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
1193 &ip_hdr(skb)->saddr,
1194 icmph->type, icmph->code,
1195 &iph->daddr, skb->dev->name);
1196 goto out;
1197 }
1198
1199 icmp_socket_deliver(skb, info);
1200
1201 out:
1202 return reason;
1203 out_err:
1204 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1205 return reason ?: SKB_DROP_REASON_NOT_SPECIFIED;
1206 }
1207
1208
1209 /*
1210 * Handle ICMP_REDIRECT.
1211 */
1212
icmp_redirect(struct sk_buff * skb)1213 static enum skb_drop_reason icmp_redirect(struct sk_buff *skb)
1214 {
1215 if (skb->len < sizeof(struct iphdr)) {
1216 __ICMP_INC_STATS(dev_net_rcu(skb->dev), ICMP_MIB_INERRORS);
1217 return SKB_DROP_REASON_PKT_TOO_SMALL;
1218 }
1219
1220 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
1221 /* there aught to be a stat */
1222 return SKB_DROP_REASON_NOMEM;
1223 }
1224
1225 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
1226 return SKB_NOT_DROPPED_YET;
1227 }
1228
1229 /*
1230 * Handle ICMP_ECHO ("ping") and ICMP_EXT_ECHO ("PROBE") requests.
1231 *
1232 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
1233 * requests.
1234 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
1235 * included in the reply.
1236 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
1237 * echo requests, MUST have default=NOT.
1238 * RFC 8335: 8 MUST have a config option to enable/disable ICMP
1239 * Extended Echo Functionality, MUST be disabled by default
1240 * See also WRT handling of options once they are done and working.
1241 */
1242
icmp_echo(struct sk_buff * skb)1243 static enum skb_drop_reason icmp_echo(struct sk_buff *skb)
1244 {
1245 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1246 IP_OPTIONS_DATA_FIXED_SIZE);
1247 struct net *net;
1248
1249 net = skb_dst_dev_net_rcu(skb);
1250 /* should there be an ICMP stat for ignored echos? */
1251 if (READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_all))
1252 return SKB_NOT_DROPPED_YET;
1253
1254 icmp_param->data.icmph = *icmp_hdr(skb);
1255 icmp_param->skb = skb;
1256 icmp_param->offset = 0;
1257 icmp_param->data_len = skb->len;
1258 icmp_param->head_len = sizeof(struct icmphdr);
1259
1260 if (icmp_param->data.icmph.type == ICMP_ECHO)
1261 icmp_param->data.icmph.type = ICMP_ECHOREPLY;
1262 else if (!icmp_build_probe(skb, &icmp_param->data.icmph))
1263 return SKB_NOT_DROPPED_YET;
1264
1265 icmp_reply(icmp_param, skb);
1266 return SKB_NOT_DROPPED_YET;
1267 }
1268
1269 /* Helper for icmp_echo and icmpv6_echo_reply.
1270 * Searches for net_device that matches PROBE interface identifier
1271 * and builds PROBE reply message in icmphdr.
1272 *
1273 * Returns false if PROBE responses are disabled via sysctl
1274 */
1275
icmp_build_probe(struct sk_buff * skb,struct icmphdr * icmphdr)1276 bool icmp_build_probe(struct sk_buff *skb, struct icmphdr *icmphdr)
1277 {
1278 struct net *net = dev_net_rcu(skb->dev);
1279 struct icmp_ext_hdr *ext_hdr, _ext_hdr;
1280 struct icmp_ext_echo_iio *iio, _iio;
1281 struct inet6_dev *in6_dev;
1282 struct in_device *in_dev;
1283 struct net_device *dev;
1284 char buff[IFNAMSIZ];
1285 u16 ident_len;
1286 u8 status;
1287
1288 if (!READ_ONCE(net->ipv4.sysctl_icmp_echo_enable_probe))
1289 return false;
1290
1291 /* We currently only support probing interfaces on the proxy node
1292 * Check to ensure L-bit is set
1293 */
1294 if (!(ntohs(icmphdr->un.echo.sequence) & 1))
1295 return false;
1296 /* Clear status bits in reply message */
1297 icmphdr->un.echo.sequence &= htons(0xFF00);
1298 if (icmphdr->type == ICMP_EXT_ECHO)
1299 icmphdr->type = ICMP_EXT_ECHOREPLY;
1300 else
1301 icmphdr->type = ICMPV6_EXT_ECHO_REPLY;
1302 ext_hdr = skb_header_pointer(skb, 0, sizeof(_ext_hdr), &_ext_hdr);
1303 /* Size of iio is class_type dependent.
