projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[NETNS]: Add namespace parameter to __ip_route_output_key.
[deliverable/linux.git] / net / ipv4 / icmp.c
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 inline 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, token = dst->rate_tokens;
279 int rc = 0;
280
281 now = jiffies;
282 token += now - dst->rate_last;
283 dst->rate_last = now;
284 if (token > XRLIM_BURST_FACTOR * timeout)
285 token = XRLIM_BURST_FACTOR * timeout;
286 if (token >= timeout) {
287 token -= timeout;
288 rc = 1;
289 }
290 dst->rate_tokens = token;
291 return rc;
292 }
293
294 static inline int icmpv4_xrlim_allow(struct rtable *rt, int type, int code)
295 {
296 struct dst_entry *dst = &rt->u.dst;
297 int rc = 1;
298
299 if (type > NR_ICMP_TYPES)
300 goto out;
301
302 /* Don't limit PMTU discovery. */
303 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
304 goto out;
305
306 /* No rate limit on loopback */
307 if (dst->dev && (dst->dev->flags&IFF_LOOPBACK))
308 goto out;
309
310 /* Limit if icmp type is enabled in ratemask. */
311 if ((1 << type) & sysctl_icmp_ratemask)
312 rc = xrlim_allow(dst, sysctl_icmp_ratelimit);
313 out:
314 return rc;
315 }
316
317 /*
318 * Maintain the counters used in the SNMP statistics for outgoing ICMP
319 */
320 void icmp_out_count(unsigned char type)
321 {
322 ICMPMSGOUT_INC_STATS(type);
323 ICMP_INC_STATS(ICMP_MIB_OUTMSGS);
324 }
325
326 /*
327 * Checksum each fragment, and on the first include the headers and final
328 * checksum.
329 */
330 static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
331 struct sk_buff *skb)
332 {
333 struct icmp_bxm *icmp_param = (struct icmp_bxm *)from;
334 __wsum csum;
335
336 csum = skb_copy_and_csum_bits(icmp_param->skb,
337 icmp_param->offset + offset,
338 to, len, 0);
339
340 skb->csum = csum_block_add(skb->csum, csum, odd);
341 if (icmp_pointers[icmp_param->data.icmph.type].error)
342 nf_ct_attach(skb, icmp_param->skb);
343 return 0;
344 }
345
346 static void icmp_push_reply(struct icmp_bxm *icmp_param,
347 struct ipcm_cookie *ipc, struct rtable *rt)
348 {
349 struct sk_buff *skb;
350
351 if (ip_append_data(icmp_socket->sk, icmp_glue_bits, icmp_param,
352 icmp_param->data_len+icmp_param->head_len,
353 icmp_param->head_len,
354 ipc, rt, MSG_DONTWAIT) < 0)
355 ip_flush_pending_frames(icmp_socket->sk);
356 else if ((skb = skb_peek(&icmp_socket->sk->sk_write_queue)) != NULL) {
357 struct icmphdr *icmph = icmp_hdr(skb);
358 __wsum csum = 0;
359 struct sk_buff *skb1;
360
361 skb_queue_walk(&icmp_socket->sk->sk_write_queue, skb1) {
362 csum = csum_add(csum, skb1->csum);
363 }
364 csum = csum_partial_copy_nocheck((void *)&icmp_param->data,
365 (char *)icmph,
366 icmp_param->head_len, csum);
367 icmph->checksum = csum_fold(csum);
368 skb->ip_summed = CHECKSUM_NONE;
369 ip_push_pending_frames(icmp_socket->sk);
370 }
371 }
372
373 /*
374 * Driving logic for building and sending ICMP messages.
