?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6/verdex.git] / net / ipv4 / icmp.c
blob803bc9f173a7118865be23c5c6ebee36228a690c
1 /*
2 * NET3: Implementation of the ICMP protocol layer.
3 *
4 * Alan Cox, <alan@redhat.com>
5 *
6 * Version: $Id: icmp.c,v 1.85 2002/02/01 22:01:03 davem Exp $
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Some of the function names and the icmp unreach table for this
14 * module were derived from [icmp.c 1.0.11 06/02/93] by
15 * Ross Biro, Fred N. van Kempen, Mark Evans, Alan Cox, Gerhard Koerting.
16 * Other than that this module is a complete rewrite.
17 *
18 * Fixes:
19 * Clemens Fruhwirth : introduce global icmp rate limiting
20 * with icmp type masking ability instead
21 * of broken per type icmp timeouts.
22 * Mike Shaver : RFC1122 checks.
23 * Alan Cox : Multicast ping reply as self.
24 * Alan Cox : Fix atomicity lockup in ip_build_xmit
25 * call.
26 * Alan Cox : Added 216,128 byte paths to the MTU
27 * code.
28 * Martin Mares : RFC1812 checks.
29 * Martin Mares : Can be configured to follow redirects
30 * if acting as a router _without_ a
31 * routing protocol (RFC 1812).
32 * Martin Mares : Echo requests may be configured to
33 * be ignored (RFC 1812).
34 * Martin Mares : Limitation of ICMP error message
35 * transmit rate (RFC 1812).
36 * Martin Mares : TOS and Precedence set correctly
37 * (RFC 1812).
38 * Martin Mares : Now copying as much data from the
39 * original packet as we can without
40 * exceeding 576 bytes (RFC 1812).
41 * Willy Konynenberg : Transparent proxying support.
42 * Keith Owens : RFC1191 correction for 4.2BSD based
43 * path MTU bug.
44 * Thomas Quinot : ICMP Dest Unreach codes up to 15 are
45 * valid (RFC 1812).
46 * Andi Kleen : Check all packet lengths properly
47 * and moved all kfree_skb() up to
48 * icmp_rcv.
49 * Andi Kleen : Move the rate limit bookkeeping
50 * into the dest entry and use a token
51 * bucket filter (thanks to ANK). Make
52 * the rates sysctl configurable.
53 * Yu Tianli : Fixed two ugly bugs in icmp_send
54 * - IP option length was accounted wrongly
55 * - ICMP header length was not accounted
56 * at all.
57 * Tristan Greaves : Added sysctl option to ignore bogus
58 * broadcast responses from broken routers.
59 *
60 * To Fix:
61 *
62 * - Should use skb_pull() instead of all the manual checking.
63 * This would also greatly simply some upper layer error handlers. --AK
64 *
65 */
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 <net/snmp.h>
80 #include <net/ip.h>
81 #include <net/route.h>
82 #include <net/protocol.h>
83 #include <net/icmp.h>
84 #include <net/tcp.h>
85 #include <net/udp.h>
86 #include <net/raw.h>
87 #include <linux/skbuff.h>
88 #include <net/sock.h>
89 #include <linux/errno.h>
90 #include <linux/timer.h>
91 #include <linux/init.h>
92 #include <asm/system.h>
93 #include <asm/uaccess.h>
94 #include <net/checksum.h>
95 #include <net/xfrm.h>
96
97 /*
98 * Build xmit assembly blocks
99 */
100
101 struct icmp_bxm {
102 struct sk_buff *skb;
103 int offset;
104 int data_len;
105
106 struct {
107 struct icmphdr icmph;
108 __be32 times[3];
109 } data;
110 int head_len;
111 struct ip_options replyopts;
112 unsigned char optbuf[40];
113 };
114
115 /*
116 * Statistics
117 */
118 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
119 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
120
121 /* An array of errno for error messages from dest unreach. */
122 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
123
124 struct icmp_err icmp_err_convert[] = {
125 {
126 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
127 .fatal = 0,
128 },
129 {
130 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
131 .fatal = 0,
132 },
133 {
134 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
135 .fatal = 1,
136 },
137 {
138 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
139 .fatal = 1,
140 },
141 {
142 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
143 .fatal = 0,
144 },
145 {
146 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
147 .fatal = 0,
148 },
149 {
150 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
151 .fatal = 1,
152 },
153 {
154 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
155 .fatal = 1,
156 },
157 {
158 .errno = ENONET, /* ICMP_HOST_ISOLATED */
159 .fatal = 1,
160 },
161 {
162 .errno = ENETUNREACH, /* ICMP_NET_ANO */
163 .fatal = 1,
164 },
165 {
166 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
167 .fatal = 1,
168 },
169 {
170 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
171 .fatal = 0,
172 },
173 {
174 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
175 .fatal = 0,
176 },
177 {
178 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
179 .fatal = 1,
180 },
181 {
182 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
183 .fatal = 1,
184 },
185 {
186 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
187 .fatal = 1,
188 },
189 };
190
191 /*
192 * ICMP control array. This specifies what to do with each ICMP.
