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 = 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, sysctl_icmp_msgs_burst);
269 if (credit) {
270 credit--;
271 rc = true;
272 }
273 WRITE_ONCE(icmp_global.credit, credit);
274 spin_unlock(&icmp_global.lock);
275 return rc;
276 }
277 EXPORT_SYMBOL(icmp_global_allow);
278
icmpv4_mask_allow(struct net * net,int type,int code)279 static bool icmpv4_mask_allow(struct net *net, int type, int code)
280 {
281 if (type > NR_ICMP_TYPES)
282 return true;
283
284 /* Don't limit PMTU discovery. */
285 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
286 return true;
287
288 /* Limit if icmp type is enabled in ratemask. */
289 if (!((1 << type) & net->ipv4.sysctl_icmp_ratemask))
290 return true;
291
292 return false;
293 }
294
icmpv4_global_allow(struct net * net,int type,int code)295 static bool icmpv4_global_allow(struct net *net, int type, int code)
296 {
297 if (icmpv4_mask_allow(net, type, code))
298 return true;
299
300 if (icmp_global_allow())
301 return true;
302
303 return false;
304 }
305
306 /*
307 * Send an ICMP frame.
308 */
309
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code)310 static bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
311 struct flowi4 *fl4, int type, int code)
312 {
313 struct dst_entry *dst = &rt->dst;
314 struct inet_peer *peer;
315 bool rc = true;
316 int vif;
317
318 if (icmpv4_mask_allow(net, type, code))
319 goto out;
320
321 /* No rate limit on loopback */
322 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
323 goto out;
324
325 vif = l3mdev_master_ifindex(dst->dev);
326 peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, vif, 1);
327 rc = inet_peer_xrlim_allow(peer, net->ipv4.sysctl_icmp_ratelimit);
328 if (peer)
329 inet_putpeer(peer);
330 out:
331 return rc;
332 }
333
334 /*
335 * Maintain the counters used in the SNMP statistics for outgoing ICMP
336 */
icmp_out_count(struct net * net,unsigned char type)337 void icmp_out_count(struct net *net, unsigned char type)
338 {
339 ICMPMSGOUT_INC_STATS(net, type);
340 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
341 }
342
343 /*
344 * Checksum each fragment, and on the first include the headers and final
345 * checksum.
346 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)347 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
348 struct sk_buff *skb)
349 {
350 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
351 __wsum csum;
352
353 csum = skb_copy_and_csum_bits(icmp_param->skb,
354 icmp_param->offset + offset,
355 to, len, 0);
356
357 skb->csum = csum_block_add(skb->csum, csum, odd);
358 if (icmp_pointers[icmp_param->data.icmph.type].error)
359 nf_ct_attach(skb, icmp_param->skb);
360 return 0;
361 }
362
icmp_push_reply(struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)363 static void icmp_push_reply(struct icmp_bxm *icmp_param,
364 struct flowi4 *fl4,
365 struct ipcm_cookie *ipc, struct rtable **rt)
366 {
367 struct sock *sk;
368 struct sk_buff *skb;
369
370 sk = icmp_sk(dev_net((*rt)->dst.dev));
371 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
372 icmp_param->data_len+icmp_param->head_len,
373 icmp_param->head_len,
374 ipc, rt, MSG_DONTWAIT) < 0) {
375 __ICMP_INC_STATS(sock_net(sk), ICMP_MIB_OUTERRORS);
376 ip_flush_pending_frames(sk);
377 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
378 struct icmphdr *icmph = icmp_hdr(skb);
379 __wsum csum = 0;
380 struct sk_buff *skb1;
381
382 skb_queue_walk(&sk->sk_write_queue, skb1) {
383 csum = csum_add(csum, skb1->csum);
384 }
385 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
386 (char *)icmph,
387 icmp_param->head_len, csum);
388 icmph->checksum = csum_fold(csum);
389 skb->ip_summed = CHECKSUM_NONE;
390 ip_push_pending_frames(sk, fl4);
391 }
392 }
393
394 /*
395 * Driving logic for building and sending ICMP messages.