375 */
376
377 static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
378 {
379 struct sock *sk = icmp_socket->sk;
380 struct inet_sock *inet = inet_sk(sk);
381 struct ipcm_cookie ipc;
382 struct rtable *rt = (struct rtable *)skb->dst;
383 __be32 daddr;
384
385 if (ip_options_echo(&icmp_param->replyopts, skb))
386 return;
387
388 if (icmp_xmit_lock())
389 return;
390
391 icmp_param->data.icmph.checksum = 0;
392
393 inet->tos = ip_hdr(skb)->tos;
394 daddr = ipc.addr = rt->rt_src;
395 ipc.opt = NULL;
396 if (icmp_param->replyopts.optlen) {
397 ipc.opt = &icmp_param->replyopts;
398 if (ipc.opt->srr)
399 daddr = icmp_param->replyopts.faddr;
400 }
401 {
402 struct flowi fl = { .nl_u = { .ip4_u =
403 { .daddr = daddr,
404 .saddr = rt->rt_spec_dst,
405 .tos = RT_TOS(ip_hdr(skb)->tos) } },
406 .proto = IPPROTO_ICMP };
407 security_skb_classify_flow(skb, &fl);
408 if (ip_route_output_key(&rt, &fl))
409 goto out_unlock;
410 }
411 if (icmpv4_xrlim_allow(rt, icmp_param->data.icmph.type,
412 icmp_param->data.icmph.code))
413 icmp_push_reply(icmp_param, &ipc, rt);
414 ip_rt_put(rt);
415 out_unlock:
416 icmp_xmit_unlock();
417 }
418
419
420 /*
421 * Send an ICMP message in response to a situation
422 *
423 * RFC 1122: 3.2.2 MUST send at least the IP header and 8 bytes of header.
424 * MAY send more (we do).
425 * MUST NOT change this header information.
426 * MUST NOT reply to a multicast/broadcast IP address.
427 * MUST NOT reply to a multicast/broadcast MAC address.
428 * MUST reply to only the first fragment.
429 */
430
431 void icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info)
432 {
433 struct iphdr *iph;
434 int room;
435 struct icmp_bxm icmp_param;
436 struct rtable *rt = (struct rtable *)skb_in->dst;
437 struct ipcm_cookie ipc;
438 __be32 saddr;
439 u8 tos;
440
441 if (!rt)
442 goto out;
443
444 /*
445 * Find the original header. It is expected to be valid, of course.
446 * Check this, icmp_send is called from the most obscure devices
447 * sometimes.
448 */
449 iph = ip_hdr(skb_in);
450
451 if ((u8 *)iph < skb_in->head ||
452 (skb_in->network_header + sizeof(*iph)) > skb_in->tail)
453 goto out;
454
455 /*
456 * No replies to physical multicast/broadcast
457 */
458 if (skb_in->pkt_type != PACKET_HOST)
459 goto out;
460
461 /*
462 * Now check at the protocol level
463 */
464 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
465 goto out;
466
467 /*
468 * Only reply to fragment 0. We byte re-order the constant
469 * mask for efficiency.
470 */
471 if (iph->frag_off & htons(IP_OFFSET))
472 goto out;
473
474 /*
475 * If we send an ICMP error to an ICMP error a mess would result..
476 */
477 if (icmp_pointers[type].error) {
478 /*
479 * We are an error, check if we are replying to an
480 * ICMP error
481 */
482 if (iph->protocol == IPPROTO_ICMP) {
483 u8 _inner_type, *itp;
484
485 itp = skb_header_pointer(skb_in,
486 skb_network_header(skb_in) +
487 (iph->ihl << 2) +
488 offsetof(struct icmphdr,
489 type) -
490 skb_in->data,
491 sizeof(_inner_type),
492 &_inner_type);
493 if (itp == NULL)
494 goto out;
495
496 /*
497 * Assume any unknown ICMP type is an error. This
498 * isn't specified by the RFC, but think about it..
499 */
500 if (*itp > NR_ICMP_TYPES ||
501 icmp_pointers[*itp].error)
502 goto out;
503 }
504 }
505
506 if (icmp_xmit_lock())
507 return;
508
509 /*
510 * Construct source address and options.
511 */
512
513 saddr = iph->daddr;
514 if (!(rt->rt_flags & RTCF_LOCAL)) {
515 struct net_device *dev = NULL;
516
517 if (rt->fl.iif && sysctl_icmp_errors_use_inbound_ifaddr)
518 dev = dev_get_by_index(&init_net, rt->fl.iif);
519
520 if (dev) {
521 saddr = inet_select_addr(dev, 0, RT_SCOPE_LINK);
522 dev_put(dev);
523 } else
524 saddr = 0;
525 }
526
527 tos = icmp_pointers[type].error ? ((iph->tos & IPTOS_TOS_MASK) |
528 IPTOS_PREC_INTERNETCONTROL) :
529 iph->tos;
530
531 if (ip_options_echo(&icmp_param.replyopts, skb_in))
532 goto out_unlock;
533
534
535 /*
536 * Prepare data for ICMP header.
