1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NET3: Implementation of the ICMP protocol layer.
4 *
5 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
6 *
7 * Some of the function names and the icmp unreach table for this
8 * module were derived from [icmp.c 1.0.11 06/02/93] by
9 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
10 * Other than that this module is a complete rewrite.
11 *
12 * Fixes:
13 * Clemens Fruhwirth : introduce global icmp rate limiting
14 * with icmp type masking ability instead
15 * of broken per type icmp timeouts.
16 * Mike Shaver : RFC1122 checks.
17 * Alan Cox : Multicast ping reply as self.
18 * Alan Cox : Fix atomicity lockup in ip_build_xmit
19 * call.
20 * Alan Cox : Added 216,128 byte paths to the MTU
21 * code.
22 * Martin Mares : RFC1812 checks.
23 * Martin Mares : Can be configured to follow redirects
24 * if acting as a router _without_ a
25 * routing protocol (RFC 1812).
26 * Martin Mares : Echo requests may be configured to
27 * be ignored (RFC 1812).
28 * Martin Mares : Limitation of ICMP error message
29 * transmit rate (RFC 1812).
30 * Martin Mares : TOS and Precedence set correctly
31 * (RFC 1812).
32 * Martin Mares : Now copying as much data from the
33 * original packet as we can without
34 * exceeding 576 bytes (RFC 1812).
35 * Willy Konynenberg : Transparent proxying support.
36 * Keith Owens : RFC1191 correction for 4.2BSD based
37 * path MTU bug.
38 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
39 * valid (RFC 1812).
40 * Andi Kleen : Check all packet lengths properly
41 * and moved all kfree_skb() up to
42 * icmp_rcv.
43 * Andi Kleen : Move the rate limit bookkeeping
44 * into the dest entry and use a token
45 * bucket filter (thanks to ANK). Make
46 * the rates sysctl configurable.
47 * Yu Tianli : Fixed two ugly bugs in icmp_send
48 * - IP option length was accounted wrongly
49 * - ICMP header length was not accounted
50 * at all.
51 * Tristan Greaves : Added sysctl option to ignore bogus
52 * broadcast responses from broken routers.
53 *
54 * To Fix:
55 *
56 * - Should use skb_pull() instead of all the manual checking.
57 * This would also greatly simply some upper layer error handlers. --AK
58 */
59
60 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
61
62 #include <linux/module.h>
63 #include <linux/types.h>
64 #include <linux/jiffies.h>
65 #include <linux/kernel.h>
66 #include <linux/fcntl.h>
67 #include <linux/socket.h>
68 #include <linux/in.h>
69 #include <linux/inet.h>
70 #include <linux/inetdevice.h>
71 #include <linux/netdevice.h>
72 #include <linux/string.h>
73 #include <linux/netfilter_ipv4.h>
74 #include <linux/slab.h>
75 #include <net/snmp.h>
76 #include <net/ip.h>
77 #include <net/route.h>
78 #include <net/protocol.h>
79 #include <net/icmp.h>
80 #include <net/tcp.h>
81 #include <net/udp.h>
82 #include <net/raw.h>
83 #include <net/ping.h>
84 #include <linux/skbuff.h>
85 #include <net/sock.h>
86 #include <linux/errno.h>
87 #include <linux/timer.h>
88 #include <linux/init.h>
89 #include <linux/uaccess.h>
90 #include <net/checksum.h>
91 #include <net/xfrm.h>
92 #include <net/inet_common.h>
93 #include <net/ip_fib.h>
94 #include <net/l3mdev.h>
95
96 /*
97 * Build xmit assembly blocks
98 */
99
100 struct icmp_bxm {
101 struct sk_buff *skb;
102 int offset;
103 int data_len;
104
105 struct {
106 struct icmphdr icmph;
107 __be32 times[3];
108 } data;
109 int head_len;
110 struct ip_options_data replyopts;
111 };
112
113 /* An array of errno for error messages from dest unreach. */
114 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
115
116 const struct icmp_err icmp_err_convert[] = {
117 {
118 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
119 .fatal = 0,
120 },
121 {
122 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
123 .fatal = 0,
124 },
125 {
126 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
127 .fatal = 1,
128 },
129 {
130 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
131 .fatal = 1,
132 },
133 {
134 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
135 .fatal = 0,
136 },
137 {
138 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
139 .fatal = 0,
140 },
141 {
142 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
143 .fatal = 1,
144 },
145 {
146 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
147 .fatal = 1,
148 },
149 {
150 .errno = ENONET, /* ICMP_HOST_ISOLATED */
151 .fatal = 1,
152 },
153 {
154 .errno = ENETUNREACH, /* ICMP_NET_ANO */
155 .fatal = 1,
156 },
157 {
158 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
163 .fatal = 0,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
167 .fatal = 0,
168 },
169 {
170 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
171 .fatal = 1,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
175 .fatal = 1,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
179 .fatal = 1,
180 },
181 };
182 EXPORT_SYMBOL(icmp_err_convert);
183
184 /*
185 * ICMP control array. This specifies what to do with each ICMP.