193 */
194
195 struct icmp_control {
196 void (*handler)(struct sk_buff *skb);
197 short error; /* This ICMP is classed as an error message */
198 };
199
200 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
201
202 /*
203 * The ICMP socket(s). This is the most convenient way to flow control
204 * our ICMP output as well as maintain a clean interface throughout
205 * all layers. All Socketless IP sends will soon be gone.
206 *
207 * On SMP we have one ICMP socket per-cpu.
208 */
209 static struct sock *icmp_sk(struct net *net)
210 {
211 return net->ipv4.icmp_sk[smp_processor_id()];
212 }
213
214 static inline int icmp_xmit_lock(struct sock *sk)
215 {
216 local_bh_disable();
217
218 if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
219 /* This can happen if the output path signals a
220 * dst_link_failure() for an outgoing ICMP packet.
221 */
222 local_bh_enable();
223 return 1;
224 }
225 return 0;
226 }
227
228 static inline void icmp_xmit_unlock(struct sock *sk)
229 {
230 spin_unlock_bh(&sk->sk_lock.slock);
231 }
232
233 /*
234 * Send an ICMP frame.
235 */
236
237 /*
238 * Check transmit rate limitation for given message.
239 * The rate information is held in the destination cache now.
240 * This function is generic and could be used for other purposes
241 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
242 *
243 * Note that the same dst_entry fields are modified by functions in
244 * route.c too, but these work for packet destinations while xrlim_allow
245 * works for icmp destinations. This means the rate limiting information
246 * for one "ip object" is shared - and these ICMPs are twice limited:
247 * by source and by destination.
248 *
249 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
250 * SHOULD allow setting of rate limits
251 *
252 * Shared between ICMPv4 and ICMPv6.
253 */
254 #define XRLIM_BURST_FACTOR 6
255 int xrlim_allow(struct dst_entry *dst, int timeout)
256 {
257 unsigned long now, token = dst->rate_tokens;
258 int rc = 0;
259
260 now = jiffies;
261 token += now - dst->rate_last;
262 dst->rate_last = now;
263 if (token > XRLIM_BURST_FACTOR * timeout)
264 token = XRLIM_BURST_FACTOR * timeout;
265 if (token >= timeout) {
266 token -= timeout;
267 rc = 1;
268 }
269 dst->rate_tokens = token;
270 return rc;
271 }
272
273 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
274 int type, int code)
275 {
276 struct dst_entry *dst = &rt->u.dst;
277 int rc = 1;
278
279 if (type > NR_ICMP_TYPES)
280 goto out;
281
282 /* Don't limit PMTU discovery. */
283 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
284 goto out;
285
286 /* No rate limit on loopback */
287 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
288 goto out;
289
290 /* Limit if icmp type is enabled in ratemask. */
291 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
292 rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
293 out:
294 return rc;
295 }
296
297 /*
298 * Maintain the counters used in the SNMP statistics for outgoing ICMP
299 */
300 void icmp_out_count(unsigned char type)
301 {
302 ICMPMSGOUT_INC_STATS(type);
303 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
304 }
305
306 /*
307 * Checksum each fragment, and on the first include the headers and final
308 * checksum.
