?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[TCP]: Avoid a divide in tcp_mtu_probing()
[linux-2.6/mini2440.git] / net / ipv4 / icmp.c
blobce5b4be559aa74fcc287405f7eb6e9417eec0977
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 /* Control parameters for ECHO replies. */
192 int sysctl_icmp_echo_ignore_all __read_mostly;
193 int sysctl_icmp_echo_ignore_broadcasts __read_mostly = 1;
194
195 /* Control parameter - ignore bogus broadcast responses? */
196 int sysctl_icmp_ignore_bogus_error_responses __read_mostly = 1;
197
198 /*
199 * Configurable global rate limit.
200 *
201 * ratelimit defines tokens/packet consumed for dst->rate_token bucket
202 * ratemask defines which icmp types are ratelimited by setting
203 * it's bit position.
204 *
205 * default:
206 * dest unreachable (3), source quench (4),
207 * time exceeded (11), parameter problem (12)
208 */
209
210 int sysctl_icmp_ratelimit __read_mostly = 1 * HZ;
211 int sysctl_icmp_ratemask __read_mostly = 0x1818;
212 int sysctl_icmp_errors_use_inbound_ifaddr __read_mostly;
213
214 /*
215 * ICMP control array. This specifies what to do with each ICMP.
216 */
217
218 struct icmp_control {
219 void (*handler)(struct sk_buff *skb);
220 short error; /* This ICMP is classed as an error message */
221 };
222
223 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
224
225 /*
226 * The ICMP socket(s). This is the most convenient way to flow control
227 * our ICMP output as well as maintain a clean interface throughout
228 * all layers. All Socketless IP sends will soon be gone.
229 *
230 * On SMP we have one ICMP socket per-cpu.
231 */
232 static DEFINE_PER_CPU(struct socket *, __icmp_socket) = NULL;
233 #define icmp_socket __get_cpu_var(__icmp_socket)
234
235 static __inline__ int icmp_xmit_lock(void)
236 {
237 local_bh_disable();
238
239 if (unlikely(!spin_trylock(&icmp_socket->sk->sk_lock.slock))) {
240 /* This can happen if the output path signals a
241 * dst_link_failure() for an outgoing ICMP packet.
242 */
243 local_bh_enable();
244 return 1;
245 }
246 return 0;
247 }
248
249 static void icmp_xmit_unlock(void)
250 {
251 spin_unlock_bh(&icmp_socket->sk->sk_lock.slock);
252 }
253
254 /*
255 * Send an ICMP frame.
256 */
257
258 /*
259 * Check transmit rate limitation for given message.
260 * The rate information is held in the destination cache now.
261 * This function is generic and could be used for other purposes
262 * too. It uses a Token bucket filter as suggested by Alexey Kuznetsov.
263 *
264 * Note that the same dst_entry fields are modified by functions in
265 * route.c too, but these work for packet destinations while xrlim_allow
266 * works for icmp destinations. This means the rate limiting information
267 * for one "ip object" is shared - and these ICMPs are twice limited:
268 * by source and by destination.
269 *
270 * RFC 1812: 4.3.2.8 SHOULD be able to limit error message rate
271 * SHOULD allow setting of rate limits
272 *
273 * Shared between ICMPv4 and ICMPv6.
274 */
275 #define XRLIM_BURST_FACTOR 6
276 int xrlim_allow(struct dst_entry *dst, int timeout)
277 {
278 unsigned long now;
279 int rc = 0;
280
281 now = jiffies;
282 dst->rate_tokens += now - dst->rate_last;
283 dst->rate_last = now;
284 if (dst->rate_tokens > XRLIM_BURST_FACTOR * timeout)
285 dst->rate_tokens = XRLIM_BURST_FACTOR * timeout;
286 if (dst->rate_tokens >= timeout) {
287 dst->rate_tokens -= timeout;
288 rc = 1;
289 }
290 return rc;
291 }
292
293 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
294 {
295 struct dst_entry *dst = &rt->u.dst;
296 int rc = 1;
297
298 if (type > NR_ICMP_TYPES)
299 goto out;
300
301 /* Don't limit PMTU discovery. */
302 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
303 goto out;
304
305 /* No rate limit on loopback */
306 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
307 goto out;
308
309 /* Limit if icmp type is enabled in ratemask. */
310 if ((1 << type) & sysctl_icmp_ratemask)
311 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
312 out:
313 return rc;
314 }
315
316 /*
317 * Maintain the counters used in the SNMP statistics for outgoing ICMP
318 */
319 void icmp_out_count(unsigned char type)
320 {
321 ICMPMSGOUT_INC_STATS(type);
322 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
323 }
324
325 /*
326 * Checksum each fragment, and on the first include the headers and final
327 * checksum.