186 */
187
188 struct icmp_control {
189 bool (*handler)(struct sk_buff *skb);
190 short error; /* This ICMP is classed as an error message */
191 };
192
193 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
194
195 /*
196 * The ICMP socket(s). This is the most convenient way to flow control
197 * our ICMP output as well as maintain a clean interface throughout
198 * all layers. All Socketless IP sends will soon be gone.
199 *
200 * On SMP we have one ICMP socket per-cpu.
201 */
icmp_sk(struct net * net)202 static struct sock *icmp_sk(struct net *net)
203 {
204 return this_cpu_read(*net->ipv4.icmp_sk);
205 }
206
207 /* Called with BH disabled */
icmp_xmit_lock(struct net * net)208 static inline struct sock *icmp_xmit_lock(struct net *net)
209 {
210 struct sock *sk;
211
212 sk = icmp_sk(net);
213
214 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
215 /* This can happen if the output path signals a
216 * dst_link_failure() for an outgoing ICMP packet.
217 */
218 return NULL;
219 }
220 return sk;
221 }
222
icmp_xmit_unlock(struct sock * sk)223 static inline void icmp_xmit_unlock(struct sock *sk)
224 {
225 spin_unlock(&sk->sk_lock.slock);
226 }
227
228 int sysctl_icmp_msgs_per_sec __read_mostly = 1000;
229 int sysctl_icmp_msgs_burst __read_mostly = 50;
230
231 static struct {
232 spinlock_t lock;
233 u32 credit;
234 u32 stamp;
235 } icmp_global = {
236 .lock = __SPIN_LOCK_UNLOCKED(icmp_global.lock),
237 };
238
239 /**
240 * icmp_global_allow - Are we allowed to send one more ICMP message ?
241 *
242 * Uses a token bucket to limit our ICMP messages to ~sysctl_icmp_msgs_per_sec.
243 * Returns false if we reached the limit and can not send another packet.
244 * Note: called with BH disabled
245 */
icmp_global_allow(void)246 bool icmp_global_allow(void)
247 {
248 u32 credit, delta, incr = 0, now = (u32)jiffies;
249 bool rc = false;
250
251 /* Check if token bucket is empty and cannot be refilled
252 * without taking the spinlock. The READ_ONCE() are paired
253 * with the following WRITE_ONCE() in this same function.
254 */
255 if (!READ_ONCE(icmp_global.credit)) {
256 delta = min_t(u32, now - READ_ONCE(icmp_global.stamp), HZ);
257 if (delta < HZ / 50)
258 return false;
259 }
260
261 spin_lock(&icmp_global.lock);
262 delta = min_t(u32, now - icmp_global.stamp, HZ);
263 if (delta >= HZ / 50) {
264 incr = READ_ONCE(sysctl_icmp_msgs_per_sec) * delta / HZ;
265 if (incr)
266 WRITE_ONCE(icmp_global.stamp, now);
267 }
268 credit = min_t(u32, icmp_global.credit + incr,
269 READ_ONCE(sysctl_icmp_msgs_burst));
270 if (credit) {
271 /* We want to use a credit of one in average, but need to randomize
272 * it for security reasons.
273 */
274 credit = max_t(int, credit - prandom_u32_max(3), 0);
275 rc = true;
276 }
277 WRITE_ONCE(icmp_global.credit, credit);
278 spin_unlock(&icmp_global.lock);
279 return rc;
280 }
281 EXPORT_SYMBOL(icmp_global_allow);
282
icmpv4_mask_allow(struct net * net,int type,int code)283 static bool icmpv4_mask_allow(struct net *net, int type, int code)
284 {
285 if (type > NR_ICMP_TYPES)
286 return true;
287
288 /* Don't limit PMTU discovery. */
289 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
290 return true;
291
292 /* Limit if icmp type is enabled in ratemask. */
293 if (!((1 << type) & READ_ONCE(net->ipv4.sysctl_icmp_ratemask)))
294 return true;
295
296 return false;
297 }
298
icmpv4_global_allow(struct net * net,int type,int code)299 static bool icmpv4_global_allow(struct net *net, int type, int code)
300 {
301 if (icmpv4_mask_allow(net, type, code))
302 return true;
303
304 if (icmp_global_allow())
305 return true;
306
307 return false;
308 }
309
310 /*
311 * Send an ICMP frame.
312 */
313
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code)314 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
315 struct flowi4 *fl4, int type, int code)
316 {
317 struct dst_entry *dst = &rt->dst;
318 struct inet_peer *peer;
319 bool rc = true;
320 int vif;
321
322 if (icmpv4_mask_allow(net, type, code))
323 goto out;
324
325 /* No rate limit on loopback */
326 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
327 goto out;
328
329 vif = l3mdev_master_ifindex(dst->dev);
330 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
331 rc = inet_peer_xrlim_allow(peer,
332 READ_ONCE(net->ipv4.sysctl_icmp_ratelimit));
333 if (peer)
334 inet_putpeer(peer);
335 out:
336 return rc;
337 }
338
339 /*
340 * Maintain the counters used in the SNMP statistics for outgoing ICMP
341 */
icmp_out_count(struct net * net,unsigned char type)342 void icmp_out_count(struct net *net, unsigned char type)
343 {
344 ICMPMSGOUT_INC_STATS(net, type);
345 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
346 }
347
348 /*
349 * Checksum each fragment, and on the first include the headers and final
350 * checksum.
