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