328 */
329 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
330 struct sk_buff *skb)
331 {
332 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
333 __wsum csum;
334
335 csum = skb_copy_and_csum_bits(icmp_param->skb,
336 icmp_param->offset + offset,
337 to, len, 0);
338
339 skb->csum = csum_block_add(skb->csum, csum, odd);
340 if (icmp_pointers[icmp_param->data.icmph.type].error)
341 nf_ct_attach(skb, icmp_param->skb);
342 return 0;
343 }
344
345 static void icmp_push_reply(struct icmp_bxm *icmp_param,
346 struct ipcm_cookie *ipc, struct rtable *rt)
347 {
348 struct sk_buff *skb;
349
350 if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
351 icmp_param->data_len+icmp_param->head_len,
352 icmp_param->head_len,
353 ipc, rt, MSG_DONTWAIT) < 0)
354 ip_flush_pending_frames(icmp_socket->sk);
355 else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
356 struct icmphdr *icmph = icmp_hdr(skb);
357 __wsum csum = 0;
358 struct sk_buff *skb1;
359
360 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
361 csum = csum_add(csum, skb1->csum);
362 }
363 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
364 (char *)icmph,
365 icmp_param->head_len, csum);
366 icmph->checksum = csum_fold(csum);
367 skb->ip_summed = CHECKSUM_NONE;
368 ip_push_pending_frames(icmp_socket->sk);
369 }
370 }
371
372 /*
373 * Driving logic for building and sending ICMP messages.
374 */
375
376 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
377 {
378 struct sock *sk = icmp_socket->sk;
379 struct inet_sock *inet = inet_sk(sk);
380 struct ipcm_cookie ipc;
381 struct rtable *rt = (struct rtable *)skb->dst;
382 __be32 daddr;
383
384 if (ip_options_echo(&icmp_param->replyopts, skb))
385 return;
386
387 if (icmp_xmit_lock())
388 return;
389
390 icmp_param->data.icmph.checksum = 0;
391
392 inet->tos = ip_hdr(skb)->tos;
393 daddr = ipc.addr = rt->rt_src;
394 ipc.opt = NULL;
395 if (icmp_param->replyopts.optlen) {
396 ipc.opt = &icmp_param->replyopts;
397 if (ipc.opt->srr)
398 daddr = icmp_param->replyopts.faddr;
399 }
400 {
401 struct flowi fl = { .nl_u = { .ip4_u =
402 { .daddr = daddr,
403 .saddr = rt->rt_spec_dst,
404 .tos = RT_TOS(ip_hdr(skb)->tos) } },
405 .proto = IPPROTO_ICMP };
406 security_skb_classify_flow(skb, &fl);
407 if (ip_route_output_key(&rt, &fl))
408 goto out_unlock;
409 }
410 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
411 icmp_param->data.icmph.code))
412 icmp_push_reply(icmp_param, &ipc, rt);
413 ip_rt_put(rt);
414 out_unlock:
415 icmp_xmit_unlock();
416 }
417
418
419 /*
420 * Send an ICMP message in response to a situation
421 *
422 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
423 * MAY send more (we do).
424 * MUST NOT change this header information.
425 * MUST NOT reply to a multicast/broadcast IP address.
426 * MUST NOT reply to a multicast/broadcast MAC address.
427 * MUST reply to only the first fragment.