1304 * Only check header here and assign length based on ctype in the switch statement
1305 */
1306 iio = skb_header_pointer(skb, sizeof(_ext_hdr), sizeof(iio->extobj_hdr), &_iio);
1307 if (!ext_hdr || !iio)
1308 goto send_mal_query;
1309 if (ntohs(iio->extobj_hdr.length) <= sizeof(iio->extobj_hdr) ||
1310 ntohs(iio->extobj_hdr.length) > sizeof(_iio))
1311 goto send_mal_query;
1312 ident_len = ntohs(iio->extobj_hdr.length) - sizeof(iio->extobj_hdr);
1313 iio = skb_header_pointer(skb, sizeof(_ext_hdr),
1314 sizeof(iio->extobj_hdr) + ident_len, &_iio);
1315 if (!iio)
1316 goto send_mal_query;
1317
1318 status = 0;
1319 dev = NULL;
1320 switch (iio->extobj_hdr.class_type) {
1321 case ICMP_EXT_ECHO_CTYPE_NAME:
1322 if (ident_len >= IFNAMSIZ)
1323 goto send_mal_query;
1324 memset(buff, 0, sizeof(buff));
1325 memcpy(buff, &iio->ident.name, ident_len);
1326 dev = dev_get_by_name(net, buff);
1327 break;
1328 case ICMP_EXT_ECHO_CTYPE_INDEX:
1329 if (ident_len != sizeof(iio->ident.ifindex))
1330 goto send_mal_query;
1331 dev = dev_get_by_index(net, ntohl(iio->ident.ifindex));
1332 break;
1333 case ICMP_EXT_ECHO_CTYPE_ADDR:
1334 if (ident_len < sizeof(iio->ident.addr.ctype3_hdr) ||
1335 ident_len != sizeof(iio->ident.addr.ctype3_hdr) +
1336 iio->ident.addr.ctype3_hdr.addrlen)
1337 goto send_mal_query;
1338 switch (ntohs(iio->ident.addr.ctype3_hdr.afi)) {
1339 case ICMP_AFI_IP:
1340 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in_addr))
1341 goto send_mal_query;
1342 dev = ip_dev_find(net, iio->ident.addr.ip_addr.ipv4_addr);
1343 break;
1344 #if IS_ENABLED(CONFIG_IPV6)
1345 case ICMP_AFI_IP6:
1346 if (iio->ident.addr.ctype3_hdr.addrlen != sizeof(struct in6_addr))
1347 goto send_mal_query;
1348 dev = ipv6_dev_find(net, &iio->ident.addr.ip_addr.ipv6_addr, dev);
1349 dev_hold(dev);
1350 break;
1351 #endif
1352 default:
1353 goto send_mal_query;
1354 }
1355 break;
1356 default:
1357 goto send_mal_query;
1358 }
1359 if (!dev) {
1360 icmphdr->code = ICMP_EXT_CODE_NO_IF;
1361 return true;
1362 }
1363 /* Fill bits in reply message */
1364 if (dev->flags & IFF_UP)
1365 status |= ICMP_EXT_ECHOREPLY_ACTIVE;
1366
1367 in_dev = __in_dev_get_rcu(dev);
1368 if (in_dev && rcu_access_pointer(in_dev->ifa_list))
1369 status |= ICMP_EXT_ECHOREPLY_IPV4;
1370
1371 in6_dev = __in6_dev_get(dev);
1372 if (in6_dev && !list_empty(&in6_dev->addr_list))
1373 status |= ICMP_EXT_ECHOREPLY_IPV6;
1374
1375 dev_put(dev);
1376 icmphdr->un.echo.sequence |= htons(status);
1377 return true;
1378 send_mal_query:
1379 icmphdr->code = ICMP_EXT_CODE_MAL_QUERY;
1380 return true;
1381 }
1382
1383 /*
1384 * Handle ICMP Timestamp requests.
1385 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1386 * SHOULD be in the kernel for minimum random latency.
1387 * MUST be accurate to a few minutes.
