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 /*
235 * Send an ICMP frame.
236 */
237
icmpv4_xrlim_allow(struct net * net,struct rtable * rt,struct flowi4 * fl4,int type,int code)238 static inline bool icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
239 struct flowi4 *fl4, int type, int code)
240 {
241 struct dst_entry *dst = &rt->dst;
242 bool rc = true;
243
244 if (type > NR_ICMP_TYPES)
245 goto out;
246
247 /* Don't limit PMTU discovery. */
248 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
249 goto out;
250
251 /* No rate limit on loopback */
252 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
253 goto out;
254
255 /* Limit if icmp type is enabled in ratemask. */
256 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask) {
257 struct inet_peer *peer = inet_getpeer_v4(net->ipv4.peers, fl4->daddr, 1);
258 rc = inet_peer_xrlim_allow(peer,
259 net->ipv4.sysctl_icmp_ratelimit);
260 if (peer)
261 inet_putpeer(peer);
262 }
263 out:
264 return rc;
265 }
266
267 /*
268 * Maintain the counters used in the SNMP statistics for outgoing ICMP
269 */
icmp_out_count(struct net * net,unsigned char type)270 void icmp_out_count(struct net *net, unsigned char type)
271 {
272 ICMPMSGOUT_INC_STATS(net, type);
273 ICMP_INC_STATS(net, ICMP_MIB_OUTMSGS);
274 }
275
276 /*
277 * Checksum each fragment, and on the first include the headers and final
278 * checksum.
279 */
icmp_glue_bits(void * from,char * to,int offset,int len,int odd,struct sk_buff * skb)280 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
281 struct sk_buff *skb)
282 {
283 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
284 __wsum csum;
285
286 csum = skb_copy_and_csum_bits(icmp_param->skb,
287 icmp_param->offset + offset,
288 to, len, 0);
289
290 skb->csum = csum_block_add(skb->csum, csum, odd);
291 if (icmp_pointers[icmp_param->data.icmph.type].error)
292 nf_ct_attach(skb, icmp_param->skb);
293 return 0;
294 }
295
icmp_push_reply(struct icmp_bxm * icmp_param,struct flowi4 * fl4,struct ipcm_cookie * ipc,struct rtable ** rt)296 static void icmp_push_reply(struct icmp_bxm *icmp_param,
297 struct flowi4 *fl4,
298 struct ipcm_cookie *ipc, struct rtable **rt)
299 {
300 struct sock *sk;
301 struct sk_buff *skb;
302
303 sk = icmp_sk(dev_net((*rt)->dst.dev));
304 if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
305 icmp_param->data_len+icmp_param->head_len,
306 icmp_param->head_len,
307 ipc, rt, MSG_DONTWAIT) < 0) {
308 ICMP_INC_STATS_BH(sock_net(sk), ICMP_MIB_OUTERRORS);
309 ip_flush_pending_frames(sk);
310 } else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
311 struct icmphdr *icmph = icmp_hdr(skb);
312 __wsum csum = 0;
313 struct sk_buff *skb1;
314
315 skb_queue_walk(&sk->sk_write_queue, skb1) {
316 csum = csum_add(csum, skb1->csum);
317 }
318 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
319 (char *)icmph,
320 icmp_param->head_len, csum);
321 icmph->checksum = csum_fold(csum);
322 skb->ip_summed = CHECKSUM_NONE;
323 ip_push_pending_frames(sk, fl4);
324 }
325 }
326
327 /*
328 * Driving logic for building and sending ICMP messages.
