?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
drm/i915: fix off by one in VGA save/restore of AR & CR regs.
[linux-2.6/linux-2.6-openrd.git] / net / ipv4 / icmp.c
blob87397351ddace424de5f0341482b408cdfaaf363
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 #include <net/inet_common.h>
97
98 /*
99 * Build xmit assembly blocks
100 */
101
102 struct icmp_bxm {
103 struct sk_buff *skb;
104 int offset;
105 int data_len;
106
107 struct {
108 struct icmphdr icmph;
109 __be32 times[3];
110 } data;
111 int head_len;
112 struct ip_options replyopts;
113 unsigned char optbuf[40];
114 };
115
116 /*
117 * Statistics
118 */
119 DEFINE_SNMP_STAT(struct icmp_mib, icmp_statistics) __read_mostly;
120 DEFINE_SNMP_STAT(struct icmpmsg_mib, icmpmsg_statistics) __read_mostly;
121
122 /* An array of errno for error messages from dest unreach. */
123 /* RFC 1122: 3.2.2.1 States that NET_UNREACH, HOST_UNREACH and SR_FAILED MUST be considered 'transient errs'. */
124
125 struct icmp_err icmp_err_convert[] = {
126 {
127 .errno = ENETUNREACH, /* ICMP_NET_UNREACH */
128 .fatal = 0,
129 },
130 {
131 .errno = EHOSTUNREACH, /* ICMP_HOST_UNREACH */
132 .fatal = 0,
133 },
134 {
135 .errno = ENOPROTOOPT /* ICMP_PROT_UNREACH */,
136 .fatal = 1,
137 },
138 {
139 .errno = ECONNREFUSED, /* ICMP_PORT_UNREACH */
140 .fatal = 1,
141 },
142 {
143 .errno = EMSGSIZE, /* ICMP_FRAG_NEEDED */
144 .fatal = 0,
145 },
146 {
147 .errno = EOPNOTSUPP, /* ICMP_SR_FAILED */
148 .fatal = 0,
149 },
150 {
151 .errno = ENETUNREACH, /* ICMP_NET_UNKNOWN */
152 .fatal = 1,
153 },
154 {
155 .errno = EHOSTDOWN, /* ICMP_HOST_UNKNOWN */
156 .fatal = 1,
157 },
158 {
159 .errno = ENONET, /* ICMP_HOST_ISOLATED */
160 .fatal = 1,
161 },
162 {
163 .errno = ENETUNREACH, /* ICMP_NET_ANO */
164 .fatal = 1,
165 },
166 {
167 .errno = EHOSTUNREACH, /* ICMP_HOST_ANO */
168 .fatal = 1,
169 },
170 {
171 .errno = ENETUNREACH, /* ICMP_NET_UNR_TOS */
172 .fatal = 0,
173 },
174 {
175 .errno = EHOSTUNREACH, /* ICMP_HOST_UNR_TOS */
176 .fatal = 0,
177 },
178 {
179 .errno = EHOSTUNREACH, /* ICMP_PKT_FILTERED */
180 .fatal = 1,
181 },
182 {
183 .errno = EHOSTUNREACH, /* ICMP_PREC_VIOLATION */
184 .fatal = 1,
185 },
186 {
187 .errno = EHOSTUNREACH, /* ICMP_PREC_CUTOFF */
188 .fatal = 1,
189 },
190 };
191
192 /*
193 * ICMP control array. This specifies what to do with each ICMP.
194 */
195
196 struct icmp_control {
197 void (*handler)(struct sk_buff *skb);
198 short error; /* This ICMP is classed as an error message */
199 };
200
201 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
202
203 /*
204 * The ICMP socket(s). This is the most convenient way to flow control
205 * our ICMP output as well as maintain a clean interface throughout
206 * all layers. All Socketless IP sends will soon be gone.
207 *
208 * On SMP we have one ICMP socket per-cpu.
209 */
210 static struct sock *icmp_sk(struct net *net)
211 {
212 return net->ipv4.icmp_sk[smp_processor_id()];
213 }
214
215 static inline int icmp_xmit_lock(struct sock *sk)
216 {
217 local_bh_disable();
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 1;
225 }
226 return 0;
227 }
228
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
238 /*
239 * Check transmit rate limitation for given message.