309 */
310 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
311 struct sk_buff *skb)
312 {
313 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
314 __wsum csum;
315
316 csum = skb_copy_and_csum_bits(icmp_param->skb,
317 icmp_param->offset + offset,
318 to, len, 0);
319
320 skb->csum = csum_block_add(skb->csum, csum, odd);
321 if (icmp_pointers[icmp_param->data.icmph.type].error)
322 nf_ct_attach(skb, icmp_param->skb);
323 return 0;
324 }
325
326 static void icmp_push_reply(struct icmp_bxm *icmp_param,
327 struct ipcm_cookie *ipc, struct rtable *rt)
328 {
329 struct sock *sk;
330 struct sk_buff *skb;
331
332 sk = icmp_sk(dev_net(rt->u.dst.dev));
333 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
334 icmp_param->data_len+icmp_param->head_len,
335 icmp_param->head_len,
336 ipc, rt, MSG_DONTWAIT) < 0)
337 ip_flush_pending_frames(sk);
338 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
339 struct icmphdr *icmph = icmp_hdr(skb);
340 __wsum csum = 0;
341 struct sk_buff *skb1;
342
343 skb_queue_walk(&sk->sk_write_queue, skb1) {
344 csum = csum_add(csum, skb1->csum);
345 }
346 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
347 (char *)icmph,
348 icmp_param->head_len, csum);
349 icmph->checksum = csum_fold(csum);
350 skb->ip_summed = CHECKSUM_NONE;
351 ip_push_pending_frames(sk);
352 }
353 }
354
355 /*
356 * Driving logic for building and sending ICMP messages.
357 */
358
359 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
360 {
361 struct ipcm_cookie ipc;
362 struct rtable *rt = skb->rtable;
363 struct net *net = dev_net(rt->u.dst.dev);
364 struct sock *sk = icmp_sk(net);
365 struct inet_sock *inet = inet_sk(sk);
366 __be32 daddr;
367
368 if (ip_options_echo(&icmp_param->replyopts, skb))
369 return;
370
371 if (icmp_xmit_lock(sk))
372 return;
373
374 icmp_param->data.icmph.checksum = 0;
375
376 inet->tos = ip_hdr(skb)->tos;
377 daddr = ipc.addr = rt->rt_src;
378 ipc.opt = NULL;
379 if (icmp_param->replyopts.optlen) {
380 ipc.opt = &icmp_param->replyopts;
381 if (ipc.opt->srr)
382 daddr = icmp_param->replyopts.faddr;
383 }
384 {
385 struct flowi fl = { .nl_u = { .ip4_u =
386 { .daddr = daddr,
387 .saddr = rt->rt_spec_dst,
388 .tos = RT_TOS(ip_hdr(skb)->tos) } },
389 .proto = IPPROTO_ICMP };
390 security_skb_classify_flow(skb, &fl);
391 if (ip_route_output_key(net, &rt, &fl))
392 goto out_unlock;
393 }
394 if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
395 icmp_param->data.icmph.code))
396 icmp_push_reply(icmp_param, &ipc, rt);
397 ip_rt_put(rt);
398 out_unlock:
399 icmp_xmit_unlock(sk);
400 }
401
402
403 /*
404 * Send an ICMP message in response to a situation
405 *
406 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
407 * MAY send more (we do).
408 * MUST NOT change this header information.
409 * MUST NOT reply to a multicast/broadcast IP address.
410 * MUST NOT reply to a multicast/broadcast MAC address.
411 * MUST reply to only the first fragment.
412 */
413
414 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
415 {
416 struct iphdr *iph;
417 int room;
418 struct icmp_bxm icmp_param;
419 struct rtable *rt = skb_in->rtable;
420 struct ipcm_cookie ipc;
421 __be32 saddr;
422 u8 tos;
423 struct net *net;
424 struct sock *sk;
425
426 if (!rt)
427 goto out;
428 net = dev_net(rt->u.dst.dev);
429 sk = icmp_sk(net);
430
431 /*
432 * Find the original header. It is expected to be valid, of course.
433 * Check this, icmp_send is called from the most obscure devices
434 * sometimes.
435 */
436 iph = ip_hdr(skb_in);
437
438 if ((u8 *)iph < skb_in->head ||
439 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
440 goto out;
441
442 /*
443 * No replies to physical multicast/broadcast
444 */
445 if (skb_in->pkt_type != PACKET_HOST)
446 goto out;
447
448 /*
449 * Now check at the protocol level
450 */
451 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
452 goto out;
453
454 /*
455 * Only reply to fragment 0. We byte re-order the constant
456 * mask for efficiency.