351 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)352 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
353 struct sk_buff *skb)
354 {
355 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
356 __wsum csum;
357
358 csum = skb_copy_and_csum_bits(icmp_param->skb,
359 icmp_param->offset + offset,
360 to, len, 0);
361
362 skb->csum = csum_block_add(skb->csum, csum, odd);
363 if (icmp_pointers[icmp_param->data.icmph.type].error)
364 nf_ct_attach(skb, icmp_param->skb);
365 return 0;
366 }
367
icmp_push_reply(struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)368 static void icmp_push_reply(struct icmp_bxm *icmp_param,
369 struct flowi4 *fl4,
370 struct ipcm_cookie *ipc, struct rtable **rt)
371 {
372 struct sock *sk;
373 struct sk_buff *skb;
374
375 sk = icmp_sk(dev_net((*rt)->dst.dev));
376 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
377 icmp_param->data_len+icmp_param->head_len,
378 icmp_param->head_len,
379 ipc, rt, MSG_DONTWAIT) < 0) {
380 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
381 ip_flush_pending_frames(sk);
382 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
383 struct icmphdr *icmph = icmp_hdr(skb);
384 __wsum csum = 0;
385 struct sk_buff *skb1;
386
387 skb_queue_walk(&sk->sk_write_queue, skb1) {
388 csum = csum_add(csum, skb1->csum);
389 }
390 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
391 (char *)icmph,
392 icmp_param->head_len, csum);
393 icmph->checksum = csum_fold(csum);
394 skb->ip_summed = CHECKSUM_NONE;
395 ip_push_pending_frames(sk, fl4);
396 }
397 }
398
399 /*
400 * Driving logic for building and sending ICMP messages.
401 */
402
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)403 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
404 {
405 struct ipcm_cookie ipc;
406 struct rtable *rt = skb_rtable(skb);
407 struct net *net = dev_net(rt->dst.dev);
408 struct flowi4 fl4;
409 struct sock *sk;
410 struct inet_sock *inet;
411 __be32 daddr, saddr;
412 u32 mark = IP4_REPLY_MARK(net, skb->mark);
413 int type = icmp_param->data.icmph.type;
414 int code = icmp_param->data.icmph.code;
415
416 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
417 return;
418
419 /* Needed by both icmp_global_allow and icmp_xmit_lock */
420 local_bh_disable();
421
422 /* global icmp_msgs_per_sec */
423 if (!icmpv4_global_allow(net, type, code))
424 goto out_bh_enable;
425
426 sk = icmp_xmit_lock(net);
427 if (!sk)
428 goto out_bh_enable;
429 inet = inet_sk(sk);
430
431 icmp_param->data.icmph.checksum = 0;
432
433 ipcm_init(&ipc);
434 inet->tos = ip_hdr(skb)->tos;
435 ipc.sockc.mark = mark;
436 daddr = ipc.addr = ip_hdr(skb)->saddr;
437 saddr = fib_compute_spec_dst(skb);
438
439 if (icmp_param->replyopts.opt.opt.optlen) {
440 ipc.opt = &icmp_param->replyopts.opt;
441 if (ipc.opt->opt.srr)
442 daddr = icmp_param->replyopts.opt.opt.faddr;
443 }
444 memset(&fl4, 0, sizeof(fl4));
445 fl4.daddr = daddr;
446 fl4.saddr = saddr;
447 fl4.flowi4_mark = mark;
448 fl4.flowi4_uid = sock_net_uid(net, NULL);
449 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
450 fl4.flowi4_proto = IPPROTO_ICMP;
451 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
452 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
453 rt = ip_route_output_key(net, &fl4);
454 if (IS_ERR(rt))
455 goto out_unlock;
456 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
457 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
458 ip_rt_put(rt);
459 out_unlock:
460 icmp_xmit_unlock(sk);
461 out_bh_enable:
462 local_bh_enable();
463 }
464
465 /*
466 * The device used for looking up which routing table to use for sending an ICMP
467 * error is preferably the source whenever it is set, which should ensure the
468 * icmp error can be sent to the source host, else lookup using the routing
469 * table of the destination device, else use the main routing table (index 0).
470 */
icmp_get_route_lookup_dev(struct sk_buff * skb)471 static struct net_device *icmp_get_route_lookup_dev(struct sk_buff *skb)
472 {
473 struct net_device *route_lookup_dev = NULL;
474
475 if (skb->dev)
476 route_lookup_dev = skb->dev;
477 else if (skb_dst(skb))
478 route_lookup_dev = skb_dst(skb)->dev;
479 return route_lookup_dev;
480 }
481
icmp_route_lookup(struct net * net,struct flowi4 * fl4,struct sk_buff * skb_in,const struct iphdr * iph,__be32 saddr,u8 tos,u32 mark,int type,int code,struct icmp_bxm * param)482 static struct rtable *icmp_route_lookup(struct net *net,
483 struct flowi4 *fl4,
484 struct sk_buff *skb_in,
485 const struct iphdr *iph,
486 __be32 saddr, u8 tos, u32 mark,
487 int type, int code,
488 struct icmp_bxm *param)
489 {
490 struct net_device *route_lookup_dev;
491 struct rtable *rt, *rt2;
492 struct flowi4 fl4_dec;
493 int err;
494
495 memset(fl4, 0, sizeof(*fl4));
496 fl4->daddr = (param->replyopts.opt.opt.srr ?