240 * The rate information is held in the destination cache now.
241 * This function is generic and could be used for other purposes
242 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
243 *
244 * Note that the same dst_entry fields are modified by functions in
245 * route.c too, but these work for packet destinations while xrlim_allow
246 * works for icmp destinations. This means the rate limiting information
247 * for one "ip object" is shared - and these ICMPs are twice limited:
248 * by source and by destination.
249 *
250 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
251 * SHOULD allow setting of rate limits
252 *
253 * Shared between ICMPv4 and ICMPv6.
254 */
255 #define XRLIM_BURST_FACTOR 6
256 int xrlim_allow(struct dst_entry *dst, int timeout)
257 {
258 unsigned long now, token = dst->rate_tokens;
259 int rc = 0;
260
261 now = jiffies;
262 token += now - dst->rate_last;
263 dst->rate_last = now;
264 if (token > XRLIM_BURST_FACTOR * timeout)
265 token = XRLIM_BURST_FACTOR * timeout;
266 if (token >= timeout) {
267 token -= timeout;
268 rc = 1;
269 }
270 dst->rate_tokens = token;
271 return rc;
272 }
273
274 static inline int icmpv4_xrlim_allow(struct net *net, struct rtable *rt,
275 int type, int code)
276 {
277 struct dst_entry *dst = &rt->u.dst;
278 int rc = 1;
279
280 if (type > NR_ICMP_TYPES)
281 goto out;
282
283 /* Don't limit PMTU discovery. */
284 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
285 goto out;
286
287 /* No rate limit on loopback */
288 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
289 goto out;
290
291 /* Limit if icmp type is enabled in ratemask. */
292 if ((1 << type) & net->ipv4.sysctl_icmp_ratemask)
293 rc = xrlim_allow(dst, net->ipv4.sysctl_icmp_ratelimit);
294 out:
295 return rc;
296 }
297
298 /*
299 * Maintain the counters used in the SNMP statistics for outgoing ICMP
300 */
301 void icmp_out_count(unsigned char type)
302 {
303 ICMPMSGOUT_INC_STATS(type);
304 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
305 }
306
307 /*
308 * Checksum each fragment, and on the first include the headers and final
309 * checksum.
310 */
311 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
312 struct sk_buff *skb)
313 {
314 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
315 __wsum csum;
316
317 csum = skb_copy_and_csum_bits(icmp_param->skb,
318 icmp_param->offset + offset,
319 to, len, 0);
320
321 skb->csum = csum_block_add(skb->csum, csum, odd);
322 if (icmp_pointers[icmp_param->data.icmph.type].error)
323 nf_ct_attach(skb, icmp_param->skb);
324 return 0;
325 }
326
327 static void icmp_push_reply(struct icmp_bxm *icmp_param,
328 struct ipcm_cookie *ipc, struct rtable *rt)
329 {
330 struct sock *sk;
331 struct sk_buff *skb;
332
333 sk = icmp_sk(dev_net(rt->u.dst.dev));
334 if (ip_append_data(sk, icmp_glue_bits, icmp_param,
335 icmp_param->data_len+icmp_param->head_len,
336 icmp_param->head_len,
337 ipc, rt, MSG_DONTWAIT) < 0)
338 ip_flush_pending_frames(sk);
339 else if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
340 struct icmphdr *icmph = icmp_hdr(skb);
341 __wsum csum = 0;
342 struct sk_buff *skb1;
343
344 skb_queue_walk(&sk->sk_write_queue, skb1) {
345 csum = csum_add(csum, skb1->csum);
346 }
347 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
348 (char *)icmph,
349 icmp_param->head_len, csum);
350 icmph->checksum = csum_fold(csum);
351 skb->ip_summed = CHECKSUM_NONE;
352 ip_push_pending_frames(sk);
353 }
354 }
355
356 /*
357 * Driving logic for building and sending ICMP messages.