457 */
458 if (iph->frag_off & htons(IP_OFFSET))
459 goto out;
460
461 /*
462 * If we send an ICMP error to an ICMP error a mess would result..
463 */
464 if (icmp_pointers[type].error) {
465 /*
466 * We are an error, check if we are replying to an
467 * ICMP error
468 */
469 if (iph->protocol == IPPROTO_ICMP) {
470 u8 _inner_type, *itp;
471
472 itp = skb_header_pointer(skb_in,
473 skb_network_header(skb_in) +
474 (iph->ihl << 2) +
475 offsetof(struct icmphdr,
476 type) -
477 skb_in->data,
478 sizeof(_inner_type),
479 &_inner_type);
480 if (itp == NULL)
481 goto out;
482
483 /*
484 * Assume any unknown ICMP type is an error. This
485 * isn't specified by the RFC, but think about it..
486 */
487 if (*itp > NR_ICMP_TYPES ||
488 icmp_pointers[*itp].error)
489 goto out;
490 }
491 }
492
493 if (icmp_xmit_lock(sk))
494 return;
495
496 /*
497 * Construct source address and options.
498 */
499
500 saddr = iph->daddr;
501 if (!(rt->rt_flags & RTCF_LOCAL)) {
502 struct net_device *dev = NULL;
503
504 if (rt->fl.iif &&
505 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
506 dev = dev_get_by_index(net, rt->fl.iif);
507
508 if (dev) {
509 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
510 dev_put(dev);
511 } else
512 saddr = 0;
513 }
514
515 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
516 IPTOS_PREC_INTERNETCONTROL) :
517 iph->tos;
518
519 if (ip_options_echo(&icmp_param.replyopts, skb_in))
520 goto out_unlock;
521
522
523 /*
524 * Prepare data for ICMP header.
525 */
526
527 icmp_param.data.icmph.type = type;
528 icmp_param.data.icmph.code = code;
529 icmp_param.data.icmph.un.gateway = info;
530 icmp_param.data.icmph.checksum = 0;
531 icmp_param.skb = skb_in;
532 icmp_param.offset = skb_network_offset(skb_in);
533 inet_sk(sk)->tos = tos;
534 ipc.addr = iph->saddr;
535 ipc.opt = &icmp_param.replyopts;
536
537 {
538 struct flowi fl = {
539 .nl_u = {
540 .ip4_u = {
541 .daddr = icmp_param.replyopts.srr ?
542 icmp_param.replyopts.faddr :
543 iph->saddr,
544 .saddr = saddr,
545 .tos = RT_TOS(tos)
546 }
547 },
548 .proto = IPPROTO_ICMP,
549 .uli_u = {
550 .icmpt = {
551 .type = type,
552 .code = code
553 }
554 }
555 };
556 int err;
557 struct rtable *rt2;
558
559 security_skb_classify_flow(skb_in, &fl);
560 if (__ip_route_output_key(net, &rt, &fl))
561 goto out_unlock;
562
563 /* No need to clone since we're just using its address. */
564 rt2 = rt;
565
566 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
567 switch (err) {
568 case 0:
569 if (rt != rt2)
570 goto route_done;
571 break;
572 case -EPERM:
573 rt = NULL;
574 break;
575 default:
576 goto out_unlock;
577 }
578
579 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
580 goto ende;
581
582 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
583 err = __ip_route_output_key(net, &rt2, &fl);
584 else {
585 struct flowi fl2 = {};
586 struct dst_entry *odst;
587
588 fl2.fl4_dst = fl.fl4_src;
589 if (ip_route_output_key(net, &rt2, &fl2))
590 goto ende;
591
592 /* Ugh! */
593 odst = skb_in->dst;
594 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
595 RT_TOS(tos), rt2->u.dst.dev);
596
597 dst_release(&rt2->u.dst);
598 rt2 = skb_in->rtable;
599 skb_in->dst = odst;
600 }
601
602 if (err)
603 goto ende;
604
605 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
606 XFRM_LOOKUP_ICMP);
607 if (err == -ENOENT) {
608 if (!rt)
609 goto out_unlock;
610 goto route_done;
611 }
612
613 dst_release(&rt->u.dst);
614 rt = rt2;
615
616 if (err)
617 goto out_unlock;
618 }
619
620 route_done:
621 if (!icmpv4_xrlim_allow(net, rt, 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->u.dst);
627 if (room > 576)
628 room = 576;
629 room -= sizeof(struct iphdr) + icmp_param.replyopts.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, &ipc, rt);
638 ende:
639 ip_rt_put(rt);
640 out_unlock:
641 icmp_xmit_unlock(sk);
642 out:;
643 }
644
645
646 /*
647 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
648 */
649
650 static void icmp_unreach(struct sk_buff *skb)
651 {
652 struct iphdr *iph;
653 struct icmphdr *icmph;
654 int hash, protocol;
655 struct net_protocol *ipprot;
656 u32 info = 0;
657 struct net *net;
658
659 net = dev_net(skb->dst->dev);
660
661 /*
662 * Incomplete header ?
