xref: /openbmc/linux/net/ipv6/icmp.c (revision a11d206d)
1 /*
2  *	Internet Control Message Protocol (ICMPv6)
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	$Id: icmp.c,v 1.38 2002/02/08 03:57:19 davem Exp $
9  *
10  *	Based on net/ipv4/icmp.c
11  *
12  *	RFC 1885
13  *
14  *	This program is free software; you can redistribute it and/or
15  *      modify it under the terms of the GNU General Public License
16  *      as published by the Free Software Foundation; either version
17  *      2 of the License, or (at your option) any later version.
18  */
19 
20 /*
21  *	Changes:
22  *
23  *	Andi Kleen		:	exception handling
24  *	Andi Kleen			add rate limits. never reply to a icmp.
25  *					add more length checks and other fixes.
26  *	yoshfuji		:	ensure to sent parameter problem for
27  *					fragments.
28  *	YOSHIFUJI Hideaki @USAGI:	added sysctl for icmp rate limit.
29  *	Randy Dunlap and
30  *	YOSHIFUJI Hideaki @USAGI:	Per-interface statistics support
31  *	Kazunori MIYAZAWA @USAGI:       change output process to use ip6_append_data
32  */
33 
34 #include <linux/module.h>
35 #include <linux/errno.h>
36 #include <linux/types.h>
37 #include <linux/socket.h>
38 #include <linux/in.h>
39 #include <linux/kernel.h>
40 #include <linux/sched.h>
41 #include <linux/sockios.h>
42 #include <linux/net.h>
43 #include <linux/skbuff.h>
44 #include <linux/init.h>
45 #include <linux/netfilter.h>
46 
47 #ifdef CONFIG_SYSCTL
48 #include <linux/sysctl.h>
49 #endif
50 
51 #include <linux/inet.h>
52 #include <linux/netdevice.h>
53 #include <linux/icmpv6.h>
54 
55 #include <net/ip.h>
56 #include <net/sock.h>
57 
58 #include <net/ipv6.h>
59 #include <net/ip6_checksum.h>
60 #include <net/protocol.h>
61 #include <net/raw.h>
62 #include <net/rawv6.h>
63 #include <net/transp_v6.h>
64 #include <net/ip6_route.h>
65 #include <net/addrconf.h>
66 #include <net/icmp.h>
67 
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 
71 DEFINE_SNMP_STAT(struct icmpv6_mib, icmpv6_statistics) __read_mostly;
72 
73 /*
74  *	The ICMP socket(s). This is the most convenient way to flow control
75  *	our ICMP output as well as maintain a clean interface throughout
76  *	all layers. All Socketless IP sends will soon be gone.
77  *
78  *	On SMP we have one ICMP socket per-cpu.
79  */
80 static DEFINE_PER_CPU(struct socket *, __icmpv6_socket) = NULL;
81 #define icmpv6_socket	__get_cpu_var(__icmpv6_socket)
82 
83 static int icmpv6_rcv(struct sk_buff **pskb);
84 
85 static struct inet6_protocol icmpv6_protocol = {
86 	.handler	=	icmpv6_rcv,
87 	.flags		=	INET6_PROTO_FINAL,
88 };
89 
90 static __inline__ int icmpv6_xmit_lock(void)
91 {
92 	local_bh_disable();
93 
94 	if (unlikely(!spin_trylock(&icmpv6_socket->sk->sk_lock.slock))) {
95 		/* This can happen if the output path (f.e. SIT or
96 		 * ip6ip6 tunnel) signals dst_link_failure() for an
97 		 * outgoing ICMP6 packet.
98 		 */
99 		local_bh_enable();
100 		return 1;
101 	}
102 	return 0;
103 }
104 
105 static __inline__ void icmpv6_xmit_unlock(void)
106 {
107 	spin_unlock_bh(&icmpv6_socket->sk->sk_lock.slock);
108 }
109 
110 /*
111  * Slightly more convenient version of icmpv6_send.
112  */
113 void icmpv6_param_prob(struct sk_buff *skb, int code, int pos)
114 {
115 	icmpv6_send(skb, ICMPV6_PARAMPROB, code, pos, skb->dev);
116 	kfree_skb(skb);
117 }
118 
119 /*
120  * Figure out, may we reply to this packet with icmp error.
121  *
122  * We do not reply, if:
123  *	- it was icmp error message.