396 */
397
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)398 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
399 {
400 struct ipcm_cookie ipc;
401 struct rtable *rt = skb_rtable(skb);
402 struct net *net = dev_net(rt->dst.dev);
403 struct flowi4 fl4;
404 struct sock *sk;
405 struct inet_sock *inet;
406 __be32 daddr, saddr;
407 u32 mark = IP4_REPLY_MARK(net, skb->mark);
408 int type = icmp_param->data.icmph.type;
409 int code = icmp_param->data.icmph.code;
410
411 if (ip_options_echo(net, &icmp_param->replyopts.opt.opt, skb))
412 return;
413
414 /* Needed by both icmp_global_allow and icmp_xmit_lock */
415 local_bh_disable();
416
417 /* global icmp_msgs_per_sec */
418 if (!icmpv4_global_allow(net, type, code))
419 goto out_bh_enable;
420
421 sk = icmp_xmit_lock(net);
422 if (!sk)
423 goto out_bh_enable;
424 inet = inet_sk(sk);
425
426 icmp_param->data.icmph.checksum = 0;
427
428 ipcm_init(&ipc);
429 inet->tos = ip_hdr(skb)->tos;
430 sk->sk_mark = mark;
431 daddr = ipc.addr = ip_hdr(skb)->saddr;
432 saddr = fib_compute_spec_dst(skb);
433
434 if (icmp_param->replyopts.opt.opt.optlen) {
435 ipc.opt = &icmp_param->replyopts.opt;
436 if (ipc.opt->opt.srr)
437 daddr = icmp_param->replyopts.opt.opt.faddr;
438 }
439 memset(&fl4, 0, sizeof(fl4));
440 fl4.daddr = daddr;
441 fl4.saddr = saddr;
442 fl4.flowi4_mark = mark;
443 fl4.flowi4_uid = sock_net_uid(net, NULL);
444 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
445 fl4.flowi4_proto = IPPROTO_ICMP;
446 fl4.flowi4_oif = l3mdev_master_ifindex(skb->dev);
447 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
448 rt = ip_route_output_key(net, &fl4);
449 if (IS_ERR(rt))
450 goto out_unlock;
451 if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
452 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
453 ip_rt_put(rt);
454 out_unlock:
455 icmp_xmit_unlock(sk);
456 out_bh_enable:
457 local_bh_enable();
458 }
459
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)460 static struct rtable *icmp_route_lookup(struct net *net,
461 struct flowi4 *fl4,
462 struct sk_buff *skb_in,
463 const struct iphdr *iph,
464 __be32 saddr, u8 tos, u32 mark,
465 int type, int code,
466 struct icmp_bxm *param)
467 {
468 struct rtable *rt, *rt2;
469 struct flowi4 fl4_dec;
470 int err;
471
472 memset(fl4, 0, sizeof(*fl4));
473 fl4->daddr = (param->replyopts.opt.opt.srr ?