663 * Only checks for the IP header, there should be an
664 * additional check for longer headers in upper levels.
665 */
666
667 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
668 goto out_err;
669
670 icmph = icmp_hdr(skb);
671 iph = (struct iphdr *)skb->data;
672
673 if (iph->ihl < 5) /* Mangled header, drop. */
674 goto out_err;
675
676 if (icmph->type == ICMP_DEST_UNREACH) {
677 switch (icmph->code & 15) {
678 case ICMP_NET_UNREACH:
679 case ICMP_HOST_UNREACH:
680 case ICMP_PROT_UNREACH:
681 case ICMP_PORT_UNREACH:
682 break;
683 case ICMP_FRAG_NEEDED:
684 if (ipv4_config.no_pmtu_disc) {
685 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
686 "fragmentation needed "
687 "and DF set.\n",
688 NIPQUAD(iph->daddr));
689 } else {
690 info = ip_rt_frag_needed(net, iph,
691 ntohs(icmph->un.frag.mtu));
692 if (!info)
693 goto out;
694 }
695 break;
696 case ICMP_SR_FAILED:
697 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
698 "Route Failed.\n",
699 NIPQUAD(iph->daddr));
700 break;
701 default:
702 break;
703 }
704 if (icmph->code > NR_ICMP_UNREACH)
705 goto out;
706 } else if (icmph->type == ICMP_PARAMETERPROB)
707 info = ntohl(icmph->un.gateway) >> 24;
708
709 /*
710 * Throw it at our lower layers
711 *
712 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
713 * header.
714 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
715 * transport layer.
716 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
717 * transport layer.
718 */
719
720 /*
721 * Check the other end isnt violating RFC 1122. Some routers send
722 * bogus responses to broadcast frames. If you see this message
723 * first check your netmask matches at both ends, if it does then
724 * get the other vendor to fix their kit.
725 */
726
727 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
728 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
729 if (net_ratelimit())
730 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
731 "type %u, code %u "
732 "error to a broadcast: %u.%u.%u.%u on %s\n",
733 NIPQUAD(ip_hdr(skb)->saddr),
734 icmph->type, icmph->code,
735 NIPQUAD(iph->daddr),
736 skb->dev->name);
737 goto out;
738 }
739
740 /* Checkin full IP header plus 8 bytes of protocol to
741 * avoid additional coding at protocol handlers.
742 */
743 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
744 goto out;
745
746 iph = (struct iphdr *)skb->data;
747 protocol = iph->protocol;
748
749 /*
750 * Deliver ICMP message to raw sockets. Pretty useless feature?
751 */
752 raw_icmp_error(skb, protocol, info);
753
754 hash = protocol & (MAX_INET_PROTOS - 1);
755 rcu_read_lock();
756 ipprot = rcu_dereference(inet_protos[hash]);
757 if (ipprot && ipprot->err_handler)
758 ipprot->err_handler(skb, info);
759 rcu_read_unlock();
760
761 out:
762 return;
763 out_err:
764 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
765 goto out;
766 }
767
768
769 /*
770 * Handle ICMP_REDIRECT.
771 */
772
773 static void icmp_redirect(struct sk_buff *skb)
774 {
775 struct iphdr *iph;
776
777 if (skb->len < sizeof(struct iphdr))
778 goto out_err;
779
780 /*
781 * Get the copied header of the packet that caused the redirect
782 */
783 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
784 goto out;
785
786 iph = (struct iphdr *)skb->data;
787
788 switch (icmp_hdr(skb)->code & 7) {
789 case ICMP_REDIR_NET:
790 case ICMP_REDIR_NETTOS:
791 /*
792 * As per RFC recommendations now handle it as a host redirect.