497 param->replyopts.opt.opt.faddr : iph->saddr);
498 fl4->saddr = saddr;
499 fl4->flowi4_mark = mark;
500 fl4->flowi4_uid = sock_net_uid(net, NULL);
501 fl4->flowi4_tos = RT_TOS(tos);
502 fl4->flowi4_proto = IPPROTO_ICMP;
503 fl4->fl4_icmp_type = type;
504 fl4->fl4_icmp_code = code;
505 route_lookup_dev = icmp_get_route_lookup_dev(skb_in);
506 fl4->flowi4_oif = l3mdev_master_ifindex(route_lookup_dev);
507
508 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
509 rt = ip_route_output_key_hash(net, fl4, skb_in);
510 if (IS_ERR(rt))
511 return rt;
512
513 /* No need to clone since we're just using its address. */
514 rt2 = rt;
515
516 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
517 flowi4_to_flowi(fl4), NULL, 0);
518 if (!IS_ERR(rt)) {
519 if (rt != rt2)
520 return rt;
521 } else if (PTR_ERR(rt) == -EPERM) {
522 rt = NULL;
523 } else
524 return rt;
525
526 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
527 if (err)
528 goto relookup_failed;
529
530 if (inet_addr_type_dev_table(net, route_lookup_dev,
531 fl4_dec.saddr) == RTN_LOCAL) {
532 rt2 = __ip_route_output_key(net, &fl4_dec);
533 if (IS_ERR(rt2))
534 err = PTR_ERR(rt2);
535 } else {
536 struct flowi4 fl4_2 = {};
537 unsigned long orefdst;
538
539 fl4_2.daddr = fl4_dec.saddr;
540 rt2 = ip_route_output_key(net, &fl4_2);
541 if (IS_ERR(rt2)) {
542 err = PTR_ERR(rt2);
543 goto relookup_failed;
544 }
545 /* Ugh! */
546 orefdst = skb_in->_skb_refdst; /* save old refdst */
547 skb_dst_set(skb_in, NULL);
548 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
549 RT_TOS(tos), rt2->dst.dev);
550
551 dst_release(&rt2->dst);
552 rt2 = skb_rtable(skb_in);
553 skb_in->_skb_refdst = orefdst; /* restore old refdst */
554 }
555
556 if (err)
557 goto relookup_failed;
558
559 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
560 flowi4_to_flowi(&fl4_dec), NULL,
561 XFRM_LOOKUP_ICMP);
562 if (!IS_ERR(rt2)) {
563 dst_release(&rt->dst);
564 memcpy(fl4, &fl4_dec, sizeof(*fl4));
565 rt = rt2;
566 } else if (PTR_ERR(rt2) == -EPERM) {
567 if (rt)
568 dst_release(&rt->dst);
569 return rt2;
570 } else {
571 err = PTR_ERR(rt2);
572 goto relookup_failed;
573 }
574 return rt;
575
576 relookup_failed:
577 if (rt)
578 return rt;
579 return ERR_PTR(err);
580 }
581
582 /*
583 * Send an ICMP message in response to a situation
584 *
585 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
586 * MAY send more (we do).
587 * MUST NOT change this header information.
588 * MUST NOT reply to a multicast/broadcast IP address.
589 * MUST NOT reply to a multicast/broadcast MAC address.
590 * MUST reply to only the first fragment.
591 */
592
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct ip_options * opt)593 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
594 const struct ip_options *opt)
595 {
596 struct iphdr *iph;
597 int room;
598 struct icmp_bxm icmp_param;
599 struct rtable *rt = skb_rtable(skb_in);
600 struct ipcm_cookie ipc;
601 struct flowi4 fl4;
602 __be32 saddr;
603 u8 tos;
604 u32 mark;
605 struct net *net;
606 struct sock *sk;
607
608 if (!rt)
609 goto out;
610
611 if (rt->dst.dev)
612 net = dev_net(rt->dst.dev);
613 else if (skb_in->dev)
614 net = dev_net(skb_in->dev);
615 else
616 goto out;
617
618 /*
619 * Find the original header. It is expected to be valid, of course.
620 * Check this, icmp_send is called from the most obscure devices
621 * sometimes.
622 */
623 iph = ip_hdr(skb_in);
624
625 if ((u8 *)iph < skb_in->head ||
626 (skb_network_header(skb_in) + sizeof(*iph)) >
627 skb_tail_pointer(skb_in))
628 goto out;
629
630 /*
631 * No replies to physical multicast/broadcast
632 */
633 if (skb_in->pkt_type != PACKET_HOST)
634 goto out;
635
636 /*
637 * Now check at the protocol level
638 */
639 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
640 goto out;
641
642 /*
643 * Only reply to fragment 0. We byte re-order the constant
644 * mask for efficiency.