428 */
429
430 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
431 {
432 struct iphdr *iph;
433 int room;
434 struct icmp_bxm icmp_param;
435 struct rtable *rt = (struct rtable *)skb_in->dst;
436 struct ipcm_cookie ipc;
437 __be32 saddr;
438 u8 tos;
439
440 if (!rt)
441 goto out;
442
443 /*
444 * Find the original header. It is expected to be valid, of course.
445 * Check this, icmp_send is called from the most obscure devices
446 * sometimes.
447 */
448 iph = ip_hdr(skb_in);
449
450 if ((u8 *)iph < skb_in->head ||
451 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
452 goto out;
453
454 /*
455 * No replies to physical multicast/broadcast
456 */
457 if (skb_in->pkt_type != PACKET_HOST)
458 goto out;
459
460 /*
461 * Now check at the protocol level
462 */
463 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
464 goto out;
465
466 /*
467 * Only reply to fragment 0. We byte re-order the constant
468 * mask for efficiency.
469 */
470 if (iph->frag_off & htons(IP_OFFSET))
471 goto out;
472
473 /*
474 * If we send an ICMP error to an ICMP error a mess would result..
475 */
476 if (icmp_pointers[type].error) {
477 /*
478 * We are an error, check if we are replying to an
479 * ICMP error
480 */
481 if (iph->protocol == IPPROTO_ICMP) {
482 u8 _inner_type, *itp;
483
484 itp = skb_header_pointer(skb_in,
485 skb_network_header(skb_in) +
486 (iph->ihl << 2) +
487 offsetof(struct icmphdr,
488 type) -
489 skb_in->data,
490 sizeof(_inner_type),
491 &_inner_type);
492 if (itp == NULL)
493 goto out;
494
495 /*
496 * Assume any unknown ICMP type is an error. This
497 * isn't specified by the RFC, but think about it..
498 */
499 if (*itp > NR_ICMP_TYPES ||
500 icmp_pointers[*itp].error)
501 goto out;
502 }
503 }
504
505 if (icmp_xmit_lock())
506 return;
507
508 /*
509 * Construct source address and options.
510 */
511
512 saddr = iph->daddr;
513 if (!(rt->rt_flags & RTCF_LOCAL)) {
514 struct net_device *dev = NULL;
515
516 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
517 dev = dev_get_by_index(&init_net, rt->fl.iif);
518
519 if (dev) {
520 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
521 dev_put(dev);
522 } else
523 saddr = 0;
524 }
525
526 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
527 IPTOS_PREC_INTERNETCONTROL) :
528 iph->tos;
529
530 if (ip_options_echo(&icmp_param.replyopts, skb_in))
531 goto out_unlock;
532
533
534 /*
535 * Prepare data for ICMP header.
536 */
537
538 icmp_param.data.icmph.type = type;
539 icmp_param.data.icmph.code = code;
540 icmp_param.data.icmph.un.gateway = info;
541 icmp_param.data.icmph.checksum = 0;
542 icmp_param.skb = skb_in;
543 icmp_param.offset = skb_network_offset(skb_in);
544 inet_sk(icmp_socket->sk)->tos = tos;
545 ipc.addr = iph->saddr;
546 ipc.opt = &icmp_param.replyopts;
547
548 {
549 struct flowi fl = {
550 .nl_u = {
551 .ip4_u = {
552 .daddr = icmp_param.replyopts.srr ?