537 */
538
539 icmp_param.data.icmph.type = type;
540 icmp_param.data.icmph.code = code;
541 icmp_param.data.icmph.un.gateway = info;
542 icmp_param.data.icmph.checksum = 0;
543 icmp_param.skb = skb_in;
544 icmp_param.offset = skb_network_offset(skb_in);
545 inet_sk(icmp_socket->sk)->tos = tos;
546 ipc.addr = iph->saddr;
547 ipc.opt = &icmp_param.replyopts;
548
549 {
550 struct flowi fl = {
551 .nl_u = {
552 .ip4_u = {
553 .daddr = icmp_param.replyopts.srr ?
554 icmp_param.replyopts.faddr :
555 iph->saddr,
556 .saddr = saddr,
557 .tos = RT_TOS(tos)
558 }
559 },
560 .proto = IPPROTO_ICMP,
561 .uli_u = {
562 .icmpt = {
563 .type = type,
564 .code = code
565 }
566 }
567 };
568 int err;
569 struct rtable *rt2;
570
571 security_skb_classify_flow(skb_in, &fl);
572 if (__ip_route_output_key(&init_net, &rt, &fl))
573 goto out_unlock;
574
575 /* No need to clone since we're just using its address. */
576 rt2 = rt;
577
578 err = xfrm_lookup((struct dst_entry **)&rt, &fl, NULL, 0);
579 switch (err) {
580 case 0:
581 if (rt != rt2)
582 goto route_done;
583 break;
584 case -EPERM:
585 rt = NULL;
586 break;
587 default:
588 goto out_unlock;
589 }
590
591 if (xfrm_decode_session_reverse(skb_in, &fl, AF_INET))
592 goto out_unlock;
593
594 if (inet_addr_type(&init_net, fl.fl4_src) == RTN_LOCAL)
595 err = __ip_route_output_key(&init_net, &rt2, &fl);
596 else {
597 struct flowi fl2 = {};
598 struct dst_entry *odst;
599
600 fl2.fl4_dst = fl.fl4_src;
601 if (ip_route_output_key(&rt2, &fl2))
602 goto out_unlock;
603
604 /* Ugh! */
605 odst = skb_in->dst;
606 err = ip_route_input(skb_in, fl.fl4_dst, fl.fl4_src,
607 RT_TOS(tos), rt2->u.dst.dev);
608
609 dst_release(&rt2->u.dst);
610 rt2 = (struct rtable *)skb_in->dst;
611 skb_in->dst = odst;
612 }
613
614 if (err)
615 goto out_unlock;
616
617 err = xfrm_lookup((struct dst_entry **)&rt2, &fl, NULL,
618 XFRM_LOOKUP_ICMP);
619 if (err == -ENOENT) {
620 if (!rt)
621 goto out_unlock;
622 goto route_done;
623 }
624
625 dst_release(&rt->u.dst);
626 rt = rt2;
627
628 if (err)
629 goto out_unlock;
630 }
631
632 route_done:
633 if (!icmpv4_xrlim_allow(rt, type, code))
634 goto ende;
635
636 /* RFC says return as much as we can without exceeding 576 bytes. */
637
638 room = dst_mtu(&rt->u.dst);
639 if (room > 576)
640 room = 576;
641 room -= sizeof(struct iphdr) + icmp_param.replyopts.optlen;
642 room -= sizeof(struct icmphdr);
643
644 icmp_param.data_len = skb_in->len - icmp_param.offset;
645 if (icmp_param.data_len > room)
646 icmp_param.data_len = room;
647 icmp_param.head_len = sizeof(struct icmphdr);
648
649 icmp_push_reply(&icmp_param, &ipc, rt);
650 ende:
651 ip_rt_put(rt);
652 out_unlock:
653 icmp_xmit_unlock();
654 out:;
655 }
656
657
658 /*
659 * Handle ICMP_DEST_UNREACH, ICMP_TIME_EXCEED, and ICMP_QUENCH.
660 */
661
662 static void icmp_unreach(struct sk_buff *skb)
663 {
664 struct iphdr *iph;
665 struct icmphdr *icmph;
666 int hash, protocol;
667 struct net_protocol *ipprot;
668 u32 info = 0;
669
670 /*
671 * Incomplete header ?
672 * Only checks for the IP header, there should be an
673 * additional check for longer headers in upper levels.