474 param->replyopts.opt.opt.faddr : iph->saddr);
475 fl4->saddr = saddr;
476 fl4->flowi4_mark = mark;
477 fl4->flowi4_uid = sock_net_uid(net, NULL);
478 fl4->flowi4_tos = RT_TOS(tos);
479 fl4->flowi4_proto = IPPROTO_ICMP;
480 fl4->fl4_icmp_type = type;
481 fl4->fl4_icmp_code = code;
482 fl4->flowi4_oif = l3mdev_master_ifindex(skb_dst(skb_in)->dev);
483
484 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
485 rt = ip_route_output_key_hash(net, fl4, skb_in);
486 if (IS_ERR(rt))
487 return rt;
488
489 /* No need to clone since we're just using its address. */
490 rt2 = rt;
491
492 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
493 flowi4_to_flowi(fl4), NULL, 0);
494 if (!IS_ERR(rt)) {
495 if (rt != rt2)
496 return rt;
497 } else if (PTR_ERR(rt) == -EPERM) {
498 rt = NULL;
499 } else
500 return rt;
501
502 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
503 if (err)
504 goto relookup_failed;
505
506 if (inet_addr_type_dev_table(net, skb_dst(skb_in)->dev,
507 fl4_dec.saddr) == RTN_LOCAL) {
508 rt2 = __ip_route_output_key(net, &fl4_dec);
509 if (IS_ERR(rt2))
510 err = PTR_ERR(rt2);
511 } else {
512 struct flowi4 fl4_2 = {};
513 unsigned long orefdst;
514
515 fl4_2.daddr = fl4_dec.saddr;
516 rt2 = ip_route_output_key(net, &fl4_2);
517 if (IS_ERR(rt2)) {
518 err = PTR_ERR(rt2);
519 goto relookup_failed;
520 }
521 /* Ugh! */
522 orefdst = skb_in->_skb_refdst; /* save old refdst */
523 skb_dst_set(skb_in, NULL);
524 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
525 RT_TOS(tos), rt2->dst.dev);
526
527 dst_release(&rt2->dst);
528 rt2 = skb_rtable(skb_in);
529 skb_in->_skb_refdst = orefdst; /* restore old refdst */
530 }
531
532 if (err)
533 goto relookup_failed;
534
535 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
536 flowi4_to_flowi(&fl4_dec), NULL,
537 XFRM_LOOKUP_ICMP);
538 if (!IS_ERR(rt2)) {
539 dst_release(&rt->dst);
540 memcpy(fl4, &fl4_dec, sizeof(*fl4));
541 rt = rt2;
542 } else if (PTR_ERR(rt2) == -EPERM) {
543 if (rt)
544 dst_release(&rt->dst);
545 return rt2;
546 } else {
547 err = PTR_ERR(rt2);
548 goto relookup_failed;
549 }
550 return rt;
551
552 relookup_failed:
553 if (rt)
554 return rt;
555 return ERR_PTR(err);
556 }
557
558 /*
559 * Send an ICMP message in response to a situation
560 *
561 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
562 * MAY send more (we do).
563 * MUST NOT change this header information.
564 * MUST NOT reply to a multicast/broadcast IP address.
565 * MUST NOT reply to a multicast/broadcast MAC address.
566 * MUST reply to only the first fragment.
567 */
568
__icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info,const struct ip_options * opt)569 void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
570 const struct ip_options *opt)
571 {
572 struct iphdr *iph;
573 int room;
574 struct icmp_bxm icmp_param;
575 struct rtable *rt = skb_rtable(skb_in);
576 struct ipcm_cookie ipc;
577 struct flowi4 fl4;
578 __be32 saddr;
579 u8 tos;
580 u32 mark;
581 struct net *net;
582 struct sock *sk;
583
584 if (!rt)
585 goto out;
586
587 if (rt->dst.dev)
588 net = dev_net(rt->dst.dev);
589 else if (skb_in->dev)
590 net = dev_net(skb_in->dev);
591 else
592 goto out;
593
594 /*
595 * Find the original header. It is expected to be valid, of course.
596 * Check this, icmp_send is called from the most obscure devices
597 * sometimes.
598 */
599 iph = ip_hdr(skb_in);
600
601 if ((u8 *)iph < skb_in->head ||
602 (skb_network_header(skb_in) + sizeof(*iph)) >
603 skb_tail_pointer(skb_in))
604 goto out;
605
606 /*
607 * No replies to physical multicast/broadcast
608 */
609 if (skb_in->pkt_type != PACKET_HOST)
610 goto out;
611
612 /*
613 * Now check at the protocol level
614 */
615 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
616 goto out;
617
618 /*
619 * Only reply to fragment 0. We byte re-order the constant
620 * mask for efficiency.