645 */
646 if (iph->frag_off & htons(IP_OFFSET))
647 goto out;
648
649 /*
650 * If we send an ICMP error to an ICMP error a mess would result..
651 */
652 if (icmp_pointers[type].error) {
653 /*
654 * We are an error, check if we are replying to an
655 * ICMP error
656 */
657 if (iph->protocol == IPPROTO_ICMP) {
658 u8 _inner_type, *itp;
659
660 itp = skb_header_pointer(skb_in,
661 skb_network_header(skb_in) +
662 (iph->ihl << 2) +
663 offsetof(struct icmphdr,
664 type) -
665 skb_in->data,
666 sizeof(_inner_type),
667 &_inner_type);
668 if (!itp)
669 goto out;
670
671 /*
672 * Assume any unknown ICMP type is an error. This
673 * isn't specified by the RFC, but think about it..
674 */
675 if (*itp > NR_ICMP_TYPES ||
676 icmp_pointers[*itp].error)
677 goto out;
678 }
679 }
680
681 /* Needed by both icmp_global_allow and icmp_xmit_lock */
682 local_bh_disable();
683
684 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
685 * incoming dev is loopback. If outgoing dev change to not be
686 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
687 */
688 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
689 !icmpv4_global_allow(net, type, code))
690 goto out_bh_enable;
691
692 sk = icmp_xmit_lock(net);
693 if (!sk)
694 goto out_bh_enable;
695
696 /*
697 * Construct source address and options.
698 */
699
700 saddr = iph->daddr;
701 if (!(rt->rt_flags & RTCF_LOCAL)) {
702 struct net_device *dev = NULL;
703
704 rcu_read_lock();
705 if (rt_is_input_route(rt) &&
706 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
707 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
708
709 if (dev)
710 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
711 else
712 saddr = 0;
713 rcu_read_unlock();
714 }
715
716 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
717 IPTOS_PREC_INTERNETCONTROL) :
718 iph->tos;
719 mark = IP4_REPLY_MARK(net, skb_in->mark);
720
721 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
722 goto out_unlock;
723
724
725 /*
726 * Prepare data for ICMP header.
727 */
728
729 icmp_param.data.icmph.type = type;
730 icmp_param.data.icmph.code = code;
731 icmp_param.data.icmph.un.gateway = info;
732 icmp_param.data.icmph.checksum = 0;
733 icmp_param.skb = skb_in;
734 icmp_param.offset = skb_network_offset(skb_in);
735 inet_sk(sk)->tos = tos;
736 ipcm_init(&ipc);
737 ipc.addr = iph->saddr;
738 ipc.opt = &icmp_param.replyopts.opt;
739 ipc.sockc.mark = mark;
740
741 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
742 type, code, &icmp_param);
743 if (IS_ERR(rt))
744 goto out_unlock;
745
746 /* peer icmp_ratelimit */
747 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
748 goto ende;
749
750 /* RFC says return as much as we can without exceeding 576 bytes. */
751
752 room = dst_mtu(&rt->dst);
753 if (room > 576)
754 room = 576;
755 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
756 room -= sizeof(struct icmphdr);
757 /* Guard against tiny mtu. We need to include at least one
758 * IP network header for this message to make any sense.
759 */
760 if (room <= (int)sizeof(struct iphdr))
761 goto ende;
762
763 icmp_param.data_len = skb_in->len - icmp_param.offset;
764 if (icmp_param.data_len > room)
765 icmp_param.data_len = room;
766 icmp_param.head_len = sizeof(struct icmphdr);
767
768 /* if we don't have a source address at this point, fall back to the
769 * dummy address instead of sending out a packet with a source address
770 * of 0.0.0.0
771 */
772 if (!fl4.saddr)
773 fl4.saddr = htonl(INADDR_DUMMY);
774
775 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
776 ende:
777 ip_rt_put(rt);
778 out_unlock:
779 icmp_xmit_unlock(sk);
780 out_bh_enable:
781 local_bh_enable();
782 out:;
783 }
784 EXPORT_SYMBOL(__icmp_send);
785
786 #if IS_ENABLED(CONFIG_NF_NAT)
787 #include <net/netfilter/nf_conntrack.h>
icmp_ndo_send(struct sk_buff * skb_in,int type,int code,__be32 info)788 void icmp_ndo_send(struct sk_buff *skb_in, int type, int code, __be32 info)
789 {
790 struct sk_buff *cloned_skb = NULL;
791 struct ip_options opts = { 0 };
792 enum ip_conntrack_info ctinfo;
793 struct nf_conn *ct;
794 __be32 orig_ip;
795
796 ct = nf_ct_get(skb_in, &ctinfo);
797 if (!ct || !(ct->status & IPS_SRC_NAT)) {
798 __icmp_send(skb_in, type, code, info, &opts);
799 return;
800 }
801
802 if (skb_shared(skb_in))
803 skb_in = cloned_skb = skb_clone(skb_in, GFP_ATOMIC);
804
805 if (unlikely(!skb_in || skb_network_header(skb_in) < skb_in->head ||
806 (skb_network_header(skb_in) + sizeof(struct iphdr)) >
807 skb_tail_pointer(skb_in) || skb_ensure_writable(skb_in,
808 skb_network_offset(skb_in) + sizeof(struct iphdr))))
809 goto out;
810
811 orig_ip = ip_hdr(skb_in)->saddr;
812 ip_hdr(skb_in)->saddr = ct->tuplehash[0].tuple.src.u3.ip;
813 __icmp_send(skb_in, type, code, info, &opts);
814 ip_hdr(skb_in)->saddr = orig_ip;
815 out:
816 consume_skb(cloned_skb);
817 }
818 EXPORT_SYMBOL(icmp_ndo_send);
819 #endif
820
icmp_socket_deliver(struct sk_buff * skb,u32 info)821 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
822 {
823 const struct iphdr *iph = (const struct iphdr *) skb->data;
824 const struct net_protocol *ipprot;
825 int protocol = iph->protocol;
826
827 /* Checkin full IP header plus 8 bytes of protocol to
828 * avoid additional coding at protocol handlers.