674 */
675
676 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
677 goto out_err;
678
679 icmph = icmp_hdr(skb);
680 iph = (struct iphdr *)skb->data;
681
682 if (iph->ihl < 5) /* Mangled header, drop. */
683 goto out_err;
684
685 if (icmph->type == ICMP_DEST_UNREACH) {
686 switch (icmph->code & 15) {
687 case ICMP_NET_UNREACH:
688 case ICMP_HOST_UNREACH:
689 case ICMP_PROT_UNREACH:
690 case ICMP_PORT_UNREACH:
691 break;
692 case ICMP_FRAG_NEEDED:
693 if (ipv4_config.no_pmtu_disc) {
694 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: "
695 "fragmentation needed "
696 "and DF set.\n",
697 NIPQUAD(iph->daddr));
698 } else {
699 info = ip_rt_frag_needed(iph,
700 ntohs(icmph->un.frag.mtu));
701 if (!info)
702 goto out;
703 }
704 break;
705 case ICMP_SR_FAILED:
706 LIMIT_NETDEBUG(KERN_INFO "ICMP: %u.%u.%u.%u: Source "
707 "Route Failed.\n",
708 NIPQUAD(iph->daddr));
709 break;
710 default:
711 break;
712 }
713 if (icmph->code > NR_ICMP_UNREACH)
714 goto out;
715 } else if (icmph->type == ICMP_PARAMETERPROB)
716 info = ntohl(icmph->un.gateway) >> 24;
717
718 /*
719 * Throw it at our lower layers
720 *
721 * RFC 1122: 3.2.2 MUST extract the protocol ID from the passed
722 * header.
723 * RFC 1122: 3.2.2.1 MUST pass ICMP unreach messages to the
724 * transport layer.
725 * RFC 1122: 3.2.2.2 MUST pass ICMP time expired messages to
726 * transport layer.
727 */
728
729 /*
730 * Check the other end isnt violating RFC 1122. Some routers send
731 * bogus responses to broadcast frames. If you see this message
732 * first check your netmask matches at both ends, if it does then
733 * get the other vendor to fix their kit.
734 */
735
736 if (!sysctl_icmp_ignore_bogus_error_responses &&
737 inet_addr_type(&init_net, iph->daddr) == RTN_BROADCAST) {
738 if (net_ratelimit())
739 printk(KERN_WARNING "%u.%u.%u.%u sent an invalid ICMP "
740 "type %u, code %u "
741 "error to a broadcast: %u.%u.%u.%u on %s\n",
742 NIPQUAD(ip_hdr(skb)->saddr),
743 icmph->type, icmph->code,
744 NIPQUAD(iph->daddr),
745 skb->dev->name);
746 goto out;
747 }
748
749 /* Checkin full IP header plus 8 bytes of protocol to
750 * avoid additional coding at protocol handlers.
751 */
752 if (!pskb_may_pull(skb, iph->ihl * 4 + 8))
753 goto out;
754
755 iph = (struct iphdr *)skb->data;
756 protocol = iph->protocol;
757
758 /*
759 * Deliver ICMP message to raw sockets. Pretty useless feature?
760 */
761 raw_icmp_error(skb, protocol, info);
762
763 hash = protocol & (MAX_INET_PROTOS - 1);
764 rcu_read_lock();
765 ipprot = rcu_dereference(inet_protos[hash]);
766 if (ipprot && ipprot->err_handler)
767 ipprot->err_handler(skb, info);
768 rcu_read_unlock();
769
770 out:
771 return;
772 out_err:
773 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
774 goto out;
775 }
776
777
778 /*
779 * Handle ICMP_REDIRECT.
780 */
781
782 static void icmp_redirect(struct sk_buff *skb)
783 {
784 struct iphdr *iph;
785
786 if (skb->len < sizeof(struct iphdr))
787 goto out_err;
788
789 /*
790 * Get the copied header of the packet that caused the redirect
791 */
792 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
793 goto out;
794
795 iph = (struct iphdr *)skb->data;
796
797 switch (icmp_hdr(skb)->code & 7) {
798 case ICMP_REDIR_NET:
799 case ICMP_REDIR_NETTOS:
800 /*
801 * As per RFC recommendations now handle it as a host redirect.