793 */
794 case ICMP_REDIR_HOST:
795 case ICMP_REDIR_HOSTTOS:
796 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
797 icmp_hdr(skb)->un.gateway,
798 iph->saddr, skb->dev);
799 break;
800 }
801 out:
802 return;
803 out_err:
804 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
805 goto out;
806 }
807
808 /*
809 * Handle ICMP_ECHO ("ping") requests.
810 *
811 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
812 * requests.
813 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
814 * included in the reply.
815 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
816 * echo requests, MUST have default=NOT.
817 * See also WRT handling of options once they are done and working.
818 */
819
820 static void icmp_echo(struct sk_buff *skb)
821 {
822 struct net *net;
823
824 net = dev_net(skb->dst->dev);
825 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
826 struct icmp_bxm icmp_param;
827
828 icmp_param.data.icmph = *icmp_hdr(skb);
829 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
830 icmp_param.skb = skb;
831 icmp_param.offset = 0;
832 icmp_param.data_len = skb->len;
833 icmp_param.head_len = sizeof(struct icmphdr);
834 icmp_reply(&icmp_param, skb);
835 }
836 }
837
838 /*
839 * Handle ICMP Timestamp requests.
840 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
841 * SHOULD be in the kernel for minimum random latency.
842 * MUST be accurate to a few minutes.
843 * MUST be updated at least at 15Hz.
844 */
845 static void icmp_timestamp(struct sk_buff *skb)
846 {
847 struct timeval tv;
848 struct icmp_bxm icmp_param;
849 /*
850 * Too short.
851 */
852 if (skb->len < 4)
853 goto out_err;
854
855 /*
856 * Fill in the current time as ms since midnight UT:
857 */
858 do_gettimeofday(&tv);
859 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
860 tv.tv_usec / 1000);
861 icmp_param.data.times[2] = icmp_param.data.times[1];
862 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
863 BUG();
864 icmp_param.data.icmph = *icmp_hdr(skb);
865 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
866 icmp_param.data.icmph.code = 0;
867 icmp_param.skb = skb;
868 icmp_param.offset = 0;
869 icmp_param.data_len = 0;
870 icmp_param.head_len = sizeof(struct icmphdr) + 12;
871 icmp_reply(&icmp_param, skb);
872 out:
873 return;
874 out_err:
875 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
876 goto out;
877 }
878
879
880 /*
881 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
882 *
883 * RFC1122 (3.2.2.9). A host MUST only send replies to
884 * ADDRESS_MASK requests if it's been configured as an address mask
885 * agent. Receiving a request doesn't constitute implicit permission to
886 * act as one. Of course, implementing this correctly requires (SHOULD)
887 * a way to turn the functionality on and off. Another one for sysctl(),
888 * I guess. -- MS
889 *
890 * RFC1812 (4.3.3.9). A router MUST implement it.
891 * A router SHOULD have switch turning it on/off.
892 * This switch MUST be ON by default.
893 *
894 * Gratuitous replies, zero-source replies are not implemented,
895 * that complies with RFC. DO NOT implement them!!! All the idea
896 * of broadcast addrmask replies as specified in RFC950 is broken.
897 * The problem is that it is not uncommon to have several prefixes
898 * on one physical interface. Moreover, addrmask agent can even be
899 * not aware of existing another prefixes.
900 * If source is zero, addrmask agent cannot choose correct prefix.
901 * Gratuitous mask announcements suffer from the same problem.
902 * RFC1812 explains it, but still allows to use ADDRMASK,
903 * that is pretty silly. --ANK
904 *
905 * All these rules are so bizarre, that I removed kernel addrmask
906 * support at all. It is wrong, it is obsolete, nobody uses it in
907 * any case. --ANK
908 *
909 * Furthermore you can do it with a usermode address agent program
910 * anyway...
911 */
912
913 static void icmp_address(struct sk_buff *skb)
914 {
915 #if 0
916 if (net_ratelimit())
917 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
918 #endif
919 }
920
921 /*
922 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
923 * loudly if an inconsistency is found.