829 */
830 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
831 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
832 return;
833 }
834
835 raw_icmp_error(skb, protocol, info);
836
837 ipprot = rcu_dereference(inet_protos[protocol]);
838 if (ipprot && ipprot->err_handler)
839 ipprot->err_handler(skb, info);
840 }
841
icmp_tag_validation(int proto)842 static bool icmp_tag_validation(int proto)
843 {
844 bool ok;
845
846 rcu_read_lock();
847 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
848 rcu_read_unlock();
849 return ok;
850 }
851
852 /*
853 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
854 * ICMP_PARAMETERPROB.
855 */
856
icmp_unreach(struct sk_buff * skb)857 static bool icmp_unreach(struct sk_buff *skb)
858 {
859 const struct iphdr *iph;
860 struct icmphdr *icmph;
861 struct net *net;
862 u32 info = 0;
863
864 net = dev_net(skb_dst(skb)->dev);
865
866 /*
867 * Incomplete header ?
868 * Only checks for the IP header, there should be an
869 * additional check for longer headers in upper levels.
870 */
871
872 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
873 goto out_err;
874
875 icmph = icmp_hdr(skb);
876 iph = (const struct iphdr *)skb->data;
877
878 if (iph->ihl < 5) /* Mangled header, drop. */
879 goto out_err;
880
881 switch (icmph->type) {
882 case ICMP_DEST_UNREACH:
883 switch (icmph->code & 15) {
884 case ICMP_NET_UNREACH:
885 case ICMP_HOST_UNREACH:
886 case ICMP_PROT_UNREACH:
887 case ICMP_PORT_UNREACH:
888 break;
889 case ICMP_FRAG_NEEDED:
890 /* for documentation of the ip_no_pmtu_disc
891 * values please see
892 * Documentation/networking/ip-sysctl.txt
893 */
894 switch (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) {
895 default:
896 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
897 &iph->daddr);
898 break;
899 case 2:
900 goto out;
901 case 3:
902 if (!icmp_tag_validation(iph->protocol))
903 goto out;
904 /* fall through */
905 case 0:
906 info = ntohs(icmph->un.frag.mtu);
907 }
908 break;
909 case ICMP_SR_FAILED:
910 net_dbg_ratelimited("%pI4: Source Route Failed\n",
911 &iph->daddr);
912 break;
913 default:
914 break;
915 }
916 if (icmph->code > NR_ICMP_UNREACH)
917 goto out;
918 break;
919 case ICMP_PARAMETERPROB:
920 info = ntohl(icmph->un.gateway) >> 24;
921 break;
922 case ICMP_TIME_EXCEEDED:
923 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
924 if (icmph->code == ICMP_EXC_FRAGTIME)
925 goto out;
926 break;
927 }
928
929 /*
930 * Throw it at our lower layers
931 *
932 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
933 * header.
934 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
935 * transport layer.
936 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
937 * transport layer.
938 */
939
940 /*
941 * Check the other end isn't violating RFC 1122. Some routers send
942 * bogus responses to broadcast frames. If you see this message
943 * first check your netmask matches at both ends, if it does then
944 * get the other vendor to fix their kit.
945 */
946
947 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
948 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
949 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
950 &ip_hdr(skb)->saddr,
951 icmph->type, icmph->code,
952 &iph->daddr, skb->dev->name);
953 goto out;
954 }
955
956 icmp_socket_deliver(skb, info);
957
958 out:
959 return true;
960 out_err:
961 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
962 return false;
963 }
964
965
966 /*
967 * Handle ICMP_REDIRECT.
968 */
969
icmp_redirect(struct sk_buff * skb)970 static bool icmp_redirect(struct sk_buff *skb)
971 {
972 if (skb->len < sizeof(struct iphdr)) {
973 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
974 return false;
975 }
976
977 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
978 /* there aught to be a stat */
979 return false;
980 }
981
982 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
983 return true;
984 }
985
986 /*
987 * Handle ICMP_ECHO ("ping") requests.