621 */
622 if (iph->frag_off & htons(IP_OFFSET))
623 goto out;
624
625 /*
626 * If we send an ICMP error to an ICMP error a mess would result..
627 */
628 if (icmp_pointers[type].error) {
629 /*
630 * We are an error, check if we are replying to an
631 * ICMP error
632 */
633 if (iph->protocol == IPPROTO_ICMP) {
634 u8 _inner_type, *itp;
635
636 itp = skb_header_pointer(skb_in,
637 skb_network_header(skb_in) +
638 (iph->ihl << 2) +
639 offsetof(struct icmphdr,
640 type) -
641 skb_in->data,
642 sizeof(_inner_type),
643 &_inner_type);
644 if (!itp)
645 goto out;
646
647 /*
648 * Assume any unknown ICMP type is an error. This
649 * isn't specified by the RFC, but think about it..
650 */
651 if (*itp > NR_ICMP_TYPES ||
652 icmp_pointers[*itp].error)
653 goto out;
654 }
655 }
656
657 /* Needed by both icmp_global_allow and icmp_xmit_lock */
658 local_bh_disable();
659
660 /* Check global sysctl_icmp_msgs_per_sec ratelimit, unless
661 * incoming dev is loopback. If outgoing dev change to not be
662 * loopback, then peer ratelimit still work (in icmpv4_xrlim_allow)
663 */
664 if (!(skb_in->dev && (skb_in->dev->flags&IFF_LOOPBACK)) &&
665 !icmpv4_global_allow(net, type, code))
666 goto out_bh_enable;
667
668 sk = icmp_xmit_lock(net);
669 if (!sk)
670 goto out_bh_enable;
671
672 /*
673 * Construct source address and options.
674 */
675
676 saddr = iph->daddr;
677 if (!(rt->rt_flags & RTCF_LOCAL)) {
678 struct net_device *dev = NULL;
679
680 rcu_read_lock();
681 if (rt_is_input_route(rt) &&
682 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
683 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
684
685 if (dev)
686 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
687 else
688 saddr = 0;
689 rcu_read_unlock();
690 }
691
692 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
693 IPTOS_PREC_INTERNETCONTROL) :
694 iph->tos;
695 mark = IP4_REPLY_MARK(net, skb_in->mark);
696
697 if (__ip_options_echo(net, &icmp_param.replyopts.opt.opt, skb_in, opt))
698 goto out_unlock;
699
700
701 /*
702 * Prepare data for ICMP header.
703 */
704
705 icmp_param.data.icmph.type = type;
706 icmp_param.data.icmph.code = code;
707 icmp_param.data.icmph.un.gateway = info;
708 icmp_param.data.icmph.checksum = 0;
709 icmp_param.skb = skb_in;
710 icmp_param.offset = skb_network_offset(skb_in);
711 inet_sk(sk)->tos = tos;
712 sk->sk_mark = mark;
713 ipcm_init(&ipc);
714 ipc.addr = iph->saddr;
715 ipc.opt = &icmp_param.replyopts.opt;
716
717 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
718 type, code, &icmp_param);
719 if (IS_ERR(rt))
720 goto out_unlock;
721
722 /* peer icmp_ratelimit */
723 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
724 goto ende;
725
726 /* RFC says return as much as we can without exceeding 576 bytes. */
727
728 room = dst_mtu(&rt->dst);
729 if (room > 576)
730 room = 576;
731 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
732 room -= sizeof(struct icmphdr);
733
734 icmp_param.data_len = skb_in->len - icmp_param.offset;
735 if (icmp_param.data_len > room)
736 icmp_param.data_len = room;
737 icmp_param.head_len = sizeof(struct icmphdr);
738
739 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
740 ende:
741 ip_rt_put(rt);
742 out_unlock:
743 icmp_xmit_unlock(sk);
744 out_bh_enable:
745 local_bh_enable();
746 out:;
747 }
748 EXPORT_SYMBOL(__icmp_send);
749
750
icmp_socket_deliver(struct sk_buff * skb,u32 info)751 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
752 {
753 const struct iphdr *iph = (const struct iphdr *) skb->data;
754 const struct net_protocol *ipprot;
755 int protocol = iph->protocol;
756
757 /* Checkin full IP header plus 8 bytes of protocol to
758 * avoid additional coding at protocol handlers.