329 */
330
icmp_reply(struct icmp_bxm * icmp_param,struct sk_buff * skb)331 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
332 {
333 struct ipcm_cookie ipc;
334 struct rtable *rt = skb_rtable(skb);
335 struct net *net = dev_net(rt->dst.dev);
336 struct flowi4 fl4;
337 struct sock *sk;
338 struct inet_sock *inet;
339 __be32 daddr, saddr;
340 u32 mark = IP4_REPLY_MARK(net, skb->mark);
341
342 if (ip_options_echo(&icmp_param->replyopts.opt.opt, skb))
343 return;
344
345 sk = icmp_xmit_lock(net);
346 if (sk == NULL)
347 return;
348 inet = inet_sk(sk);
349
350 icmp_param->data.icmph.checksum = 0;
351
352 inet->tos = ip_hdr(skb)->tos;
353 sk->sk_mark = mark;
354 daddr = ipc.addr = ip_hdr(skb)->saddr;
355 saddr = fib_compute_spec_dst(skb);
356 ipc.opt = NULL;
357 ipc.tx_flags = 0;
358 if (icmp_param->replyopts.opt.opt.optlen) {
359 ipc.opt = &icmp_param->replyopts.opt;
360 if (ipc.opt->opt.srr)
361 daddr = icmp_param->replyopts.opt.opt.faddr;
362 }
363 memset(&fl4, 0, sizeof(fl4));
364 fl4.daddr = daddr;
365 fl4.saddr = saddr;
366 fl4.flowi4_mark = mark;
367 fl4.flowi4_uid = sock_net_uid(net, NULL);
368 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
369 fl4.flowi4_proto = IPPROTO_ICMP;
370 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
371 rt = ip_route_output_key(net, &fl4);
372 if (IS_ERR(rt))
373 goto out_unlock;
374 if (icmpv4_xrlim_allow(net, rt, &fl4, icmp_param->data.icmph.type,
375 icmp_param->data.icmph.code))
376 icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
377 ip_rt_put(rt);
378 out_unlock:
379 icmp_xmit_unlock(sk);
380 }
381
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)382 static struct rtable *icmp_route_lookup(struct net *net,
383 struct flowi4 *fl4,
384 struct sk_buff *skb_in,
385 const struct iphdr *iph,
386 __be32 saddr, u8 tos, u32 mark,
387 int type, int code,
388 struct icmp_bxm *param)
389 {
390 struct rtable *rt, *rt2;
391 struct flowi4 fl4_dec;
392 int err;
393
394 memset(fl4, 0, sizeof(*fl4));
395 fl4->daddr = (param->replyopts.opt.opt.srr ?
396 param->replyopts.opt.opt.faddr : iph->saddr);
397 fl4->saddr = saddr;
398 fl4->flowi4_mark = mark;
399 fl4->flowi4_uid = sock_net_uid(net, NULL);
400 fl4->flowi4_tos = RT_TOS(tos);
401 fl4->flowi4_proto = IPPROTO_ICMP;
402 fl4->fl4_icmp_type = type;
403 fl4->fl4_icmp_code = code;
404 security_skb_classify_flow(skb_in, flowi4_to_flowi(fl4));
405 rt = __ip_route_output_key(net, fl4);
406 if (IS_ERR(rt))
407 return rt;
408
409 /* No need to clone since we're just using its address. */
410 rt2 = rt;
411
412 rt = (struct rtable *) xfrm_lookup(net, &rt->dst,
413 flowi4_to_flowi(fl4), NULL, 0);
414 if (!IS_ERR(rt)) {
415 if (rt != rt2)
416 return rt;
417 } else if (PTR_ERR(rt) == -EPERM) {
418 rt = NULL;
419 } else
420 return rt;
421
422 err = xfrm_decode_session_reverse(skb_in, flowi4_to_flowi(&fl4_dec), AF_INET);
423 if (err)
424 goto relookup_failed;
425
426 if (inet_addr_type(net, fl4_dec.saddr) == RTN_LOCAL) {
427 rt2 = __ip_route_output_key(net, &fl4_dec);
428 if (IS_ERR(rt2))
429 err = PTR_ERR(rt2);
430 } else {
431 struct flowi4 fl4_2 = {};
432 unsigned long orefdst;
433
434 fl4_2.daddr = fl4_dec.saddr;
435 rt2 = ip_route_output_key(net, &fl4_2);
436 if (IS_ERR(rt2)) {
437 err = PTR_ERR(rt2);
438 goto relookup_failed;
439 }
440 /* Ugh! */
441 orefdst = skb_in->_skb_refdst; /* save old refdst */
442 err = ip_route_input(skb_in, fl4_dec.daddr, fl4_dec.saddr,
443 RT_TOS(tos), rt2->dst.dev);
444
445 dst_release(&rt2->dst);
446 rt2 = skb_rtable(skb_in);
447 skb_in->_skb_refdst = orefdst; /* restore old refdst */
448 }
449
450 if (err)
451 goto relookup_failed;
452
453 rt2 = (struct rtable *) xfrm_lookup(net, &rt2->dst,
454 flowi4_to_flowi(&fl4_dec), NULL,
455 XFRM_LOOKUP_ICMP);
456 if (!IS_ERR(rt2)) {
457 dst_release(&rt->dst);
458 memcpy(fl4, &fl4_dec, sizeof(*fl4));
459 rt = rt2;
460 } else if (PTR_ERR(rt2) == -EPERM) {
461 if (rt)
462 dst_release(&rt->dst);
463 return rt2;
464 } else {
465 err = PTR_ERR(rt2);
466 goto relookup_failed;
467 }
468 return rt;
469
470 relookup_failed:
471 if (rt)
472 return rt;
473 return ERR_PTR(err);
474 }
475
476 /*
477 * Send an ICMP message in response to a situation
478 *
479 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
480 * MAY send more (we do).