988 *
989 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
990 * requests.
991 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
992 * included in the reply.
993 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
994 * echo requests, MUST have default=NOT.
995 * See also WRT handling of options once they are done and working.
996 */
997
icmp_echo(struct sk_buff * skb)998 static bool icmp_echo(struct sk_buff *skb)
999 {
1000 struct net *net;
1001
1002 net = dev_net(skb_dst(skb)->dev);
1003 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
1004 struct icmp_bxm icmp_param;
1005
1006 icmp_param.data.icmph = *icmp_hdr(skb);
1007 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
1008 icmp_param.skb = skb;
1009 icmp_param.offset = 0;
1010 icmp_param.data_len = skb->len;
1011 icmp_param.head_len = sizeof(struct icmphdr);
1012 icmp_reply(&icmp_param, skb);
1013 }
1014 /* should there be an ICMP stat for ignored echos? */
1015 return true;
1016 }
1017
1018 /*
1019 * Handle ICMP Timestamp requests.
1020 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
1021 * SHOULD be in the kernel for minimum random latency.
1022 * MUST be accurate to a few minutes.
1023 * MUST be updated at least at 15Hz.
1024 */
icmp_timestamp(struct sk_buff * skb)1025 static bool icmp_timestamp(struct sk_buff *skb)
1026 {
1027 struct icmp_bxm icmp_param;
1028 /*
1029 * Too short.
1030 */
1031 if (skb->len < 4)
1032 goto out_err;
1033
1034 /*
1035 * Fill in the current time as ms since midnight UT:
1036 */
1037 icmp_param.data.times[1] = inet_current_timestamp();
1038 icmp_param.data.times[2] = icmp_param.data.times[1];
1039
1040 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
1041
1042 icmp_param.data.icmph = *icmp_hdr(skb);
1043 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
1044 icmp_param.data.icmph.code = 0;
1045 icmp_param.skb = skb;
1046 icmp_param.offset = 0;
1047 icmp_param.data_len = 0;
1048 icmp_param.head_len = sizeof(struct icmphdr) + 12;
1049 icmp_reply(&icmp_param, skb);
1050 return true;
1051
1052 out_err:
1053 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
1054 return false;
1055 }
1056
icmp_discard(struct sk_buff * skb)1057 static bool icmp_discard(struct sk_buff *skb)
1058 {
1059 /* pretend it was a success */
1060 return true;
1061 }
1062
1063 /*
1064 * Deal with incoming ICMP packets.
1065 */
icmp_rcv(struct sk_buff * skb)1066 int icmp_rcv(struct sk_buff *skb)
1067 {
1068 struct icmphdr *icmph;
1069 struct rtable *rt = skb_rtable(skb);
1070 struct net *net = dev_net(rt->dst.dev);
1071 bool success;
1072
1073 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1074 struct sec_path *sp = skb_sec_path(skb);
1075 int nh;
1076
1077 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1078 XFRM_STATE_ICMP))
1079 goto drop;
1080
1081 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1082 goto drop;
1083
1084 nh = skb_network_offset(skb);
1085 skb_set_network_header(skb, sizeof(*icmph));
1086
1087 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1088 goto drop;
1089
1090 skb_set_network_header(skb, nh);
1091 }
1092
1093 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1094
1095 if (skb_checksum_simple_validate(skb))
1096 goto csum_error;
1097
1098 if (!pskb_pull(skb, sizeof(*icmph)))
1099 goto error;
1100
1101 icmph = icmp_hdr(skb);
1102
1103 ICMPMSGIN_INC_STATS(net, icmph->type);
1104 /*
1105 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1106 *
1107 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1108 * discarded.
1109 */
1110 if (icmph->type > NR_ICMP_TYPES)
1111 goto error;
1112
1113
1114 /*
1115 * Parse the ICMP message
1116 */
1117
1118 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1119 /*
1120 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1121 * silently ignored (we let user decide with a sysctl).
1122 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1123 * discarded if to broadcast/multicast.
1124 */
1125 if ((icmph->type == ICMP_ECHO ||
1126 icmph->type == ICMP_TIMESTAMP) &&
1127 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1128 goto error;
1129 }
1130 if (icmph->type != ICMP_ECHO &&
1131 icmph->type != ICMP_TIMESTAMP &&
1132 icmph->type != ICMP_ADDRESS &&
1133 icmph->type != ICMP_ADDRESSREPLY) {
1134 goto error;
1135 }
1136 }
1137
1138 success = icmp_pointers[icmph->type].handler(skb);
1139
1140 if (success) {
1141 consume_skb(skb);
1142 return NET_RX_SUCCESS;
1143 }
1144
1145 drop:
1146 kfree_skb(skb);
1147 return NET_RX_DROP;
1148 csum_error:
1149 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1150 error:
1151 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1152 goto drop;
1153 }
1154
icmp_err(struct sk_buff * skb,u32 info)1155 int icmp_err(struct sk_buff *skb, u32 info)
1156 {
1157 struct iphdr *iph = (struct iphdr *)skb->data;
1158 int offset = iph->ihl<<2;
1159 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1160 int type = icmp_hdr(skb)->type;
1161 int code = icmp_hdr(skb)->code;
1162 struct net *net = dev_net(skb->dev);
1163
1164 /*
1165 * Use ping_err to handle all icmp errors except those
1166 * triggered by ICMP_ECHOREPLY which sent from kernel.