924 */
925
926 static void icmp_address_reply(struct sk_buff *skb)
927 {
928 struct rtable *rt = skb->rtable;
929 struct net_device *dev = skb->dev;
930 struct in_device *in_dev;
931 struct in_ifaddr *ifa;
932
933 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
934 goto out;
935
936 in_dev = in_dev_get(dev);
937 if (!in_dev)
938 goto out;
939 rcu_read_lock();
940 if (in_dev->ifa_list &&
941 IN_DEV_LOG_MARTIANS(in_dev) &&
942 IN_DEV_FORWARD(in_dev)) {
943 __be32 _mask, *mp;
944
945 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
946 BUG_ON(mp == NULL);
947 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
948 if (*mp == ifa->ifa_mask &&
949 inet_ifa_match(rt->rt_src, ifa))
950 break;
951 }
952 if (!ifa && net_ratelimit()) {
953 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
954 "%s/%u.%u.%u.%u\n",
955 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
956 }
957 }
958 rcu_read_unlock();
959 in_dev_put(in_dev);
960 out:;
961 }
962
963 static void icmp_discard(struct sk_buff *skb)
964 {
965 }
966
967 /*
968 * Deal with incoming ICMP packets.
969 */
970 int icmp_rcv(struct sk_buff *skb)
971 {
972 struct icmphdr *icmph;
973 struct rtable *rt = skb->rtable;
974
975 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
976 int nh;
977
978 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
979 XFRM_STATE_ICMP))
980 goto drop;
981
982 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
983 goto drop;
984
985 nh = skb_network_offset(skb);
986 skb_set_network_header(skb, sizeof(*icmph));
987
988 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
989 goto drop;
990
991 skb_set_network_header(skb, nh);
992 }
993
994 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
995
996 switch (skb->ip_summed) {
997 case CHECKSUM_COMPLETE:
998 if (!csum_fold(skb->csum))
999 break;
1000 /* fall through */
1001 case CHECKSUM_NONE:
1002 skb->csum = 0;
1003 if (__skb_checksum_complete(skb))
1004 goto error;
1005 }
1006
1007 if (!pskb_pull(skb, sizeof(*icmph)))
1008 goto error;
1009
1010 icmph = icmp_hdr(skb);
1011
1012 ICMPMSGIN_INC_STATS_BH(icmph->type);
1013 /*
1014 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1015 *
1016 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1017 * discarded.
1018 */
1019 if (icmph->type > NR_ICMP_TYPES)
1020 goto error;
1021
1022
1023 /*
1024 * Parse the ICMP message
1025 */
1026
1027 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1028 struct net *net;
1029
1030 net = dev_net(rt->u.dst.dev);
1031 /*
1032 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1033 * silently ignored (we let user decide with a sysctl).
1034 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1035 * discarded if to broadcast/multicast.
1036 */
1037 if ((icmph->type == ICMP_ECHO ||
1038 icmph->type == ICMP_TIMESTAMP) &&
1039 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1040 goto error;
1041 }
1042 if (icmph->type != ICMP_ECHO &&
1043 icmph->type != ICMP_TIMESTAMP &&
1044 icmph->type != ICMP_ADDRESS &&
1045 icmph->type != ICMP_ADDRESSREPLY) {
1046 goto error;
1047 }
1048 }
1049
1050 icmp_pointers[icmph->type].handler(skb);
1051
1052 drop:
1053 kfree_skb(skb);
1054 return 0;
1055 error:
1056 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1057 goto drop;
1058 }
1059
1060 /*
1061 * This table is the definition of how we handle ICMP.