481 * MUST NOT change this header information.
482 * MUST NOT reply to a multicast/broadcast IP address.
483 * MUST NOT reply to a multicast/broadcast MAC address.
484 * MUST reply to only the first fragment.
485 */
486
icmp_send(struct sk_buff * skb_in,int type,int code,__be32 info)487 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
488 {
489 struct iphdr *iph;
490 int room;
491 struct icmp_bxm icmp_param;
492 struct rtable *rt = skb_rtable(skb_in);
493 struct ipcm_cookie ipc;
494 struct flowi4 fl4;
495 __be32 saddr;
496 u8 tos;
497 u32 mark;
498 struct net *net;
499 struct sock *sk;
500
501 if (!rt)
502 goto out;
503 net = dev_net(rt->dst.dev);
504
505 /*
506 * Find the original header. It is expected to be valid, of course.
507 * Check this, icmp_send is called from the most obscure devices
508 * sometimes.
509 */
510 iph = ip_hdr(skb_in);
511
512 if ((u8 *)iph < skb_in->head ||
513 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
514 goto out;
515
516 /*
517 * No replies to physical multicast/broadcast
518 */
519 if (skb_in->pkt_type != PACKET_HOST)
520 goto out;
521
522 /*
523 * Now check at the protocol level
524 */
525 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
526 goto out;
527
528 /*
529 * Only reply to fragment 0. We byte re-order the constant
530 * mask for efficiency.
531 */
532 if (iph->frag_off & htons(IP_OFFSET))
533 goto out;
534
535 /*
536 * If we send an ICMP error to an ICMP error a mess would result..
537 */
538 if (icmp_pointers[type].error) {
539 /*
540 * We are an error, check if we are replying to an
541 * ICMP error
542 */
543 if (iph->protocol == IPPROTO_ICMP) {
544 u8 _inner_type, *itp;
545
546 itp = skb_header_pointer(skb_in,
547 skb_network_header(skb_in) +
548 (iph->ihl << 2) +
549 offsetof(struct icmphdr,
550 type) -
551 skb_in->data,
552 sizeof(_inner_type),
553 &_inner_type);
554 if (itp == NULL)
555 goto out;
556
557 /*
558 * Assume any unknown ICMP type is an error. This
559 * isn't specified by the RFC, but think about it..
560 */
561 if (*itp > NR_ICMP_TYPES ||
562 icmp_pointers[*itp].error)
563 goto out;
564 }
565 }
566
567 sk = icmp_xmit_lock(net);
568 if (sk == NULL)
569 return;
570
571 /*
572 * Construct source address and options.
573 */
574
575 saddr = iph->daddr;
576 if (!(rt->rt_flags & RTCF_LOCAL)) {
577 struct net_device *dev = NULL;
578
579 rcu_read_lock();
580 if (rt_is_input_route(rt) &&
581 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
582 dev = dev_get_by_index_rcu(net, inet_iif(skb_in));
583
584 if (dev)
585 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
586 else
587 saddr = 0;
588 rcu_read_unlock();
589 }
590
591 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
592 IPTOS_PREC_INTERNETCONTROL) :
593 iph->tos;
594 mark = IP4_REPLY_MARK(net, skb_in->mark);
595
596 if (ip_options_echo(&icmp_param.replyopts.opt.opt, skb_in))
597 goto out_unlock;
598
599
600 /*
601 * Prepare data for ICMP header.