553 icmp_param.replyopts.faddr :
554 iph->saddr,
555 .saddr = saddr,
556 .tos = RT_TOS(tos)
557 }
558 },
559 .proto = IPPROTO_ICMP,
560 .uli_u = {
561 .icmpt = {
562 .type = type,
563 .code = code
564 }
565 }
566 };
567 int err;
568 struct rtable *rt2;
569
570 security_skb_classify_flow(skb_in, &fl);
571 if (__ip_route_output_key(&rt, &fl))
572 goto out_unlock;
573
574 /* No need to clone since we're just using its address. */
575 rt2 = rt;
576
577 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
578 switch (err) {
579 case 0:
580 if (rt != rt2)
581 goto route_done;
582 break;
583 case -EPERM:
584 rt = NULL;
585 break;
586 default:
587 goto out_unlock;
588 }
589
590 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
591 goto out_unlock;
592
593 if (inet_addr_type(fl.fl4_src) == RTN_LOCAL)
594 err = __ip_route_output_key(&rt2, &fl);
595 else {
596 struct flowi fl2 = {};
597 struct dst_entry *odst;
598
599 fl2.fl4_dst = fl.fl4_src;
600 if (ip_route_output_key(&rt2, &fl2))
601 goto out_unlock;
602
603 /* Ugh! */
604 odst = skb_in->dst;
605 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
606 RT_TOS(tos), rt2->u.dst.dev);
607
608 dst_release(&rt2->u.dst);
609 rt2 = (struct rtable *)skb_in->dst;
610 skb_in->dst = odst;
611 }
612
613 if (err)
614 goto out_unlock;
615
616 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
617 XFRM_LOOKUP_ICMP);
618 if (err == -ENOENT) {
619 if (!rt)
620 goto out_unlock;
621 goto route_done;
622 }
623
624 dst_release(&rt->u.dst);
625 rt = rt2;
626
627 if (err)
628 goto out_unlock;
629 }
630
631 route_done:
632 if (!icmpv4_xrlim_allow(rt, type, code))
633 goto ende;
634
635 /* RFC says return as much as we can without exceeding 576 bytes. */
636
637 room = dst_mtu(&rt->u.dst);
638 if (room > 576)
639 room = 576;
640 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
641 room -= sizeof(struct icmphdr);
642
643 icmp_param.data_len = skb_in->len - icmp_param.offset;
644 if (icmp_param.data_len > room)
645 icmp_param.data_len = room;
646 icmp_param.head_len = sizeof(struct icmphdr);
647
648 icmp_push_reply(&icmp_param, &ipc, rt);
649 ende:
650 ip_rt_put(rt);
651 out_unlock:
652 icmp_xmit_unlock();
653 out:;
654 }
655
656
657 /*
658 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
659 */
660
661 static void icmp_unreach(struct sk_buff *skb)
662 {
663 struct iphdr *iph;
664 struct icmphdr *icmph;
665 int hash, protocol;
666 struct net_protocol *ipprot;
667 u32 info = 0;
668
669 /*
670 * Incomplete header ?
671 * Only checks for the IP header, there should be an
672 * additional check for longer headers in upper levels.
673 */
674
675 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
676 goto out_err;
677
678 icmph = icmp_hdr(skb);
679 iph = (struct iphdr *)skb->data;
680
681 if (iph->ihl < 5) /* Mangled header, drop. */
682 goto out_err;
683
684 if (icmph->type == ICMP_DEST_UNREACH) {
685 switch (icmph->code & 15) {
686 case ICMP_NET_UNREACH:
687 case ICMP_HOST_UNREACH:
688 case ICMP_PROT_UNREACH:
689 case ICMP_PORT_UNREACH:
690 break;
691 case ICMP_FRAG_NEEDED:
692 if (ipv4_config.no_pmtu_disc) {
693 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
694 "fragmentation needed "
695 "and DF set.\n",
696 NIPQUAD(iph->daddr));
697 } else {
698 info = ip_rt_frag_needed(iph,
699 ntohs(icmph->un.frag.mtu));
700 if (!info)
701 goto out;
702 }
703 break;
704 case ICMP_SR_FAILED:
705 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
706 "Route Failed.\n",
707 NIPQUAD(iph->daddr));
708 break;
709 default:
710 break;
711 }
712 if (icmph->code > NR_ICMP_UNREACH)
713 goto out;
714 } else if (icmph->type == ICMP_PARAMETERPROB)
715 info = ntohl(icmph->un.gateway) >> 24;
716
717 /*
718 * Throw it at our lower layers
719 *
720 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
721 * header.
722 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
723 * transport layer.
724 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
725 * transport layer.
726 */
727
728 /*
729 * Check the other end isnt violating RFC 1122. Some routers send
730 * bogus responses to broadcast frames. If you see this message
731 * first check your netmask matches at both ends, if it does then
732 * get the other vendor to fix their kit.