358 */
359
360 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
361 {
362 struct ipcm_cookie ipc;
363 struct rtable *rt = skb->rtable;
364 struct net *net = dev_net(rt->u.dst.dev);
365 struct sock *sk = icmp_sk(net);
366 struct inet_sock *inet = inet_sk(sk);
367 __be32 daddr;
368
369 if (ip_options_echo(&icmp_param->replyopts, skb))
370 return;
371
372 if (icmp_xmit_lock(sk))
373 return;
374
375 icmp_param->data.icmph.checksum = 0;
376
377 inet->tos = ip_hdr(skb)->tos;
378 daddr = ipc.addr = rt->rt_src;
379 ipc.opt = NULL;
380 if (icmp_param->replyopts.optlen) {
381 ipc.opt = &icmp_param->replyopts;
382 if (ipc.opt->srr)
383 daddr = icmp_param->replyopts.faddr;
384 }
385 {
386 struct flowi fl = { .nl_u = { .ip4_u =
387 { .daddr = daddr,
388 .saddr = rt->rt_spec_dst,
389 .tos = RT_TOS(ip_hdr(skb)->tos) } },
390 .proto = IPPROTO_ICMP };
391 security_skb_classify_flow(skb, &fl);
392 if (ip_route_output_key(net, &rt, &fl))
393 goto out_unlock;
394 }
395 if (icmpv4_xrlim_allow(net, rt, icmp_param->data.icmph.type,
396 icmp_param->data.icmph.code))
397 icmp_push_reply(icmp_param, &ipc, rt);
398 ip_rt_put(rt);
399 out_unlock:
400 icmp_xmit_unlock(sk);
401 }
402
403
404 /*
405 * Send an ICMP message in response to a situation
406 *
407 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
408 * MAY send more (we do).
409 * MUST NOT change this header information.
410 * MUST NOT reply to a multicast/broadcast IP address.
411 * MUST NOT reply to a multicast/broadcast MAC address.
412 * MUST reply to only the first fragment.
413 */
414
415 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
416 {
417 struct iphdr *iph;
418 int room;
419 struct icmp_bxm icmp_param;
420 struct rtable *rt = skb_in->rtable;
421 struct ipcm_cookie ipc;
422 __be32 saddr;
423 u8 tos;
424 struct net *net;
425 struct sock *sk;
426
427 if (!rt)
428 goto out;
429 net = dev_net(rt->u.dst.dev);
430 sk = icmp_sk(net);
431
432 /*
433 * Find the original header. It is expected to be valid, of course.
434 * Check this, icmp_send is called from the most obscure devices
435 * sometimes.
436 */
437 iph = ip_hdr(skb_in);
438
439 if ((u8 *)iph < skb_in->head ||
440 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
441 goto out;
442
443 /*
444 * No replies to physical multicast/broadcast
445 */
446 if (skb_in->pkt_type != PACKET_HOST)
447 goto out;
448
449 /*
450 * Now check at the protocol level
451 */
452 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
453 goto out;
454
455 /*
456 * Only reply to fragment 0. We byte re-order the constant
457 * mask for efficiency.
458 */
459 if (iph->frag_off & htons(IP_OFFSET))
460 goto out;
461
462 /*
463 * If we send an ICMP error to an ICMP error a mess would result..
464 */
465 if (icmp_pointers[type].error) {
466 /*
467 * We are an error, check if we are replying to an
468 * ICMP error
469 */
470 if (iph->protocol == IPPROTO_ICMP) {
471 u8 _inner_type, *itp;
472
473 itp = skb_header_pointer(skb_in,
474 skb_network_header(skb_in) +
475 (iph->ihl << 2) +
476 offsetof(struct icmphdr,
477 type) -
478 skb_in->data,
479 sizeof(_inner_type),
480 &_inner_type);
481 if (itp == NULL)
482 goto out;
483
484 /*
485 * Assume any unknown ICMP type is an error. This
486 * isn't specified by the RFC, but think about it..
487 */
488 if (*itp > NR_ICMP_TYPES ||
489 icmp_pointers[*itp].error)
490 goto out;
491 }
492 }
493
494 if (icmp_xmit_lock(sk))
495 return;
496
497 /*
498 * Construct source address and options.