602 */
603
604 icmp_param.data.icmph.type = type;
605 icmp_param.data.icmph.code = code;
606 icmp_param.data.icmph.un.gateway = info;
607 icmp_param.data.icmph.checksum = 0;
608 icmp_param.skb = skb_in;
609 icmp_param.offset = skb_network_offset(skb_in);
610 inet_sk(sk)->tos = tos;
611 sk->sk_mark = mark;
612 ipc.addr = iph->saddr;
613 ipc.opt = &icmp_param.replyopts.opt;
614 ipc.tx_flags = 0;
615
616 rt = icmp_route_lookup(net, &fl4, skb_in, iph, saddr, tos, mark,
617 type, code, &icmp_param);
618 if (IS_ERR(rt))
619 goto out_unlock;
620
621 if (!icmpv4_xrlim_allow(net, rt, &fl4, type, code))
622 goto ende;
623
624 /* RFC says return as much as we can without exceeding 576 bytes. */
625
626 room = dst_mtu(&rt->dst);
627 if (room > 576)
628 room = 576;
629 room -= sizeof(struct iphdr) + icmp_param.replyopts.opt.opt.optlen;
630 room -= sizeof(struct icmphdr);
631
632 icmp_param.data_len = skb_in->len - icmp_param.offset;
633 if (icmp_param.data_len > room)
634 icmp_param.data_len = room;
635 icmp_param.head_len = sizeof(struct icmphdr);
636
637 icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
638 ende:
639 ip_rt_put(rt);
640 out_unlock:
641 icmp_xmit_unlock(sk);
642 out:;
643 }
644 EXPORT_SYMBOL(icmp_send);
645
646
icmp_socket_deliver(struct sk_buff * skb,u32 info)647 static void icmp_socket_deliver(struct sk_buff *skb, u32 info)
648 {
649 const struct iphdr *iph = (const struct iphdr *) skb->data;
650 const struct net_protocol *ipprot;
651 int protocol = iph->protocol;
652
653 /* Checkin full IP header plus 8 bytes of protocol to
654 * avoid additional coding at protocol handlers.
655 */
656 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
657 return;
658
659 raw_icmp_error(skb, protocol, info);
660
661 rcu_read_lock();
662 ipprot = rcu_dereference(inet_protos[protocol]);
663 if (ipprot && ipprot->err_handler)
664 ipprot->err_handler(skb, info);
665 rcu_read_unlock();
666 }
667
668 /*
669 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
670 */
671
icmp_unreach(struct sk_buff * skb)672 static void icmp_unreach(struct sk_buff *skb)
673 {
674 const struct iphdr *iph;
675 struct icmphdr *icmph;
676 struct net *net;
677 u32 info = 0;
678
679 net = dev_net(skb_dst(skb)->dev);
680
681 /*
682 * Incomplete header ?
683 * Only checks for the IP header, there should be an
684 * additional check for longer headers in upper levels.
685 */
686
687 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
688 goto out_err;
689
690 icmph = icmp_hdr(skb);
691 iph = (const struct iphdr *)skb->data;
692
693 if (iph->ihl < 5) /* Mangled header, drop. */
694 goto out_err;
695
696 if (icmph->type == ICMP_DEST_UNREACH) {
697 switch (icmph->code & 15) {
698 case ICMP_NET_UNREACH:
699 case ICMP_HOST_UNREACH:
700 case ICMP_PROT_UNREACH:
701 case ICMP_PORT_UNREACH:
702 break;
703 case ICMP_FRAG_NEEDED:
704 if (ipv4_config.no_pmtu_disc) {
705 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: fragmentation needed and DF set\n"),
706 &iph->daddr);
707 } else {
708 info = ntohs(icmph->un.frag.mtu);
709 if (!info)
710 goto out;
711 }
712 break;
713 case ICMP_SR_FAILED:
714 LIMIT_NETDEBUG(KERN_INFO pr_fmt("%pI4: Source Route Failed\n"),
715 &iph->daddr);
716 break;
717 default:
718 break;
719 }
720 if (icmph->code > NR_ICMP_UNREACH)
721 goto out;
722 } else if (icmph->type == ICMP_PARAMETERPROB)
723 info = ntohl(icmph->un.gateway) >> 24;
724
725 /*
726 * Throw it at our lower layers
727 *
728 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
729 * header.