802 */
803 case ICMP_REDIR_HOST:
804 case ICMP_REDIR_HOSTTOS:
805 ip_rt_redirect(ip_hdr(skb)->saddr, iph->daddr,
806 icmp_hdr(skb)->un.gateway,
807 iph->saddr, skb->dev);
808 break;
809 }
810 out:
811 return;
812 out_err:
813 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
814 goto out;
815 }
816
817 /*
818 * Handle ICMP_ECHO ("ping") requests.
819 *
820 * RFC 1122: 3.2.2.6 MUST have an echo server that answers ICMP echo
821 * requests.
822 * RFC 1122: 3.2.2.6 Data received in the ICMP_ECHO request MUST be
823 * included in the reply.
824 * RFC 1812: 4.3.3.6 SHOULD have a config option for silently ignoring
825 * echo requests, MUST have default=NOT.
826 * See also WRT handling of options once they are done and working.
827 */
828
829 static void icmp_echo(struct sk_buff *skb)
830 {
831 if (!sysctl_icmp_echo_ignore_all) {
832 struct icmp_bxm icmp_param;
833
834 icmp_param.data.icmph = *icmp_hdr(skb);
835 icmp_param.data.icmph.type = ICMP_ECHOREPLY;
836 icmp_param.skb = skb;
837 icmp_param.offset = 0;
838 icmp_param.data_len = skb->len;
839 icmp_param.head_len = sizeof(struct icmphdr);
840 icmp_reply(&icmp_param, skb);
841 }
842 }
843
844 /*
845 * Handle ICMP Timestamp requests.
846 * RFC 1122: 3.2.2.8 MAY implement ICMP timestamp requests.
847 * SHOULD be in the kernel for minimum random latency.
848 * MUST be accurate to a few minutes.
849 * MUST be updated at least at 15Hz.
850 */
851 static void icmp_timestamp(struct sk_buff *skb)
852 {
853 struct timeval tv;
854 struct icmp_bxm icmp_param;
855 /*
856 * Too short.
857 */
858 if (skb->len < 4)
859 goto out_err;
860
861 /*
862 * Fill in the current time as ms since midnight UT:
863 */
864 do_gettimeofday(&tv);
865 icmp_param.data.times[1] = htonl((tv.tv_sec % 86400) * 1000 +
866 tv.tv_usec / 1000);
867 icmp_param.data.times[2] = icmp_param.data.times[1];
868 if (skb_copy_bits(skb, 0, &icmp_param.data.times[0], 4))
869 BUG();
870 icmp_param.data.icmph = *icmp_hdr(skb);
871 icmp_param.data.icmph.type = ICMP_TIMESTAMPREPLY;
872 icmp_param.data.icmph.code = 0;
873 icmp_param.skb = skb;
874 icmp_param.offset = 0;
875 icmp_param.data_len = 0;
876 icmp_param.head_len = sizeof(struct icmphdr) + 12;
877 icmp_reply(&icmp_param, skb);
878 out:
879 return;
880 out_err:
881 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
882 goto out;
883 }
884
885
886 /*
887 * Handle ICMP_ADDRESS_MASK requests. (RFC950)
888 *
889 * RFC1122 (3.2.2.9). A host MUST only send replies to
890 * ADDRESS_MASK requests if it's been configured as an address mask
891 * agent. Receiving a request doesn't constitute implicit permission to
892 * act as one. Of course, implementing this correctly requires (SHOULD)
893 * a way to turn the functionality on and off. Another one for sysctl(),
894 * I guess. -- MS
895 *
896 * RFC1812 (4.3.3.9). A router MUST implement it.
897 * A router SHOULD have switch turning it on/off.
898 * This switch MUST be ON by default.
899 *
900 * Gratuitous replies, zero-source replies are not implemented,
901 * that complies with RFC. DO NOT implement them!!! All the idea
902 * of broadcast addrmask replies as specified in RFC950 is broken.
903 * The problem is that it is not uncommon to have several prefixes
904 * on one physical interface. Moreover, addrmask agent can even be
905 * not aware of existing another prefixes.
906 * If source is zero, addrmask agent cannot choose correct prefix.
907 * Gratuitous mask announcements suffer from the same problem.
908 * RFC1812 explains it, but still allows to use ADDRMASK,
909 * that is pretty silly. --ANK
910 *
911 * All these rules are so bizarre, that I removed kernel addrmask
912 * support at all. It is wrong, it is obsolete, nobody uses it in
913 * any case. --ANK
914 *
915 * Furthermore you can do it with a usermode address agent program
916 * anyway...