499 */
500
501 saddr = iph->daddr;
502 if (!(rt->rt_flags & RTCF_LOCAL)) {
503 struct net_device *dev = NULL;
504
505 if (rt->fl.iif &&
506 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr)
507 dev = dev_get_by_index(net, rt->fl.iif);
508
509 if (dev) {
510 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
511 dev_put(dev);
512 } else
513 saddr = 0;
514 }
515
516 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
517 IPTOS_PREC_INTERNETCONTROL) :
518 iph->tos;
519
520 if (ip_options_echo(&icmp_param.replyopts, skb_in))
521 goto out_unlock;
522
523
524 /*
525 * Prepare data for ICMP header.
526 */
527
528 icmp_param.data.icmph.type = type;
529 icmp_param.data.icmph.code = code;
530 icmp_param.data.icmph.un.gateway = info;
531 icmp_param.data.icmph.checksum = 0;
532 icmp_param.skb = skb_in;
533 icmp_param.offset = skb_network_offset(skb_in);
534 inet_sk(sk)->tos = tos;
535 ipc.addr = iph->saddr;
536 ipc.opt = &icmp_param.replyopts;
537
538 {
539 struct flowi fl = {
540 .nl_u = {
541 .ip4_u = {
542 .daddr = icmp_param.replyopts.srr ?
543 icmp_param.replyopts.faddr :
544 iph->saddr,
545 .saddr = saddr,
546 .tos = RT_TOS(tos)
547 }
548 },
549 .proto = IPPROTO_ICMP,
550 .uli_u = {
551 .icmpt = {
552 .type = type,
553 .code = code
554 }
555 }
556 };
557 int err;
558 struct rtable *rt2;
559
560 security_skb_classify_flow(skb_in, &fl);
561 if (__ip_route_output_key(net, &rt, &fl))
562 goto out_unlock;
563
564 /* No need to clone since we're just using its address. */
565 rt2 = rt;
566
567 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
568 switch (err) {
569 case 0:
570 if (rt != rt2)
571 goto route_done;
572 break;
573 case -EPERM:
574 rt = NULL;
575 break;
576 default:
577 goto out_unlock;
578 }
579
580 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
581 goto relookup_failed;
582
583 if (inet_addr_type(net, fl.fl4_src) == RTN_LOCAL)
584 err = __ip_route_output_key(net, &rt2, &fl);
585 else {
586 struct flowi fl2 = {};
587 struct dst_entry *odst;
588
589 fl2.fl4_dst = fl.fl4_src;
590 if (ip_route_output_key(net, &rt2, &fl2))
591 goto relookup_failed;
592
593 /* Ugh! */
594 odst = skb_in->dst;
595 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
596 RT_TOS(tos), rt2->u.dst.dev);
597
598 dst_release(&rt2->u.dst);
599 rt2 = skb_in->rtable;
600 skb_in->dst = odst;
601 }
602
603 if (err)
604 goto relookup_failed;
605
606 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
607 XFRM_LOOKUP_ICMP);
608 switch (err) {
609 case 0:
610 dst_release(&rt->u.dst);
611 rt = rt2;
612 break;
613 case -EPERM:
614 goto ende;
615 default:
616 relookup_failed:
617 if (!rt)
618 goto out_unlock;
619 break;
620 }
621 }
622
623 route_done:
624 if (!icmpv4_xrlim_allow(net, rt, type, code))
625 goto ende;
626
627 /* RFC says return as much as we can without exceeding 576 bytes. */
628
629 room = dst_mtu(&rt->u.dst);
630 if (room > 576)
631 room = 576;
632 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
633 room -= sizeof(struct icmphdr);
634
635 icmp_param.data_len = skb_in->len - icmp_param.offset;
636 if (icmp_param.data_len > room)
637 icmp_param.data_len = room;
638 icmp_param.head_len = sizeof(struct icmphdr);
639
640 icmp_push_reply(&icmp_param, &ipc, rt);
641 ende:
642 ip_rt_put(rt);
643 out_unlock:
644 icmp_xmit_unlock(sk);
645 out:;
646 }
647
648
649 /*
650 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
651 */
652
653 static void icmp_unreach(struct sk_buff *skb)
654 {
655 struct iphdr *iph;
656 struct icmphdr *icmph;
657 int hash, protocol;
658 struct net_protocol *ipprot;
659 u32 info = 0;
660 struct net *net;
661
662 net = dev_net(skb->dst->dev);
663
664 /*
665 * Incomplete header ?
666 * Only checks for the IP header, there should be an
667 * additional check for longer headers in upper levels.