730 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
731 * transport layer.
732 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
733 * transport layer.
734 */
735
736 /*
737 * Check the other end isn't violating RFC 1122. Some routers send
738 * bogus responses to broadcast frames. If you see this message
739 * first check your netmask matches at both ends, if it does then
740 * get the other vendor to fix their kit.
741 */
742
743 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
744 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
745 net_warn_ratelimited("%pI4 sent an invalid ICMP type %u, code %u error to a broadcast: %pI4 on %s\n",
746 &ip_hdr(skb)->saddr,
747 icmph->type, icmph->code,
748 &iph->daddr, skb->dev->name);
749 goto out;
750 }
751
752 icmp_socket_deliver(skb, info);
753
754 out:
755 return;
756 out_err:
757 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
758 goto out;
759 }
760
761
762 /*
763 * Handle ICMP_REDIRECT.
764 */
765
icmp_redirect(struct sk_buff * skb)766 static void icmp_redirect(struct sk_buff *skb)
767 {
768 if (skb->len < sizeof(struct iphdr)) {
769 ICMP_INC_STATS_BH(dev_net(skb->dev), ICMP_MIB_INERRORS);
770 return;
771 }
772
773 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
774 return;
775
776 icmp_socket_deliver(skb, icmp_hdr(skb)->un.gateway);
777 }
778
779 /*
780 * Handle ICMP_ECHO ("ping") requests.
781 *
782 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
783 * requests.
784 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
785 * included in the reply.
786 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
787 * echo requests, MUST have default=NOT.
788 * See also WRT handling of options once they are done and working.
789 */
790
icmp_echo(struct sk_buff * skb)791 static void icmp_echo(struct sk_buff *skb)
792 {
793 struct net *net;
794
795 net = dev_net(skb_dst(skb)->dev);
796 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
797 struct icmp_bxm icmp_param;
798
799 icmp_param.data.icmph = *icmp_hdr(skb);
800 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
801 icmp_param.skb = skb;
802 icmp_param.offset = 0;
803 icmp_param.data_len = skb->len;
804 icmp_param.head_len = sizeof(struct icmphdr);
805 icmp_reply(&icmp_param, skb);
806 }
807 }
808
809 /*
810 * Handle ICMP Timestamp requests.
811 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
812 * SHOULD be in the kernel for minimum random latency.
813 * MUST be accurate to a few minutes.
814 * MUST be updated at least at 15Hz.
815 */
icmp_timestamp(struct sk_buff * skb)816 static void icmp_timestamp(struct sk_buff *skb)
817 {
818 struct timespec tv;
819 struct icmp_bxm icmp_param;
820 /*
821 * Too short.
822 */
823 if (skb->len < 4)
824 goto out_err;
825
826 /*
827 * Fill in the current time as ms since midnight UT:
828 */
829 getnstimeofday(&tv);
830 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
831 tv.tv_nsec / NSEC_PER_MSEC);
832 icmp_param.data.times[2] = icmp_param.data.times[1];
833 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
834 BUG();
835 icmp_param.data.icmph = *icmp_hdr(skb);
836 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
837 icmp_param.data.icmph.code = 0;
838 icmp_param.skb = skb;
839 icmp_param.offset = 0;
840 icmp_param.data_len = 0;
841 icmp_param.head_len = sizeof(struct icmphdr) + 12;
842 icmp_reply(&icmp_param, skb);
843 out:
844 return;
845 out_err:
846 ICMP_INC_STATS_BH(dev_net(skb_dst(skb)->dev), ICMP_MIB_INERRORS);
847 goto out;
848 }
849
icmp_discard(struct sk_buff * skb)850 static void icmp_discard(struct sk_buff *skb)
851 {
852 }
853
854 /*
855 * Deal with incoming ICMP packets.