733 */
734
735 if (!sysctl_icmp_ignore_bogus_error_responses &&
736 inet_addr_type(iph->daddr) == RTN_BROADCAST) {
737 if (net_ratelimit())
738 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
739 "type %u, code %u "
740 "error to a broadcast: %u.%u.%u.%u on %s\n",
741 NIPQUAD(ip_hdr(skb)->saddr),
742 icmph->type, icmph->code,
743 NIPQUAD(iph->daddr),
744 skb->dev->name);
745 goto out;
746 }
747
748 /* Checkin full IP header plus 8 bytes of protocol to
749 * avoid additional coding at protocol handlers.
750 */
751 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
752 goto out;
753
754 iph = (struct iphdr *)skb->data;
755 protocol = iph->protocol;
756
757 /*
758 * Deliver ICMP message to raw sockets. Pretty useless feature?
759 */
760 raw_icmp_error(skb, protocol, info);
761
762 hash = protocol & (MAX_INET_PROTOS - 1);
763 rcu_read_lock();
764 ipprot = rcu_dereference(inet_protos[hash]);
765 if (ipprot && ipprot->err_handler)
766 ipprot->err_handler(skb, info);
767 rcu_read_unlock();
768
769 out:
770 return;
771 out_err:
772 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
773 goto out;
774 }
775
776
777 /*
778 * Handle ICMP_REDIRECT.
779 */
780
781 static void icmp_redirect(struct sk_buff *skb)
782 {
783 struct iphdr *iph;
784
785 if (skb->len < sizeof(struct iphdr))
786 goto out_err;
787
788 /*
789 * Get the copied header of the packet that caused the redirect
790 */
791 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
792 goto out;
793
794 iph = (struct iphdr *)skb->data;
795
796 switch (icmp_hdr(skb)->code & 7) {
797 case ICMP_REDIR_NET:
798 case ICMP_REDIR_NETTOS:
799 /*
800 * As per RFC recommendations now handle it as a host redirect.
801 */
802 case ICMP_REDIR_HOST:
803 case ICMP_REDIR_HOSTTOS:
804 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
805 icmp_hdr(skb)->un.gateway,
806 iph->saddr, skb->dev);
807 break;
808 }
809 out:
810 return;
811 out_err:
812 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
813 goto out;
814 }
815
816 /*
817 * Handle ICMP_ECHO ("ping") requests.
818 *
819 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
820 * requests.
821 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
822 * included in the reply.
823 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
824 * echo requests, MUST have default=NOT.
825 * See also WRT handling of options once they are done and working.
826 */
827
828 static void icmp_echo(struct sk_buff *skb)
829 {
830 if (!sysctl_icmp_echo_ignore_all) {
831 struct icmp_bxm icmp_param;
832
833 icmp_param.data.icmph = *icmp_hdr(skb);
834 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
835 icmp_param.skb = skb;
836 icmp_param.offset = 0;
837 icmp_param.data_len = skb->len;
838 icmp_param.head_len = sizeof(struct icmphdr);
839 icmp_reply(&icmp_param, skb);
840 }
841 }
842
843 /*
844 * Handle ICMP Timestamp requests.
845 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
846 * SHOULD be in the kernel for minimum random latency.
847 * MUST be accurate to a few minutes.
848 * MUST be updated at least at 15Hz.
849 */
850 static void icmp_timestamp(struct sk_buff *skb)
851 {
852 struct timeval tv;
853 struct icmp_bxm icmp_param;
854 /*
855 * Too short.
856 */
857 if (skb->len < 4)
858 goto out_err;
859
860 /*
861 * Fill in the current time as ms since midnight UT:
862 */
863 do_gettimeofday(&tv);
864 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
865 tv.tv_usec / 1000);
866 icmp_param.data.times[2] = icmp_param.data.times[1];
867 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
868 BUG();
869 icmp_param.data.icmph = *icmp_hdr(skb);
870 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
871 icmp_param.data.icmph.code = 0;
872 icmp_param.skb = skb;
873 icmp_param.offset = 0;
874 icmp_param.data_len = 0;
875 icmp_param.head_len = sizeof(struct icmphdr) + 12;
876 icmp_reply(&icmp_param, skb);
877 out:
878 return;
879 out_err:
880 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
881 goto out;
882 }
883
884
885 /*
886 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
887 *
888 * RFC1122 (3.2.2.9). A host MUST only send replies to
889 * ADDRESS_MASK requests if it's been configured as an address mask
890 * agent. Receiving a request doesn't constitute implicit permission to
891 * act as one. Of course, implementing this correctly requires (SHOULD)
892 * a way to turn the functionality on and off. Another one for sysctl(),
893 * I guess. -- MS
894 *
895 * RFC1812 (4.3.3.9). A router MUST implement it.