1388 * MUST be updated at least at 15Hz.
1389 */
icmp_timestamp(struct sk_buff * skb)1390 static enum skb_drop_reason icmp_timestamp(struct sk_buff *skb)
1391 {
1392 DEFINE_RAW_FLEX(struct icmp_bxm, icmp_param, replyopts.opt.__data,
1393 IP_OPTIONS_DATA_FIXED_SIZE);
1394 /*
1395 * Too short.
1396 */
1397 if (skb->len < 4)
1398 goto out_err;
1399
1400 /*
1401 * Fill in the current time as ms since midnight UT:
1402 */
1403 icmp_param->data.times[1] = inet_current_timestamp();
1404 icmp_param->data.times[2] = icmp_param->data.times[1];
1405
1406 BUG_ON(skb_copy_bits(skb, 0, &icmp_param->data.times[0], 4));
1407
1408 icmp_param->data.icmph = *icmp_hdr(skb);
1409 icmp_param->data.icmph.type = ICMP_TIMESTAMPREPLY;
1410 icmp_param->data.icmph.code = 0;
1411 icmp_param->skb = skb;
1412 icmp_param->offset = 0;
1413 icmp_param->data_len = 0;
1414 icmp_param->head_len = sizeof(struct icmphdr) + 12;
1415 icmp_reply(icmp_param, skb);
1416 return SKB_NOT_DROPPED_YET;
1417
1418 out_err:
1419 __ICMP_INC_STATS(skb_dst_dev_net_rcu(skb), ICMP_MIB_INERRORS);
1420 return SKB_DROP_REASON_PKT_TOO_SMALL;
1421 }
1422
icmp_discard(struct sk_buff * skb)1423 static enum skb_drop_reason icmp_discard(struct sk_buff *skb)
1424 {
1425 /* pretend it was a success */
1426 return SKB_NOT_DROPPED_YET;
1427 }
1428
1429 /*
1430 * Deal with incoming ICMP packets.
1431 */
icmp_rcv(struct sk_buff * skb)1432 int icmp_rcv(struct sk_buff *skb)
1433 {
1434 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
1435 struct rtable *rt = skb_rtable(skb);
1436 struct net *net = dev_net_rcu(rt->dst.dev);
1437 struct icmphdr *icmph;
1438
1439 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1440 struct sec_path *sp = skb_sec_path(skb);
1441 int nh;
1442
1443 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1444 XFRM_STATE_ICMP)) {
1445 reason = SKB_DROP_REASON_XFRM_POLICY;
1446 goto drop;
1447 }
1448
1449 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1450 goto drop;
1451
1452 nh = skb_network_offset(skb);
1453 skb_set_network_header(skb, sizeof(*icmph));
1454
1455 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN,
1456 skb)) {
1457 reason = SKB_DROP_REASON_XFRM_POLICY;
1458 goto drop;
1459 }
1460
1461 skb_set_network_header(skb, nh);
1462 }
1463
1464 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1465
1466 if (skb_checksum_simple_validate(skb))
1467 goto csum_error;
1468
1469 if (!pskb_pull(skb, sizeof(*icmph)))
1470 goto error;
1471
1472 icmph = icmp_hdr(skb);
1473
1474 ICMPMSGIN_INC_STATS(net, icmph->type);
1475
1476 /* Check for ICMP Extended Echo (PROBE) messages */
1477 if (icmph->type == ICMP_EXT_ECHO) {
1478 /* We can't use icmp_pointers[].handler() because it is an array of
1479 * size NR_ICMP_TYPES + 1 (19 elements) and PROBE has code 42.
1480 */
1481 reason = icmp_echo(skb);
1482 goto reason_check;
1483 }
1484
1485 /*
1486 * Parse the ICMP message
1487 */
1488
1489 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1490 /*
1491 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1492 * silently ignored (we let user decide with a sysctl).
1493 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1494 * discarded if to broadcast/multicast.