1062 */
1063 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1064 [ICMP_ECHOREPLY] = {
1065 .handler = icmp_discard,
1066 },
1067 [1] = {
1068 .handler = icmp_discard,
1069 .error = 1,
1070 },
1071 [2] = {
1072 .handler = icmp_discard,
1073 .error = 1,
1074 },
1075 [ICMP_DEST_UNREACH] = {
1076 .handler = icmp_unreach,
1077 .error = 1,
1078 },
1079 [ICMP_SOURCE_QUENCH] = {
1080 .handler = icmp_unreach,
1081 .error = 1,
1082 },
1083 [ICMP_REDIRECT] = {
1084 .handler = icmp_redirect,
1085 .error = 1,
1086 },
1087 [6] = {
1088 .handler = icmp_discard,
1089 .error = 1,
1090 },
1091 [7] = {
1092 .handler = icmp_discard,
1093 .error = 1,
1094 },
1095 [ICMP_ECHO] = {
1096 .handler = icmp_echo,
1097 },
1098 [9] = {
1099 .handler = icmp_discard,
1100 .error = 1,
1101 },
1102 [10] = {
1103 .handler = icmp_discard,
1104 .error = 1,
1105 },
1106 [ICMP_TIME_EXCEEDED] = {
1107 .handler = icmp_unreach,
1108 .error = 1,
1109 },
1110 [ICMP_PARAMETERPROB] = {
1111 .handler = icmp_unreach,
1112 .error = 1,
1113 },
1114 [ICMP_TIMESTAMP] = {
1115 .handler = icmp_timestamp,
1116 },
1117 [ICMP_TIMESTAMPREPLY] = {
1118 .handler = icmp_discard,
1119 },
1120 [ICMP_INFO_REQUEST] = {
1121 .handler = icmp_discard,
1122 },
1123 [ICMP_INFO_REPLY] = {
1124 .handler = icmp_discard,
1125 },
1126 [ICMP_ADDRESS] = {
1127 .handler = icmp_address,
1128 },
1129 [ICMP_ADDRESSREPLY] = {
1130 .handler = icmp_address_reply,
1131 },
1132 };
1133
1134 static void __net_exit icmp_sk_exit(struct net *net)
1135 {
1136 int i;
1137
1138 for_each_possible_cpu(i)
1139 sk_release_kernel(net->ipv4.icmp_sk[i]);
1140 kfree(net->ipv4.icmp_sk);
1141 net->ipv4.icmp_sk = NULL;
1142 }
1143
1144 int __net_init icmp_sk_init(struct net *net)
1145 {
1146 int i, err;
1147
1148 net->ipv4.icmp_sk =
1149 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1150 if (net->ipv4.icmp_sk == NULL)
1151 return -ENOMEM;
1152
1153 for_each_possible_cpu(i) {
1154 struct sock *sk;
1155 struct socket *sock;
1156 struct inet_sock *inet;
1157
1158 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP, &sock);
1159 if (err < 0)
1160 goto fail;
1161
1162 net->ipv4.icmp_sk[i] = sk = sock->sk;
1163 sk_change_net(sk, net);
1164
1165 sk->sk_allocation = GFP_ATOMIC;
1166
1167 /* Enough space for 2 64K ICMP packets, including
1168 * sk_buff struct overhead.
1169 */
1170 sk->sk_sndbuf =
1171 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1172
1173 inet = inet_sk(sk);
1174 inet->uc_ttl = -1;
1175 inet->pmtudisc = IP_PMTUDISC_DONT;
1176
1177 /* Unhash it so that IP input processing does not even
1178 * see it, we do not wish this socket to see incoming
1179 * packets.
1180 */
1181 sk->sk_prot->unhash(sk);
1182 }
1183
1184 /* Control parameters for ECHO replies. */
1185 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1186 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1187
1188 /* Control parameter - ignore bogus broadcast responses? */
1189 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1190
1191 /*
1192 * Configurable global rate limit.
1193 *
1194 * ratelimit defines tokens/packet consumed for dst->rate_token
1195 * bucket ratemask defines which icmp types are ratelimited by
1196 * setting it's bit position.
1197 *
1198 * default:
1199 * dest unreachable (3), source quench (4),
1200 * time exceeded (11), parameter problem (12)
1201 */
1202
1203 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1204 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1205 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1206
1207 return 0;
1208
1209 fail:
1210 for_each_possible_cpu(i)
1211 sk_release_kernel(net->ipv4.icmp_sk[i]);
1212 kfree(net->ipv4.icmp_sk);
1213 return err;
1214 }
1215
1216 static struct pernet_operations __net_initdata icmp_sk_ops = {
1217 .init = icmp_sk_init,
1218 .exit = icmp_sk_exit,
1219 };
1220
1221 int __init icmp_init(void)
1222 {
1223 return register_pernet_device(&icmp_sk_ops);
1224 }
1225
1226 EXPORT_SYMBOL(icmp_err_convert);
1227 EXPORT_SYMBOL(icmp_send);
1228 EXPORT_SYMBOL(icmp_statistics);
1229 EXPORT_SYMBOL(xrlim_allow);
Verdex Pro support