917 */
918
919 static void icmp_address(struct sk_buff *skb)
920 {
921 #if 0
922 if (net_ratelimit())
923 printk(KERN_DEBUG "a guy asks for address mask. Who is it?\n");
924 #endif
925 }
926
927 /*
928 * RFC1812 (4.3.3.9). A router SHOULD listen all replies, and complain
929 * loudly if an inconsistency is found.
930 */
931
932 static void icmp_address_reply(struct sk_buff *skb)
933 {
934 struct rtable *rt = (struct rtable *)skb->dst;
935 struct net_device *dev = skb->dev;
936 struct in_device *in_dev;
937 struct in_ifaddr *ifa;
938
939 if (skb->len < 4 || !(rt->rt_flags&RTCF_DIRECTSRC))
940 goto out;
941
942 in_dev = in_dev_get(dev);
943 if (!in_dev)
944 goto out;
945 rcu_read_lock();
946 if (in_dev->ifa_list &&
947 IN_DEV_LOG_MARTIANS(in_dev) &&
948 IN_DEV_FORWARD(in_dev)) {
949 __be32 _mask, *mp;
950
951 mp = skb_header_pointer(skb, 0, sizeof(_mask), &_mask);
952 BUG_ON(mp == NULL);
953 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next) {
954 if (*mp == ifa->ifa_mask &&
955 inet_ifa_match(rt->rt_src, ifa))
956 break;
957 }
958 if (!ifa && net_ratelimit()) {
959 printk(KERN_INFO "Wrong address mask %u.%u.%u.%u from "
960 "%s/%u.%u.%u.%u\n",
961 NIPQUAD(*mp), dev->name, NIPQUAD(rt->rt_src));
962 }
963 }
964 rcu_read_unlock();
965 in_dev_put(in_dev);
966 out:;
967 }
968
969 static void icmp_discard(struct sk_buff *skb)
970 {
971 }
972
973 /*
974 * Deal with incoming ICMP packets.
975 */
976 int icmp_rcv(struct sk_buff *skb)
977 {
978 struct icmphdr *icmph;
979 struct rtable *rt = (struct rtable *)skb->dst;
980
981 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
982 int nh;
983
984 if (!(skb->sp && skb->sp->xvec[skb->sp->len - 1]->props.flags &
985 XFRM_STATE_ICMP))
986 goto drop;
987
988 if (!pskb_may_pull(skb, sizeof(*icmph) + sizeof(struct iphdr)))
989 goto drop;
990
991 nh = skb_network_offset(skb);
992 skb_set_network_header(skb, sizeof(*icmph));
993
994 if (!xfrm4_policy_check_reverse(NULL, XFRM_POLICY_IN, skb))
995 goto drop;
996
997 skb_set_network_header(skb, nh);
998 }
999
1000 ICMP_INC_STATS_BH(ICMP_MIB_INMSGS);
1001
1002 switch (skb->ip_summed) {
1003 case CHECKSUM_COMPLETE:
1004 if (!csum_fold(skb->csum))
1005 break;
1006 /* fall through */
1007 case CHECKSUM_NONE:
1008 skb->csum = 0;
1009 if (__skb_checksum_complete(skb))
1010 goto error;
1011 }
1012
1013 __skb_pull(skb, sizeof(*icmph));
1014
1015 icmph = icmp_hdr(skb);
1016
1017 ICMPMSGIN_INC_STATS_BH(icmph->type);
1018 /*
1019 * 18 is the highest 'known' ICMP type. Anything else is a mystery
1020 *
1021 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
1022 * discarded.
1023 */
1024 if (icmph->type > NR_ICMP_TYPES)
1025 goto error;
1026
1027
1028 /*
1029 * Parse the ICMP message
1030 */
1031
1032 if (rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) {
1033 /*
1034 * RFC 1122: 3.2.2.6 An ICMP_ECHO to broadcast MAY be
1035 * silently ignored (we let user decide with a sysctl).
1036 * RFC 1122: 3.2.2.8 An ICMP_TIMESTAMP MAY be silently
1037 * discarded if to broadcast/multicast.