896 * A router SHOULD have switch turning it on/off.
897 * This switch MUST be ON by default.
898 *
899 * Gratuitous replies, zero-source replies are not implemented,
900 * that complies with RFC. DO NOT implement them!!! All the idea
901 * of broadcast addrmask replies as specified in RFC950 is broken.
902 * The problem is that it is not uncommon to have several prefixes
903 * on one physical interface. Moreover, addrmask agent can even be
904 * not aware of existing another prefixes.
905 * If source is zero, addrmask agent cannot choose correct prefix.
906 * Gratuitous mask announcements suffer from the same problem.
907 * RFC1812 explains it, but still allows to use ADDRMASK,
908 * that is pretty silly. --ANK
909 *
910 * All these rules are so bizarre, that I removed kernel addrmask
911 * support at all. It is wrong, it is obsolete, nobody uses it in
912 * any case. --ANK
913 *
914 * Furthermore you can do it with a usermode address agent program
915 * anyway...
916 */
917
918 static void icmp_address(struct sk_buff *skb)
919 {
920 #if 0
921 if (net_ratelimit())
922 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
923 #endif
924 }
925
926 /*
927 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
928 * loudly if an inconsistency is found.
929 */
930
931 static void icmp_address_reply(struct sk_buff *skb)
932 {
933 struct rtable *rt = (struct rtable *)skb->dst;
934 struct net_device *dev = skb->dev;
935 struct in_device *in_dev;
936 struct in_ifaddr *ifa;
937
938 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
939 goto out;
940
941 in_dev = in_dev_get(dev);
942 if (!in_dev)
943 goto out;
944 rcu_read_lock();
945 if (in_dev->ifa_list &&
946 IN_DEV_LOG_MARTIANS(in_dev) &&
947 IN_DEV_FORWARD(in_dev)) {
948 __be32 _mask, *mp;
949
950 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
951 BUG_ON(mp == NULL);
952 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
953 if (*mp == ifa->ifa_mask &&
954 inet_ifa_match(rt->rt_src, ifa))
955 break;
956 }
957 if (!ifa && net_ratelimit()) {
958 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
959 "%s/%u.%u.%u.%u\n",
960 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
961 }
962 }
963 rcu_read_unlock();
964 in_dev_put(in_dev);
965 out:;
966 }
967
968 static void icmp_discard(struct sk_buff *skb)
969 {
970 }
971
972 /*
973 * Deal with incoming ICMP packets.
974 */
975 int icmp_rcv(struct sk_buff *skb)
976 {
977 struct icmphdr *icmph;
978 struct rtable *rt = (struct rtable *)skb->dst;
979
980 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
981 int nh;
982
983 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
984 XFRM_STATE_ICMP))
985 goto drop;
986
987 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
988 goto drop;
989
990 nh = skb_network_offset(skb);
991 skb_set_network_header(skb, sizeof(*icmph));
992
993 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
994 goto drop;
995
996 skb_set_network_header(skb, nh);
997 }
998
999 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1000
1001 switch (skb->ip_summed) {
1002 case CHECKSUM_COMPLETE:
1003 if (!csum_fold(skb->csum))
1004 break;
1005 /* fall through */
1006 case CHECKSUM_NONE:
1007 skb->csum = 0;
1008 if (__skb_checksum_complete(skb))
1009 goto error;
1010 }
1011
1012 __skb_pull(skb, sizeof(*icmph));
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 /*
1033 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1034 * silently ignored (we let user decide with a sysctl).
1035 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1036 * discarded if to broadcast/multicast.