668 */
669
670 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
671 goto out_err;
672
673 icmph = icmp_hdr(skb);
674 iph = (struct iphdr *)skb->data;
675
676 if (iph->ihl < 5) /* Mangled header, drop. */
677 goto out_err;
678
679 if (icmph->type == ICMP_DEST_UNREACH) {
680 switch (icmph->code & 15) {
681 case ICMP_NET_UNREACH:
682 case ICMP_HOST_UNREACH:
683 case ICMP_PROT_UNREACH:
684 case ICMP_PORT_UNREACH:
685 break;
686 case ICMP_FRAG_NEEDED:
687 if (ipv4_config.no_pmtu_disc) {
688 LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": "
689 "fragmentation needed "
690 "and DF set.\n",
691 NIPQUAD(iph->daddr));
692 } else {
693 info = ip_rt_frag_needed(net, iph,
694 ntohs(icmph->un.frag.mtu),
695 skb->dev);
696 if (!info)
697 goto out;
698 }
699 break;
700 case ICMP_SR_FAILED:
701 LIMIT_NETDEBUG(KERN_INFO "ICMP: " NIPQUAD_FMT ": Source "
702 "Route Failed.\n",
703 NIPQUAD(iph->daddr));
704 break;
705 default:
706 break;
707 }
708 if (icmph->code > NR_ICMP_UNREACH)
709 goto out;
710 } else if (icmph->type == ICMP_PARAMETERPROB)
711 info = ntohl(icmph->un.gateway) >> 24;
712
713 /*
714 * Throw it at our lower layers
715 *
716 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
717 * header.
718 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
719 * transport layer.
720 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
721 * transport layer.
722 */
723
724 /*
725 * Check the other end isnt violating RFC 1122. Some routers send
726 * bogus responses to broadcast frames. If you see this message
727 * first check your netmask matches at both ends, if it does then
728 * get the other vendor to fix their kit.
729 */
730
731 if (!net->ipv4.sysctl_icmp_ignore_bogus_error_responses &&
732 inet_addr_type(net, iph->daddr) == RTN_BROADCAST) {
733 if (net_ratelimit())
734 printk(KERN_WARNING NIPQUAD_FMT " sent an invalid ICMP "
735 "type %u, code %u "
736 "error to a broadcast: " NIPQUAD_FMT " on %s\n",
737 NIPQUAD(ip_hdr(skb)->saddr),
738 icmph->type, icmph->code,
739 NIPQUAD(iph->daddr),
740 skb->dev->name);
741 goto out;
742 }
743
744 /* Checkin full IP header plus 8 bytes of protocol to
745 * avoid additional coding at protocol handlers.
746 */
747 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
748 goto out;
749
750 iph = (struct iphdr *)skb->data;
751 protocol = iph->protocol;
752
753 /*
754 * Deliver ICMP message to raw sockets. Pretty useless feature?
755 */
756 raw_icmp_error(skb, protocol, info);
757
758 hash = protocol & (MAX_INET_PROTOS - 1);
759 rcu_read_lock();
760 ipprot = rcu_dereference(inet_protos[hash]);
761 if (ipprot && ipprot->err_handler)
762 ipprot->err_handler(skb, info);
763 rcu_read_unlock();
764
765 out:
766 return;
767 out_err:
768 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
769 goto out;
770 }
771
772
773 /*
774 * Handle ICMP_REDIRECT.
775 */
776
777 static void icmp_redirect(struct sk_buff *skb)
778 {
779 struct iphdr *iph;
780
781 if (skb->len < sizeof(struct iphdr))
782 goto out_err;
783
784 /*
785 * Get the copied header of the packet that caused the redirect
786 */
787 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
788 goto out;
789
790 iph = (struct iphdr *)skb->data;
791
792 switch (icmp_hdr(skb)->code & 7) {
793 case ICMP_REDIR_NET:
794 case ICMP_REDIR_NETTOS:
795 /*
796 * As per RFC recommendations now handle it as a host redirect.
797 */
798 case ICMP_REDIR_HOST:
799 case ICMP_REDIR_HOSTTOS:
800 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
801 icmp_hdr(skb)->un.gateway,
802 iph->saddr, skb->dev);
803 break;
804 }
805 out:
806 return;
807 out_err:
808 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
809 goto out;
810 }
811
812 /*
813 * Handle ICMP_ECHO ("ping") requests.