856 */
icmp_rcv(struct sk_buff * skb)857 int icmp_rcv(struct sk_buff *skb)
858 {
859 struct icmphdr *icmph;
860 struct rtable *rt = skb_rtable(skb);
861 struct net *net = dev_net(rt->dst.dev);
862
863 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
864 struct sec_path *sp = skb_sec_path(skb);
865 int nh;
866
867 if (!(sp && sp->xvec[sp->len - 1]->props.flags &
868 XFRM_STATE_ICMP))
869 goto drop;
870
871 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
872 goto drop;
873
874 nh = skb_network_offset(skb);
875 skb_set_network_header(skb, sizeof(*icmph));
876
877 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
878 goto drop;
879
880 skb_set_network_header(skb, nh);
881 }
882
883 ICMP_INC_STATS_BH(net, ICMP_MIB_INMSGS);
884
885 switch (skb->ip_summed) {
886 case CHECKSUM_COMPLETE:
887 if (!csum_fold(skb->csum))
888 break;
889 /* fall through */
890 case CHECKSUM_NONE:
891 skb->csum = 0;
892 if (__skb_checksum_complete(skb))
893 goto csum_error;
894 }
895
896 if (!pskb_pull(skb, sizeof(*icmph)))
897 goto error;
898
899 icmph = icmp_hdr(skb);
900
901 ICMPMSGIN_INC_STATS_BH(net, icmph->type);
902 /*
903 * 18 is the highest 'known' ICMP type. Anything else is a mystery
904 *
905 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
906 * discarded.
907 */
908 if (icmph->type > NR_ICMP_TYPES)
909 goto error;
910
911
912 /*
913 * Parse the ICMP message
914 */
915
916 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
917 /*
918 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
919 * silently ignored (we let user decide with a sysctl).
920 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
921 * discarded if to broadcast/multicast.
922 */
923 if ((icmph->type == ICMP_ECHO ||
924 icmph->type == ICMP_TIMESTAMP) &&
925 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
926 goto error;
927 }
928 if (icmph->type != ICMP_ECHO &&
929 icmph->type != ICMP_TIMESTAMP &&
930 icmph->type != ICMP_ADDRESS &&
931 icmph->type != ICMP_ADDRESSREPLY) {
932 goto error;
933 }
934 }
935
936 icmp_pointers[icmph->type].handler(skb);
937
938 drop:
939 kfree_skb(skb);
940 return 0;
941 csum_error:
942 ICMP_INC_STATS_BH(net, ICMP_MIB_CSUMERRORS);
943 error:
944 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
945 goto drop;
946 }
947
icmp_err(struct sk_buff * skb,u32 info)948 void icmp_err(struct sk_buff *skb, u32 info)
949 {
950 struct iphdr *iph = (struct iphdr *)skb->data;
951 int offset = iph->ihl<<2;
952 struct icmphdr *icmph = (struct icmphdr *)(skb->data + offset);
953 int type = icmp_hdr(skb)->type;
954 int code = icmp_hdr(skb)->code;
955 struct net *net = dev_net(skb->dev);
956
957 /*
958 * Use ping_err to handle all icmp errors except those
959 * triggered by ICMP_ECHOREPLY which sent from kernel.
960 */
961 if (icmph->type != ICMP_ECHOREPLY) {
962 ping_err(skb, offset, info);
963 return;
964 }
965
966 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
967 ipv4_update_pmtu(skb, net, info, 0, 0, IPPROTO_ICMP, 0);
968 else if (type == ICMP_REDIRECT)
969 ipv4_redirect(skb, net, 0, 0, IPPROTO_ICMP, 0);
970 }
971
972 /*
973 * This table is the definition of how we handle ICMP.