1495 */
1496 if ((icmph->type == ICMP_ECHO ||
1497 icmph->type == ICMP_TIMESTAMP) &&
1498 READ_ONCE(net->ipv4.sysctl_icmp_echo_ignore_broadcasts)) {
1499 reason = SKB_DROP_REASON_INVALID_PROTO;
1500 goto error;
1501 }
1502 if (icmph->type != ICMP_ECHO &&
1503 icmph->type != ICMP_TIMESTAMP &&
1504 icmph->type != ICMP_ADDRESS &&
1505 icmph->type != ICMP_ADDRESSREPLY) {
1506 reason = SKB_DROP_REASON_INVALID_PROTO;
1507 goto error;
1508 }
1509 }
1510
1511 if (icmph->type == ICMP_EXT_ECHOREPLY ||
1512 icmph->type == ICMP_ECHOREPLY) {
1513 reason = ping_rcv(skb);
1514 return reason ? NET_RX_DROP : NET_RX_SUCCESS;
1515 }
1516
1517 /*
1518 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1519 *
1520 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1521 * discarded.
1522 */
1523 if (icmph->type > NR_ICMP_TYPES) {
1524 reason = SKB_DROP_REASON_UNHANDLED_PROTO;
1525 goto error;
1526 }
1527
1528 reason = icmp_pointers[icmph->type].handler(skb);
1529 reason_check:
1530 if (!reason) {
1531 consume_skb(skb);
1532 return NET_RX_SUCCESS;
1533 }
1534
1535 drop:
1536 kfree_skb_reason(skb, reason);
1537 return NET_RX_DROP;
1538 csum_error:
1539 reason = SKB_DROP_REASON_ICMP_CSUM;
1540 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1541 error:
1542 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1543 goto drop;
1544 }
1545
ip_icmp_error_rfc4884_validate(const struct sk_buff * skb,int off)1546 static bool ip_icmp_error_rfc4884_validate(const struct sk_buff *skb, int off)
1547 {
1548 struct icmp_extobj_hdr *objh, _objh;
1549 struct icmp_ext_hdr *exth, _exth;
1550 u16 olen;
1551
1552 exth = skb_header_pointer(skb, off, sizeof(_exth), &_exth);
1553 if (!exth)
1554 return false;
1555 if (exth->version != 2)
1556 return true;
1557
1558 if (exth->checksum &&
1559 csum_fold(skb_checksum(skb, off, skb->len - off, 0)))
1560 return false;
1561
1562 off += sizeof(_exth);
1563 while (off < skb->len) {
1564 objh = skb_header_pointer(skb, off, sizeof(_objh), &_objh);
1565 if (!objh)
1566 return false;
1567
1568 olen = ntohs(objh->length);
1569 if (olen < sizeof(_objh))
1570 return false;
1571
1572 off += olen;
1573 if (off > skb->len)
1574 return false;
1575 }
1576
1577 return true;
1578 }
1579
ip_icmp_error_rfc4884(const struct sk_buff * skb,struct sock_ee_data_rfc4884 * out,int thlen,int off)1580 void ip_icmp_error_rfc4884(const struct sk_buff *skb,
1581 struct sock_ee_data_rfc4884 *out,
1582 int thlen, int off)
1583 {
1584 int hlen;
1585
1586 /* original datagram headers: end of icmph to payload (skb->data) */
1587 hlen = -skb_transport_offset(skb) - thlen;
1588
1589 /* per rfc 4884: minimal datagram length of 128 bytes */
1590 if (off < 128 || off < hlen)
1591 return;
1592
1593 /* kernel has stripped headers: return payload offset in bytes */
1594 off -= hlen;
1595 if (off + sizeof(struct icmp_ext_hdr) > skb->len)
1596 return;
1597
1598 out->len = off;
1599
1600 if (!ip_icmp_error_rfc4884_validate(skb, off))
1601 out->flags |= SO_EE_RFC4884_FLAG_INVALID;
1602 }
1603
icmp_err(struct sk_buff * skb,u32 info)1604 int icmp_err(struct sk_buff *skb, u32 info)
1605 {
1606 struct iphdr *iph = (struct iphdr *)skb->data;
1607 int offset = iph->ihl<<2;
1608 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1609 struct net *net = dev_net_rcu(skb->dev);
1610 int type = icmp_hdr(skb)->type;
1611 int code = icmp_hdr(skb)->code;
1612
1613 /*
1614 * Use ping_err to handle all icmp errors except those
1615 * triggered by ICMP_ECHOREPLY which sent from kernel.
1616 */
1617 if (icmph->type != ICMP_ECHOREPLY) {
1618 ping_err(skb, offset, info);
1619 return 0;
1620 }
1621
1622 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1623 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1624 else if (type == ICMP_REDIRECT)
1625 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1626
1627 return 0;
1628 }
1629
1630 /*
1631 * This table is the definition of how we handle ICMP.