814 *
815 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
816 * requests.
817 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
818 * included in the reply.
819 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
820 * echo requests, MUST have default=NOT.
821 * See also WRT handling of options once they are done and working.
822 */
823
824 static void icmp_echo(struct sk_buff *skb)
825 {
826 struct net *net;
827
828 net = dev_net(skb->dst->dev);
829 if (!net->ipv4.sysctl_icmp_echo_ignore_all) {
830 struct icmp_bxm icmp_param;
831
832 icmp_param.data.icmph = *icmp_hdr(skb);
833 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
834 icmp_param.skb = skb;
835 icmp_param.offset = 0;
836 icmp_param.data_len = skb->len;
837 icmp_param.head_len = sizeof(struct icmphdr);
838 icmp_reply(&icmp_param, skb);
839 }
840 }
841
842 /*
843 * Handle ICMP Timestamp requests.
844 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
845 * SHOULD be in the kernel for minimum random latency.
846 * MUST be accurate to a few minutes.
847 * MUST be updated at least at 15Hz.
848 */
849 static void icmp_timestamp(struct sk_buff *skb)
850 {
851 struct timespec tv;
852 struct icmp_bxm icmp_param;
853 /*
854 * Too short.
855 */
856 if (skb->len < 4)
857 goto out_err;
858
859 /*
860 * Fill in the current time as ms since midnight UT:
861 */
862 getnstimeofday(&tv);
863 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * MSEC_PER_SEC +
864 tv.tv_nsec / NSEC_PER_MSEC);
865 icmp_param.data.times[2] = icmp_param.data.times[1];
866 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
867 BUG();
868 icmp_param.data.icmph = *icmp_hdr(skb);
869 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
870 icmp_param.data.icmph.code = 0;
871 icmp_param.skb = skb;
872 icmp_param.offset = 0;
873 icmp_param.data_len = 0;
874 icmp_param.head_len = sizeof(struct icmphdr) + 12;
875 icmp_reply(&icmp_param, skb);
876 out:
877 return;
878 out_err:
879 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
880 goto out;
881 }
882
883
884 /*
885 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
886 *
887 * RFC1122 (3.2.2.9). A host MUST only send replies to
888 * ADDRESS_MASK requests if it's been configured as an address mask
889 * agent. Receiving a request doesn't constitute implicit permission to
890 * act as one. Of course, implementing this correctly requires (SHOULD)
891 * a way to turn the functionality on and off. Another one for sysctl(),
892 * I guess. -- MS
893 *
894 * RFC1812 (4.3.3.9). A router MUST implement it.
895 * A router SHOULD have switch turning it on/off.
896 * This switch MUST be ON by default.
897 *
898 * Gratuitous replies, zero-source replies are not implemented,
899 * that complies with RFC. DO NOT implement them!!! All the idea
900 * of broadcast addrmask replies as specified in RFC950 is broken.
901 * The problem is that it is not uncommon to have several prefixes
902 * on one physical interface. Moreover, addrmask agent can even be
903 * not aware of existing another prefixes.
904 * If source is zero, addrmask agent cannot choose correct prefix.
905 * Gratuitous mask announcements suffer from the same problem.
906 * RFC1812 explains it, but still allows to use ADDRMASK,
907 * that is pretty silly. --ANK
908 *
909 * All these rules are so bizarre, that I removed kernel addrmask
910 * support at all. It is wrong, it is obsolete, nobody uses it in
911 * any case. --ANK
912 *
913 * Furthermore you can do it with a usermode address agent program
914 * anyway...
915 */
916
917 static void icmp_address(struct sk_buff *skb)
918 {
919 #if 0
920 if (net_ratelimit())
921 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
922 #endif
923 }
924
925 /*
926 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
927 * loudly if an inconsistency is found.