759 */
760 if (!pskb_may_pull(skb, iph->ihl * 4 + 8)) {
761 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
762 return;
763 }
764
765 raw_icmp_error(skb, protocol, info);
766
767 ipprot = rcu_dereference(inet_protos[protocol]);
768 if (ipprot && ipprot->err_handler)
769 ipprot->err_handler(skb, info);
770 }
771
icmp_tag_validation(int proto)772 static bool icmp_tag_validation(int proto)
773 {
774 bool ok;
775
776 rcu_read_lock();
777 ok = rcu_dereference(inet_protos[proto])->icmp_strict_tag_validation;
778 rcu_read_unlock();
779 return ok;
780 }
781
782 /*
783 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEEDED, ICMP_QUENCH, and
784 * ICMP_PARAMETERPROB.
785 */
786
icmp_unreach(struct sk_buff * skb)787 static bool icmp_unreach(struct sk_buff *skb)
788 {
789 const struct iphdr *iph;
790 struct icmphdr *icmph;
791 struct net *net;
792 u32 info = 0;
793
794 net = dev_net(skb_dst(skb)->dev);
795
796 /*
797 * Incomplete header ?
798 * Only checks for the IP header, there should be an
799 * additional check for longer headers in upper levels.
800 */
801
802 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
803 goto out_err;
804
805 icmph = icmp_hdr(skb);
806 iph = (const struct iphdr *)skb->data;
807
808 if (iph->ihl < 5) /* Mangled header, drop. */
809 goto out_err;
810
811 switch (icmph->type) {
812 case ICMP_DEST_UNREACH:
813 switch (icmph->code & 15) {
814 case ICMP_NET_UNREACH:
815 case ICMP_HOST_UNREACH:
816 case ICMP_PROT_UNREACH:
817 case ICMP_PORT_UNREACH:
818 break;
819 case ICMP_FRAG_NEEDED:
820 /* for documentation of the ip_no_pmtu_disc
821 * values please see
822 * Documentation/networking/ip-sysctl.txt
823 */
824 switch (net->ipv4.sysctl_ip_no_pmtu_disc) {
825 default:
826 net_dbg_ratelimited("%pI4: fragmentation needed and DF set\n",
827 &iph->daddr);
828 break;
829 case 2:
830 goto out;
831 case 3:
832 if (!icmp_tag_validation(iph->protocol))
833 goto out;
834 /* fall through */
835 case 0:
836 info = ntohs(icmph->un.frag.mtu);
837 }
838 break;
839 case ICMP_SR_FAILED:
840 net_dbg_ratelimited("%pI4: Source Route Failed\n",
841 &iph->daddr);
842 break;
843 default:
844 break;
845 }
846 if (icmph->code > NR_ICMP_UNREACH)
847 goto out;
848 break;
849 case ICMP_PARAMETERPROB:
850 info = ntohl(icmph->un.gateway) >> 24;
851 break;
852 case ICMP_TIME_EXCEEDED:
853 __ICMP_INC_STATS(net, ICMP_MIB_INTIMEEXCDS);
854 if (icmph->code == ICMP_EXC_FRAGTIME)
855 goto out;
856 break;
857 }
858
859 /*
860 * Throw it at our lower layers
861 *
862 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
863 * header.
864 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
865 * transport layer.
866 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
867 * transport layer.
868 */
869
870 /*
871 * Check the other end isn't violating RFC 1122. Some routers send
872 * bogus responses to broadcast frames. If you see this message
873 * first check your netmask matches at both ends, if it does then
874 * get the other vendor to fix their kit.