1632 */
1633 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1634 [ICMP_ECHOREPLY] = {
1635 .handler = ping_rcv,
1636 },
1637 [1] = {
1638 .handler = icmp_discard,
1639 .error = 1,
1640 },
1641 [2] = {
1642 .handler = icmp_discard,
1643 .error = 1,
1644 },
1645 [ICMP_DEST_UNREACH] = {
1646 .handler = icmp_unreach,
1647 .error = 1,
1648 },
1649 [ICMP_SOURCE_QUENCH] = {
1650 .handler = icmp_unreach,
1651 .error = 1,
1652 },
1653 [ICMP_REDIRECT] = {
1654 .handler = icmp_redirect,
1655 .error = 1,
1656 },
1657 [6] = {
1658 .handler = icmp_discard,
1659 .error = 1,
1660 },
1661 [7] = {
1662 .handler = icmp_discard,
1663 .error = 1,
1664 },
1665 [ICMP_ECHO] = {
1666 .handler = icmp_echo,
1667 },
1668 [9] = {
1669 .handler = icmp_discard,
1670 .error = 1,
1671 },
1672 [10] = {
1673 .handler = icmp_discard,
1674 .error = 1,
1675 },
1676 [ICMP_TIME_EXCEEDED] = {
1677 .handler = icmp_unreach,
1678 .error = 1,
1679 },
1680 [ICMP_PARAMETERPROB] = {
1681 .handler = icmp_unreach,
1682 .error = 1,
1683 },
1684 [ICMP_TIMESTAMP] = {
1685 .handler = icmp_timestamp,
1686 },
1687 [ICMP_TIMESTAMPREPLY] = {
1688 .handler = icmp_discard,
1689 },
1690 [ICMP_INFO_REQUEST] = {
1691 .handler = icmp_discard,
1692 },
1693 [ICMP_INFO_REPLY] = {
1694 .handler = icmp_discard,
1695 },
1696 [ICMP_ADDRESS] = {
1697 .handler = icmp_discard,
1698 },
1699 [ICMP_ADDRESSREPLY] = {
1700 .handler = icmp_discard,
1701 },
1702 };
1703
icmp_sk_init(struct net * net)1704 static int __net_init icmp_sk_init(struct net *net)
1705 {
1706 /* Control parameters for ECHO replies. */
1707 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1708 net->ipv4.sysctl_icmp_echo_enable_probe = 0;
1709 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1710
1711 /* Control parameter - ignore bogus broadcast responses? */
1712 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1713
1714 /*
1715 * Configurable global rate limit.
1716 *
1717 * ratelimit defines tokens/packet consumed for dst->rate_token
1718 * bucket ratemask defines which icmp types are ratelimited by
1719 * setting it's bit position.
1720 *
1721 * default:
1722 * dest unreachable (3), source quench (4),
1723 * time exceeded (11), parameter problem (12)
1724 */
1725
1726 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1727 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1728 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1729 net->ipv4.sysctl_icmp_errors_extension_mask = 0;
1730 net->ipv4.sysctl_icmp_msgs_per_sec = 10000;
1731 net->ipv4.sysctl_icmp_msgs_burst = 10000;
1732
1733 return 0;
1734 }
1735
1736 static struct pernet_operations __net_initdata icmp_sk_ops = {
1737 .init = icmp_sk_init,
1738 };
1739
icmp_init(void)1740 int __init icmp_init(void)
1741 {
1742 int err, i;
1743
1744 for_each_possible_cpu(i) {
1745 struct sock *sk;
1746
1747 err = inet_ctl_sock_create(&sk, PF_INET,
1748 SOCK_RAW, IPPROTO_ICMP, &init_net);
1749 if (err < 0)
1750 return err;
1751
1752 per_cpu(ipv4_icmp_sk, i) = sk;
1753
1754 /* Enough space for 2 64K ICMP packets, including
1755 * sk_buff/skb_shared_info struct overhead.
1756 */
1757 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1758
1759 /*
1760 * Speedup sock_wfree()
1761 */
1762 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1763 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1764 }
1765 return register_pernet_subsys(&icmp_sk_ops);
1766 }
1767