974 */
975 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
976 [ICMP_ECHOREPLY] = {
977 .handler = ping_rcv,
978 },
979 [1] = {
980 .handler = icmp_discard,
981 .error = 1,
982 },
983 [2] = {
984 .handler = icmp_discard,
985 .error = 1,
986 },
987 [ICMP_DEST_UNREACH] = {
988 .handler = icmp_unreach,
989 .error = 1,
990 },
991 [ICMP_SOURCE_QUENCH] = {
992 .handler = icmp_unreach,
993 .error = 1,
994 },
995 [ICMP_REDIRECT] = {
996 .handler = icmp_redirect,
997 .error = 1,
998 },
999 [6] = {
1000 .handler = icmp_discard,
1001 .error = 1,
1002 },
1003 [7] = {
1004 .handler = icmp_discard,
1005 .error = 1,
1006 },
1007 [ICMP_ECHO] = {
1008 .handler = icmp_echo,
1009 },
1010 [9] = {
1011 .handler = icmp_discard,
1012 .error = 1,
1013 },
1014 [10] = {
1015 .handler = icmp_discard,
1016 .error = 1,
1017 },
1018 [ICMP_TIME_EXCEEDED] = {
1019 .handler = icmp_unreach,
1020 .error = 1,
1021 },
1022 [ICMP_PARAMETERPROB] = {
1023 .handler = icmp_unreach,
1024 .error = 1,
1025 },
1026 [ICMP_TIMESTAMP] = {
1027 .handler = icmp_timestamp,
1028 },
1029 [ICMP_TIMESTAMPREPLY] = {
1030 .handler = icmp_discard,
1031 },
1032 [ICMP_INFO_REQUEST] = {
1033 .handler = icmp_discard,
1034 },
1035 [ICMP_INFO_REPLY] = {
1036 .handler = icmp_discard,
1037 },
1038 [ICMP_ADDRESS] = {
1039 .handler = icmp_discard,
1040 },
1041 [ICMP_ADDRESSREPLY] = {
1042 .handler = icmp_discard,
1043 },
1044 };
1045
icmp_sk_exit(struct net * net)1046 static void __net_exit icmp_sk_exit(struct net *net)
1047 {
1048 int i;
1049
1050 for_each_possible_cpu(i)
1051 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1052 kfree(net->ipv4.icmp_sk);
1053 net->ipv4.icmp_sk = NULL;
1054 }
1055
icmp_sk_init(struct net * net)1056 static int __net_init icmp_sk_init(struct net *net)
1057 {
1058 int i, err;
1059
1060 net->ipv4.icmp_sk =
1061 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1062 if (net->ipv4.icmp_sk == NULL)
1063 return -ENOMEM;
1064
1065 for_each_possible_cpu(i) {
1066 struct sock *sk;
1067
1068 err = inet_ctl_sock_create(&sk, PF_INET,
1069 SOCK_RAW, IPPROTO_ICMP, net);
1070 if (err < 0)
1071 goto fail;
1072
1073 net->ipv4.icmp_sk[i] = sk;
1074
1075 /* Enough space for 2 64K ICMP packets, including
1076 * sk_buff/skb_shared_info struct overhead.
1077 */
1078 sk->sk_sndbuf = 2 * SKB_TRUESIZE(64 * 1024);
1079
1080 /*
1081 * Speedup sock_wfree()
1082 */
1083 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1084 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1085 }
1086
1087 /* Control parameters for ECHO replies. */
1088 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1089 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1090
1091 /* Control parameter - ignore bogus broadcast responses? */
1092 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1093
1094 /*
1095 * Configurable global rate limit.
1096 *
1097 * ratelimit defines tokens/packet consumed for dst->rate_token
1098 * bucket ratemask defines which icmp types are ratelimited by
1099 * setting it's bit position.
1100 *
1101 * default:
1102 * dest unreachable (3), source quench (4),
1103 * time exceeded (11), parameter problem (12)
1104 */
1105
1106 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1107 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1108 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1109
1110 return 0;
1111
1112 fail:
1113 for_each_possible_cpu(i)
1114 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1115 kfree(net->ipv4.icmp_sk);
1116 return err;
1117 }
1118
1119 static struct pernet_operations __net_initdata icmp_sk_ops = {
1120 .init = icmp_sk_init,
1121 .exit = icmp_sk_exit,
1122 };
1123
icmp_init(void)1124 int __init icmp_init(void)
1125 {
1126 return register_pernet_subsys(&icmp_sk_ops);
1127 }
1128