875 */
876
877 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
878 inet_addr_type_dev_table(net, skb->dev, iph->daddr) == RTN_BROADCAST) {
879 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
880 &ip_hdr(skb)->saddr,
881 icmph->type, icmph->code,
882 &iph->daddr, skb->dev->name);
883 goto out;
884 }
885
886 icmp_socket_deliver(skb, info);
887
888 out:
889 return true;
890 out_err:
891 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
892 return false;
893 }
894
895
896 /*
897 * Handle ICMP_REDIRECT.
898 */
899
icmp_redirect(struct sk_buff * skb)900 static bool icmp_redirect(struct sk_buff *skb)
901 {
902 if (skb->len < sizeof(struct iphdr)) {
903 __ICMP_INC_STATS(dev_net(skb->dev), ICMP_MIB_INERRORS);
904 return false;
905 }
906
907 if (!pskb_may_pull(skb, sizeof(struct iphdr))) {
908 /* there aught to be a stat */
909 return false;
910 }
911
912 icmp_socket_deliver(skb, ntohl(icmp_hdr(skb)->un.gateway));
913 return true;
914 }
915
916 /*
917 * Handle ICMP_ECHO ("ping") requests.
918 *
919 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
920 * requests.
921 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
922 * included in the reply.
923 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
924 * echo requests, MUST have default=NOT.
925 * See also WRT handling of options once they are done and working.
926 */
927
icmp_echo(struct sk_buff * skb)928 static bool icmp_echo(struct sk_buff *skb)
929 {
930 struct net *net;
931
932 net = dev_net(skb_dst(skb)->dev);
933 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
934 struct icmp_bxm icmp_param;
935
936 icmp_param.data.icmph = *icmp_hdr(skb);
937 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
938 icmp_param.skb = skb;
939 icmp_param.offset = 0;
940 icmp_param.data_len = skb->len;
941 icmp_param.head_len = sizeof(struct icmphdr);
942 icmp_reply(&icmp_param, skb);
943 }
944 /* should there be an ICMP stat for ignored echos? */
945 return true;
946 }
947
948 /*
949 * Handle ICMP Timestamp requests.
950 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
951 * SHOULD be in the kernel for minimum random latency.
952 * MUST be accurate to a few minutes.
953 * MUST be updated at least at 15Hz.
954 */
icmp_timestamp(struct sk_buff * skb)955 static bool icmp_timestamp(struct sk_buff *skb)
956 {
957 struct icmp_bxm icmp_param;
958 /*
959 * Too short.
960 */
961 if (skb->len < 4)
962 goto out_err;
963
964 /*
965 * Fill in the current time as ms since midnight UT:
966 */
967 icmp_param.data.times[1] = inet_current_timestamp();
968 icmp_param.data.times[2] = icmp_param.data.times[1];
969
970 BUG_ON(skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4));
971
972 icmp_param.data.icmph = *icmp_hdr(skb);
973 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
974 icmp_param.data.icmph.code = 0;
975 icmp_param.skb = skb;
976 icmp_param.offset = 0;
977 icmp_param.data_len = 0;
978 icmp_param.head_len = sizeof(struct icmphdr) + 12;
979 icmp_reply(&icmp_param, skb);
980 return true;
981
982 out_err:
983 __ICMP_INC_STATS(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
984 return false;
985 }
986
icmp_discard(struct sk_buff * skb)987 static bool icmp_discard(struct sk_buff *skb)
988 {
989 /* pretend it was a success */
990 return true;
991 }
992
993 /*
994 * Deal with incoming ICMP packets.