1167 */
1168 if (icmph->type != ICMP_ECHOREPLY) {
1169 ping_err(skb, offset, info);
1170 return 0;
1171 }
1172
1173 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1174 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1175 else if (type == ICMP_REDIRECT)
1176 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1177
1178 return 0;
1179 }
1180
1181 /*
1182 * This table is the definition of how we handle ICMP.
1183 */
1184 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1185 [ICMP_ECHOREPLY] = {
1186 .handler = ping_rcv,
1187 },
1188 [1] = {
1189 .handler = icmp_discard,
1190 .error = 1,
1191 },
1192 [2] = {
1193 .handler = icmp_discard,
1194 .error = 1,
1195 },
1196 [ICMP_DEST_UNREACH] = {
1197 .handler = icmp_unreach,
1198 .error = 1,
1199 },
1200 [ICMP_SOURCE_QUENCH] = {
1201 .handler = icmp_unreach,
1202 .error = 1,
1203 },
1204 [ICMP_REDIRECT] = {
1205 .handler = icmp_redirect,
1206 .error = 1,
1207 },
1208 [6] = {
1209 .handler = icmp_discard,
1210 .error = 1,
1211 },
1212 [7] = {
1213 .handler = icmp_discard,
1214 .error = 1,
1215 },
1216 [ICMP_ECHO] = {
1217 .handler = icmp_echo,
1218 },
1219 [9] = {
1220 .handler = icmp_discard,
1221 .error = 1,
1222 },
1223 [10] = {
1224 .handler = icmp_discard,
1225 .error = 1,
1226 },
1227 [ICMP_TIME_EXCEEDED] = {
1228 .handler = icmp_unreach,
1229 .error = 1,
1230 },
1231 [ICMP_PARAMETERPROB] = {
1232 .handler = icmp_unreach,
1233 .error = 1,
1234 },
1235 [ICMP_TIMESTAMP] = {
1236 .handler = icmp_timestamp,
1237 },
1238 [ICMP_TIMESTAMPREPLY] = {
1239 .handler = icmp_discard,
1240 },
1241 [ICMP_INFO_REQUEST] = {
1242 .handler = icmp_discard,
1243 },
1244 [ICMP_INFO_REPLY] = {
1245 .handler = icmp_discard,
1246 },
1247 [ICMP_ADDRESS] = {
1248 .handler = icmp_discard,
1249 },
1250 [ICMP_ADDRESSREPLY] = {
1251 .handler = icmp_discard,
1252 },
1253 };
1254
icmp_sk_exit(struct net * net)1255 static void __net_exit icmp_sk_exit(struct net *net)
1256 {
1257 int i;
1258
1259 for_each_possible_cpu(i)
1260 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1261 free_percpu(net->ipv4.icmp_sk);
1262 net->ipv4.icmp_sk = NULL;
1263 }
1264
icmp_sk_init(struct net * net)1265 static int __net_init icmp_sk_init(struct net *net)
1266 {
1267 int i, err;
1268
1269 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1270 if (!net->ipv4.icmp_sk)
1271 return -ENOMEM;
1272
1273 for_each_possible_cpu(i) {
1274 struct sock *sk;
1275
1276 err = inet_ctl_sock_create(&sk, PF_INET,
1277 SOCK_RAW, IPPROTO_ICMP, net);
1278 if (err < 0)
1279 goto fail;
1280
1281 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1282
1283 /* Enough space for 2 64K ICMP packets, including
1284 * sk_buff/skb_shared_info struct overhead.
1285 */
1286 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1287
1288 /*
1289 * Speedup sock_wfree()
1290 */
1291 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1292 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1293 }
1294
1295 /* Control parameters for ECHO replies. */
1296 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1297 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1298
1299 /* Control parameter - ignore bogus broadcast responses? */
1300 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1301
1302 /*
1303 * Configurable global rate limit.
1304 *
1305 * ratelimit defines tokens/packet consumed for dst->rate_token
1306 * bucket ratemask defines which icmp types are ratelimited by
1307 * setting it's bit position.
1308 *
1309 * default:
1310 * dest unreachable (3), source quench (4),
1311 * time exceeded (11), parameter problem (12)
1312 */
1313
1314 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1315 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1316 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1317
1318 return 0;
1319
1320 fail:
1321 icmp_sk_exit(net);
1322 return err;
1323 }
1324
1325 static struct pernet_operations __net_initdata icmp_sk_ops = {
1326 .init = icmp_sk_init,
1327 .exit = icmp_sk_exit,
1328 };
1329
icmp_init(void)1330 int __init icmp_init(void)
1331 {
1332 return register_pernet_subsys(&icmp_sk_ops);
1333 }
1334