928 */
929
930 static void icmp_address_reply(struct sk_buff *skb)
931 {
932 struct rtable *rt = skb->rtable;
933 struct net_device *dev = skb->dev;
934 struct in_device *in_dev;
935 struct in_ifaddr *ifa;
936
937 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
938 goto out;
939
940 in_dev = in_dev_get(dev);
941 if (!in_dev)
942 goto out;
943 rcu_read_lock();
944 if (in_dev->ifa_list &&
945 IN_DEV_LOG_MARTIANS(in_dev) &&
946 IN_DEV_FORWARD(in_dev)) {
947 __be32 _mask, *mp;
948
949 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
950 BUG_ON(mp == NULL);
951 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
952 if (*mp == ifa->ifa_mask &&
953 inet_ifa_match(rt->rt_src, ifa))
954 break;
955 }
956 if (!ifa && net_ratelimit()) {
957 printk(KERN_INFO "Wrong address mask " NIPQUAD_FMT " from "
958 "%s/" NIPQUAD_FMT "\n",
959 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
960 }
961 }
962 rcu_read_unlock();
963 in_dev_put(in_dev);
964 out:;
965 }
966
967 static void icmp_discard(struct sk_buff *skb)
968 {
969 }
970
971 /*
972 * Deal with incoming ICMP packets.
973 */
974 int icmp_rcv(struct sk_buff *skb)
975 {
976 struct icmphdr *icmph;
977 struct rtable *rt = skb->rtable;
978
979 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
980 int nh;
981
982 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
983 XFRM_STATE_ICMP))
984 goto drop;
985
986 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
987 goto drop;
988
989 nh = skb_network_offset(skb);
990 skb_set_network_header(skb, sizeof(*icmph));
991
992 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
993 goto drop;
994
995 skb_set_network_header(skb, nh);
996 }
997
998 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
999
1000 switch (skb->ip_summed) {
1001 case CHECKSUM_COMPLETE:
1002 if (!csum_fold(skb->csum))
1003 break;
1004 /* fall through */
1005 case CHECKSUM_NONE:
1006 skb->csum = 0;
1007 if (__skb_checksum_complete(skb))
1008 goto error;
1009 }
1010
1011 if (!pskb_pull(skb, sizeof(*icmph)))
1012 goto error;
1013
1014 icmph = icmp_hdr(skb);
1015
1016 ICMPMSGIN_INC_STATS_BH(icmph->type);
1017 /*
1018 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1019 *
1020 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1021 * discarded.
1022 */
1023 if (icmph->type > NR_ICMP_TYPES)
1024 goto error;
1025
1026
1027 /*
1028 * Parse the ICMP message
1029 */
1030
1031 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1032 struct net *net;
1033
1034 net = dev_net(rt->u.dst.dev);
1035 /*
1036 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1037 * silently ignored (we let user decide with a sysctl).
1038 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1039 * discarded if to broadcast/multicast.
1040 */
1041 if ((icmph->type == ICMP_ECHO ||
1042 icmph->type == ICMP_TIMESTAMP) &&
1043 net->ipv4.sysctl_icmp_echo_ignore_broadcasts) {
1044 goto error;
1045 }
1046 if (icmph->type != ICMP_ECHO &&
1047 icmph->type != ICMP_TIMESTAMP &&
1048 icmph->type != ICMP_ADDRESS &&
1049 icmph->type != ICMP_ADDRESSREPLY) {
1050 goto error;
1051 }
1052 }
1053
1054 icmp_pointers[icmph->type].handler(skb);
1055
1056 drop:
1057 kfree_skb(skb);
1058 return 0;
1059 error:
1060 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1061 goto drop;
1062 }
1063
1064 /*
1065 * This table is the definition of how we handle ICMP.