995 */
icmp_rcv(struct sk_buff * skb)996 int icmp_rcv(struct sk_buff *skb)
997 {
998 struct icmphdr *icmph;
999 struct rtable *rt = skb_rtable(skb);
1000 struct net *net = dev_net(rt->dst.dev);
1001 bool success;
1002
1003 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1004 struct sec_path *sp = skb_sec_path(skb);
1005 int nh;
1006
1007 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
1008 XFRM_STATE_ICMP))
1009 goto drop;
1010
1011 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
1012 goto drop;
1013
1014 nh = skb_network_offset(skb);
1015 skb_set_network_header(skb, sizeof(*icmph));
1016
1017 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
1018 goto drop;
1019
1020 skb_set_network_header(skb, nh);
1021 }
1022
1023 __ICMP_INC_STATS(net, ICMP_MIB_INMSGS);
1024
1025 if (skb_checksum_simple_validate(skb))
1026 goto csum_error;
1027
1028 if (!pskb_pull(skb, sizeof(*icmph)))
1029 goto error;
1030
1031 icmph = icmp_hdr(skb);
1032
1033 ICMPMSGIN_INC_STATS(net, icmph->type);
1034 /*
1035 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1036 *
1037 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1038 * discarded.
1039 */
1040 if (icmph->type > NR_ICMP_TYPES)
1041 goto error;
1042
1043
1044 /*
1045 * Parse the ICMP message
1046 */
1047
1048 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1049 /*
1050 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1051 * silently ignored (we let user decide with a sysctl).
1052 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1053 * discarded if to broadcast/multicast.
1054 */
1055 if ((icmph->type == ICMP_ECHO ||
1056 icmph->type == ICMP_TIMESTAMP) &&
1057 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1058 goto error;
1059 }
1060 if (icmph->type != ICMP_ECHO &&
1061 icmph->type != ICMP_TIMESTAMP &&
1062 icmph->type != ICMP_ADDRESS &&
1063 icmph->type != ICMP_ADDRESSREPLY) {
1064 goto error;
1065 }
1066 }
1067
1068 success = icmp_pointers[icmph->type].handler(skb);
1069
1070 if (success) {
1071 consume_skb(skb);
1072 return NET_RX_SUCCESS;
1073 }
1074
1075 drop:
1076 kfree_skb(skb);
1077 return NET_RX_DROP;
1078 csum_error:
1079 __ICMP_INC_STATS(net, ICMP_MIB_CSUMERRORS);
1080 error:
1081 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
1082 goto drop;
1083 }
1084
icmp_err(struct sk_buff * skb,u32 info)1085 int icmp_err(struct sk_buff *skb, u32 info)
1086 {
1087 struct iphdr *iph = (struct iphdr *)skb->data;
1088 int offset = iph->ihl<<2;
1089 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
1090 int type = icmp_hdr(skb)->type;
1091 int code = icmp_hdr(skb)->code;
1092 struct net *net = dev_net(skb->dev);
1093
1094 /*
1095 * Use ping_err to handle all icmp errors except those
1096 * triggered by ICMP_ECHOREPLY which sent from kernel.
1097 */
1098 if (icmph->type != ICMP_ECHOREPLY) {
1099 ping_err(skb, offset, info);
1100 return 0;
1101 }
1102
1103 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
1104 ipv4_update_pmtu(skb, net, info, 0, IPPROTO_ICMP);
1105 else if (type == ICMP_REDIRECT)
1106 ipv4_redirect(skb, net, 0, IPPROTO_ICMP);
1107
1108 return 0;
1109 }
1110
1111 /*
1112 * This table is the definition of how we handle ICMP.