1037 */
1038 if ((icmph->type == ICMP_ECHO ||
1039 icmph->type == ICMP_TIMESTAMP) &&
1040 sysctl_icmp_echo_ignore_broadcasts) {
1041 goto error;
1042 }
1043 if (icmph->type != ICMP_ECHO &&
1044 icmph->type != ICMP_TIMESTAMP &&
1045 icmph->type != ICMP_ADDRESS &&
1046 icmph->type != ICMP_ADDRESSREPLY) {
1047 goto error;
1048 }
1049 }
1050
1051 icmp_pointers[icmph->type].handler(skb);
1052
1053 drop:
1054 kfree_skb(skb);
1055 return 0;
1056 error:
1057 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1058 goto drop;
1059 }
1060
1061 /*
1062 * This table is the definition of how we handle ICMP.
1063 */
1064 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1065 [ICMP_ECHOREPLY] = {
1066 .handler = icmp_discard,
1067 },
1068 [1] = {
1069 .handler = icmp_discard,
1070 .error = 1,
1071 },
1072 [2] = {
1073 .handler = icmp_discard,
1074 .error = 1,
1075 },
1076 [ICMP_DEST_UNREACH] = {
1077 .handler = icmp_unreach,
1078 .error = 1,
1079 },
1080 [ICMP_SOURCE_QUENCH] = {
1081 .handler = icmp_unreach,
1082 .error = 1,
1083 },
1084 [ICMP_REDIRECT] = {
1085 .handler = icmp_redirect,
1086 .error = 1,
1087 },
1088 [6] = {
1089 .handler = icmp_discard,
1090 .error = 1,
1091 },
1092 [7] = {
1093 .handler = icmp_discard,
1094 .error = 1,
1095 },
1096 [ICMP_ECHO] = {
1097 .handler = icmp_echo,
1098 },
1099 [9] = {
1100 .handler = icmp_discard,
1101 .error = 1,
1102 },
1103 [10] = {
1104 .handler = icmp_discard,
1105 .error = 1,
1106 },
1107 [ICMP_TIME_EXCEEDED] = {
1108 .handler = icmp_unreach,
1109 .error = 1,
1110 },
1111 [ICMP_PARAMETERPROB] = {
1112 .handler = icmp_unreach,
1113 .error = 1,
1114 },
1115 [ICMP_TIMESTAMP] = {
1116 .handler = icmp_timestamp,
1117 },
1118 [ICMP_TIMESTAMPREPLY] = {
1119 .handler = icmp_discard,
1120 },
1121 [ICMP_INFO_REQUEST] = {
1122 .handler = icmp_discard,
1123 },
1124 [ICMP_INFO_REPLY] = {
1125 .handler = icmp_discard,
1126 },
1127 [ICMP_ADDRESS] = {
1128 .handler = icmp_address,
1129 },
1130 [ICMP_ADDRESSREPLY] = {
1131 .handler = icmp_address_reply,
1132 },
1133 };
1134
1135 void __init icmp_init(struct net_proto_family *ops)
1136 {
1137 struct inet_sock *inet;
1138 int i;
1139
1140 for_each_possible_cpu(i) {
1141 int err;
1142
1143 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1144 &per_cpu(__icmp_socket, i));
1145
1146 if (err < 0)
1147 panic("Failed to create the ICMP control socket.\n");
1148
1149 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1150
1151 /* Enough space for 2 64K ICMP packets, including
1152 * sk_buff struct overhead.
1153 */
1154 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1155 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1156
1157 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1158 inet->uc_ttl = -1;
1159 inet->pmtudisc = IP_PMTUDISC_DONT;
1160
1161 /* Unhash it so that IP input processing does not even
1162 * see it, we do not wish this socket to see incoming
1163 * packets.
1164 */
1165 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1166 }
1167 }
1168
1169 EXPORT_SYMBOL(icmp_err_convert);
1170 EXPORT_SYMBOL(icmp_send);
1171 EXPORT_SYMBOL(icmp_statistics);
1172 EXPORT_SYMBOL(xrlim_allow);
Support for the Chinese Samsung S3C2440 based development boards