1066 */
1067 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1068 [ICMP_ECHOREPLY] = {
1069 .handler = icmp_discard,
1070 },
1071 [1] = {
1072 .handler = icmp_discard,
1073 .error = 1,
1074 },
1075 [2] = {
1076 .handler = icmp_discard,
1077 .error = 1,
1078 },
1079 [ICMP_DEST_UNREACH] = {
1080 .handler = icmp_unreach,
1081 .error = 1,
1082 },
1083 [ICMP_SOURCE_QUENCH] = {
1084 .handler = icmp_unreach,
1085 .error = 1,
1086 },
1087 [ICMP_REDIRECT] = {
1088 .handler = icmp_redirect,
1089 .error = 1,
1090 },
1091 [6] = {
1092 .handler = icmp_discard,
1093 .error = 1,
1094 },
1095 [7] = {
1096 .handler = icmp_discard,
1097 .error = 1,
1098 },
1099 [ICMP_ECHO] = {
1100 .handler = icmp_echo,
1101 },
1102 [9] = {
1103 .handler = icmp_discard,
1104 .error = 1,
1105 },
1106 [10] = {
1107 .handler = icmp_discard,
1108 .error = 1,
1109 },
1110 [ICMP_TIME_EXCEEDED] = {
1111 .handler = icmp_unreach,
1112 .error = 1,
1113 },
1114 [ICMP_PARAMETERPROB] = {
1115 .handler = icmp_unreach,
1116 .error = 1,
1117 },
1118 [ICMP_TIMESTAMP] = {
1119 .handler = icmp_timestamp,
1120 },
1121 [ICMP_TIMESTAMPREPLY] = {
1122 .handler = icmp_discard,
1123 },
1124 [ICMP_INFO_REQUEST] = {
1125 .handler = icmp_discard,
1126 },
1127 [ICMP_INFO_REPLY] = {
1128 .handler = icmp_discard,
1129 },
1130 [ICMP_ADDRESS] = {
1131 .handler = icmp_address,
1132 },
1133 [ICMP_ADDRESSREPLY] = {
1134 .handler = icmp_address_reply,
1135 },
1136 };
1137
1138 static void __net_exit icmp_sk_exit(struct net *net)
1139 {
1140 int i;
1141
1142 for_each_possible_cpu(i)
1143 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1144 kfree(net->ipv4.icmp_sk);
1145 net->ipv4.icmp_sk = NULL;
1146 }
1147
1148 static int __net_init icmp_sk_init(struct net *net)
1149 {
1150 int i, err;
1151
1152 net->ipv4.icmp_sk =
1153 kzalloc(nr_cpu_ids * sizeof(struct sock *), GFP_KERNEL);
1154 if (net->ipv4.icmp_sk == NULL)
1155 return -ENOMEM;
1156
1157 for_each_possible_cpu(i) {
1158 struct sock *sk;
1159
1160 err = inet_ctl_sock_create(&sk, PF_INET,
1161 SOCK_RAW, IPPROTO_ICMP, net);
1162 if (err < 0)
1163 goto fail;
1164
1165 net->ipv4.icmp_sk[i] = sk;
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_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
1174 }
1175
1176 /* Control parameters for ECHO replies. */
1177 net->ipv4.sysctl_icmp_echo_ignore_all = 0;
1178 net->ipv4.sysctl_icmp_echo_ignore_broadcasts = 1;
1179
1180 /* Control parameter - ignore bogus broadcast responses? */
1181 net->ipv4.sysctl_icmp_ignore_bogus_error_responses = 1;
1182
1183 /*
1184 * Configurable global rate limit.
1185 *
1186 * ratelimit defines tokens/packet consumed for dst->rate_token
1187 * bucket ratemask defines which icmp types are ratelimited by
1188 * setting it's bit position.
1189 *
1190 * default:
1191 * dest unreachable (3), source quench (4),
1192 * time exceeded (11), parameter problem (12)
1193 */
1194
1195 net->ipv4.sysctl_icmp_ratelimit = 1 * HZ;
1196 net->ipv4.sysctl_icmp_ratemask = 0x1818;
1197 net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
1198
1199 return 0;
1200
1201 fail:
1202 for_each_possible_cpu(i)
1203 inet_ctl_sock_destroy(net->ipv4.icmp_sk[i]);
1204 kfree(net->ipv4.icmp_sk);
1205 return err;
1206 }
1207
1208 static struct pernet_operations __net_initdata icmp_sk_ops = {
1209 .init = icmp_sk_init,
1210 .exit = icmp_sk_exit,
1211 };
1212
1213 int __init icmp_init(void)
1214 {
1215 return register_pernet_device(&icmp_sk_ops);
1216 }
1217
1218 EXPORT_SYMBOL(icmp_err_convert);
1219 EXPORT_SYMBOL(icmp_send);
1220 EXPORT_SYMBOL(icmp_statistics);
1221 EXPORT_SYMBOL(xrlim_allow);
linux 2.6 git repository for openrd