1113 */
1114 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1115 [ICMP_ECHOREPLY] = {
1116 .handler = ping_rcv,
1117 },
1118 [1] = {
1119 .handler = icmp_discard,
1120 .error = 1,
1121 },
1122 [2] = {
1123 .handler = icmp_discard,
1124 .error = 1,
1125 },
1126 [ICMP_DEST_UNREACH] = {
1127 .handler = icmp_unreach,
1128 .error = 1,
1129 },
1130 [ICMP_SOURCE_QUENCH] = {
1131 .handler = icmp_unreach,
1132 .error = 1,
1133 },
1134 [ICMP_REDIRECT] = {
1135 .handler = icmp_redirect,
1136 .error = 1,
1137 },
1138 [6] = {
1139 .handler = icmp_discard,
1140 .error = 1,
1141 },
1142 [7] = {
1143 .handler = icmp_discard,
1144 .error = 1,
1145 },
1146 [ICMP_ECHO] = {
1147 .handler = icmp_echo,
1148 },
1149 [9] = {
1150 .handler = icmp_discard,
1151 .error = 1,
1152 },
1153 [10] = {
1154 .handler = icmp_discard,
1155 .error = 1,
1156 },
1157 [ICMP_TIME_EXCEEDED] = {
1158 .handler = icmp_unreach,
1159 .error = 1,
1160 },
1161 [ICMP_PARAMETERPROB] = {
1162 .handler = icmp_unreach,
1163 .error = 1,
1164 },
1165 [ICMP_TIMESTAMP] = {
1166 .handler = icmp_timestamp,
1167 },
1168 [ICMP_TIMESTAMPREPLY] = {
1169 .handler = icmp_discard,
1170 },
1171 [ICMP_INFO_REQUEST] = {
1172 .handler = icmp_discard,
1173 },
1174 [ICMP_INFO_REPLY] = {
1175 .handler = icmp_discard,
1176 },
1177 [ICMP_ADDRESS] = {
1178 .handler = icmp_discard,
1179 },
1180 [ICMP_ADDRESSREPLY] = {
1181 .handler = icmp_discard,
1182 },
1183 };
1184
icmp_sk_exit(struct net * net)1185 static void __net_exit icmp_sk_exit(struct net *net)
1186 {
1187 int i;
1188
1189 for_each_possible_cpu(i)
1190 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
1191 free_percpu(net->ipv4.icmp_sk);
1192 net->ipv4.icmp_sk = NULL;
1193 }
1194
icmp_sk_init(struct net * net)1195 static int __net_init icmp_sk_init(struct net *net)
1196 {
1197 int i, err;
1198
1199 net->ipv4.icmp_sk = alloc_percpu(struct sock *);
1200 if (!net->ipv4.icmp_sk)
1201 return -ENOMEM;
1202
1203 for_each_possible_cpu(i) {
1204 struct sock *sk;
1205
1206 err = inet_ctl_sock_create(&sk, PF_INET,
1207 SOCK_RAW, IPPROTO_ICMP, net);
1208 if (err < 0)
1209 goto fail;
1210
1211 *per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
1212
1213 /* Enough space for 2 64K ICMP packets, including
1214 * sk_buff/skb_shared_info struct overhead.
1215 */
1216 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1217
1218 /*
1219 * Speedup sock_wfree()
1220 */
1221 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1222 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1223 }
1224
1225 /* Control parameters for ECHO replies. */
1226 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1227 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1228
1229 /* Control parameter - ignore bogus broadcast responses? */
1230 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1231
1232 /*
1233 * Configurable global rate limit.
1234 *
1235 * ratelimit defines tokens/packet consumed for dst->rate_token
1236 * bucket ratemask defines which icmp types are ratelimited by
1237 * setting it's bit position.
1238 *
1239 * default:
1240 * dest unreachable (3), source quench (4),
1241 * time exceeded (11), parameter problem (12)
1242 */
1243
1244 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1245 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1246 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1247
1248 return 0;
1249
1250 fail:
1251 icmp_sk_exit(net);
1252 return err;
1253 }
1254
1255 static struct pernet_operations __net_initdata icmp_sk_ops = {
1256 .init = icmp_sk_init,
1257 .exit = icmp_sk_exit,
1258 };
1259
icmp_init(void)1260 int __init icmp_init(void)
1261 {
1262 return register_pernet_subsys(&icmp_sk_ops);
1263 }
1264