1038 */
1039 if ((icmph->type == ICMP_ECHO ||
1040 icmph->type == ICMP_TIMESTAMP) &&
1041 sysctl_icmp_echo_ignore_broadcasts) {
1042 goto error;
1043 }
1044 if (icmph->type != ICMP_ECHO &&
1045 icmph->type != ICMP_TIMESTAMP &&
1046 icmph->type != ICMP_ADDRESS &&
1047 icmph->type != ICMP_ADDRESSREPLY) {
1048 goto error;
1049 }
1050 }
1051
1052 icmp_pointers[icmph->type].handler(skb);
1053
1054 drop:
1055 kfree_skb(skb);
1056 return 0;
1057 error:
1058 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
1059 goto drop;
1060 }
1061
1062 /*
1063 * This table is the definition of how we handle ICMP.
1064 */
1065 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
1066 [ICMP_ECHOREPLY] = {
1067 .handler = icmp_discard,
1068 },
1069 [1] = {
1070 .handler = icmp_discard,
1071 .error = 1,
1072 },
1073 [2] = {
1074 .handler = icmp_discard,
1075 .error = 1,
1076 },
1077 [ICMP_DEST_UNREACH] = {
1078 .handler = icmp_unreach,
1079 .error = 1,
1080 },
1081 [ICMP_SOURCE_QUENCH] = {
1082 .handler = icmp_unreach,
1083 .error = 1,
1084 },
1085 [ICMP_REDIRECT] = {
1086 .handler = icmp_redirect,
1087 .error = 1,
1088 },
1089 [6] = {
1090 .handler = icmp_discard,
1091 .error = 1,
1092 },
1093 [7] = {
1094 .handler = icmp_discard,
1095 .error = 1,
1096 },
1097 [ICMP_ECHO] = {
1098 .handler = icmp_echo,
1099 },
1100 [9] = {
1101 .handler = icmp_discard,
1102 .error = 1,
1103 },
1104 [10] = {
1105 .handler = icmp_discard,
1106 .error = 1,
1107 },
1108 [ICMP_TIME_EXCEEDED] = {
1109 .handler = icmp_unreach,
1110 .error = 1,
1111 },
1112 [ICMP_PARAMETERPROB] = {
1113 .handler = icmp_unreach,
1114 .error = 1,
1115 },
1116 [ICMP_TIMESTAMP] = {
1117 .handler = icmp_timestamp,
1118 },
1119 [ICMP_TIMESTAMPREPLY] = {
1120 .handler = icmp_discard,
1121 },
1122 [ICMP_INFO_REQUEST] = {
1123 .handler = icmp_discard,
1124 },
1125 [ICMP_INFO_REPLY] = {
1126 .handler = icmp_discard,
1127 },
1128 [ICMP_ADDRESS] = {
1129 .handler = icmp_address,
1130 },
1131 [ICMP_ADDRESSREPLY] = {
1132 .handler = icmp_address_reply,
1133 },
1134 };
1135
1136 void __init icmp_init(struct net_proto_family *ops)
1137 {
1138 struct inet_sock *inet;
1139 int i;
1140
1141 for_each_possible_cpu(i) {
1142 int err;
1143
1144 err = sock_create_kern(PF_INET, SOCK_RAW, IPPROTO_ICMP,
1145 &per_cpu(__icmp_socket, i));
1146
1147 if (err < 0)
1148 panic("Failed to create the ICMP control socket.\n");
1149
1150 per_cpu(__icmp_socket, i)->sk->sk_allocation = GFP_ATOMIC;
1151
1152 /* Enough space for 2 64K ICMP packets, including
1153 * sk_buff struct overhead.
1154 */
1155 per_cpu(__icmp_socket, i)->sk->sk_sndbuf =
1156 (2 * ((64 * 1024) + sizeof(struct sk_buff)));
1157
1158 inet = inet_sk(per_cpu(__icmp_socket, i)->sk);
1159 inet->uc_ttl = -1;
1160 inet->pmtudisc = IP_PMTUDISC_DONT;
1161
1162 /* Unhash it so that IP input processing does not even
1163 * see it, we do not wish this socket to see incoming
1164 * packets.
1165 */
1166 per_cpu(__icmp_socket, i)->sk->sk_prot->unhash(per_cpu(__icmp_socket, i)->sk);
1167 }
1168 }
1169
1170 EXPORT_SYMBOL(icmp_err_convert);
1171 EXPORT_SYMBOL(icmp_send);
1172 EXPORT_SYMBOL(icmp_statistics);
1173 EXPORT_SYMBOL(xrlim_allow);
Linux kernel - Deliverables
This page took 0.070861 seconds and 6 git commands to generate.