124  *	- it is truncated, so that it is known, that protocol is ICMPV6
125  *	  (i.e. in the middle of some exthdr)
126  *
127  *	--ANK (980726)
128  */
129 
130 static int is_ineligible(struct sk_buff *skb)
131 {
132 	int ptr = (u8*)(skb->nh.ipv6h+1) - skb->data;
133 	int len = skb->len - ptr;
134 	__u8 nexthdr = skb->nh.ipv6h->nexthdr;
135 
136 	if (len < 0)
137 		return 1;
138 
139 	ptr = ipv6_skip_exthdr(skb, ptr, &nexthdr);
140 	if (ptr < 0)
141 		return 0;
142 	if (nexthdr == IPPROTO_ICMPV6) {
143 		u8 _type, *tp;
144 		tp = skb_header_pointer(skb,
145 			ptr+offsetof(struct icmp6hdr, icmp6_type),
146 			sizeof(_type), &_type);
147 		if (tp == NULL ||
148 		    !(*tp & ICMPV6_INFOMSG_MASK))
149 			return 1;
150 	}
151 	return 0;
152 }
153 
154 static int sysctl_icmpv6_time __read_mostly = 1*HZ;
155 
156 /*
157  * Check the ICMP output rate limit
158  */
159 static inline int icmpv6_xrlim_allow(struct sock *sk, int type,
160 				     struct flowi *fl)
161 {
162 	struct dst_entry *dst;
163 	int res = 0;
164 
165 	/* Informational messages are not limited. */
166 	if (type & ICMPV6_INFOMSG_MASK)
167 		return 1;
168 
169 	/* Do not limit pmtu discovery, it would break it. */
170 	if (type == ICMPV6_PKT_TOOBIG)
171 		return 1;
172 
173 	/*
174 	 * Look up the output route.
175 	 * XXX: perhaps the expire for routing entries cloned by
176 	 * this lookup should be more aggressive (not longer than timeout).
177 	 */
178 	dst = ip6_route_output(sk, fl);
179 	if (dst->error) {
180 		IP6_INC_STATS(ip6_dst_idev(dst),
181 			      IPSTATS_MIB_OUTNOROUTES);
182 	} else if (dst->dev && (dst->dev->flags&IFF_LOOPBACK)) {
183 		res = 1;
184 	} else {
185 		struct rt6_info *rt = (struct rt6_info *)dst;
186 		int tmo = sysctl_icmpv6_time;
187 
188 		/* Give more bandwidth to wider prefixes. */
189 		if (rt->rt6i_dst.plen < 128)
190 			tmo >>= ((128 - rt->rt6i_dst.plen)>>5);
191 
192 		res = xrlim_allow(dst, tmo);
193 	}
194 	dst_release(dst);
195 	return res;
196 }
197 
198 /*
199  *	an inline helper for the "simple" if statement below
200  *	checks if parameter problem report is caused by an
201  *	unrecognized IPv6 option that has the Option Type
202  *	highest-order two bits set to 10
203  */
204 
205 static __inline__ int opt_unrec(struct sk_buff *skb, __u32 offset)
206 {
207 	u8 _optval, *op;
208 
209 	offset += skb->nh.raw - skb->data;
210 	op = skb_header_pointer(skb, offset, sizeof(_optval), &_optval);
211 	if (op == NULL)
212 		return 1;
213 	return (*op & 0xC0) == 0x80;
214 }
215 
216 static int icmpv6_push_pending_frames(struct sock *sk, struct flowi *fl, struct icmp6hdr *thdr, int len)
217 {
218 	struct sk_buff *skb;
219 	struct icmp6hdr *icmp6h;
220 	int err = 0;
221 
222 	if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
223 		goto out;
224 
225 	icmp6h = (struct icmp6hdr*) skb->h.raw;
226 	memcpy(icmp6h, thdr, sizeof(struct icmp6hdr));
227 	icmp6h->icmp6_cksum = 0;
228 
229 	if (skb_queue_len(&sk->sk_write_queue) == 1) {
230 		skb->csum = csum_partial((char *)icmp6h,
231 					sizeof(struct icmp6hdr), skb->csum);
232 		icmp6h->icmp6_cksum = csum_ipv6_magic(&fl->fl6_src,
233 						      &fl->fl6_dst,
234 						      len, fl->proto,
235 						      skb->csum);
236 	} else {
237 		u32 tmp_csum = 0;
238 
239 		skb_queue_walk(&sk->sk_write_queue, skb) {
240 			tmp_csum = csum_add(tmp_csum, skb->csum);
241 		}
242 
243 		tmp_csum = csum_partial((char *)icmp6h,
244 					sizeof(struct icmp6hdr), tmp_csum);
245 		tmp_csum = csum_ipv6_magic(&fl->fl6_src,
246 					   &fl->fl6_dst,
247 					   len, fl->proto, tmp_csum);
248 		icmp6h->icmp6_cksum = tmp_csum;
249 	}
250 	if (icmp6h->icmp6_cksum == 0)
251 		icmp6h->icmp6_cksum = -1;
252 	ip6_push_pending_frames(sk);
253 out:
254 	return err;
255 }
256 
257 struct icmpv6_msg {
258 	struct sk_buff	*skb;
259 	int		offset;
260 	uint8_t		type;
261 };
262 
263 static int icmpv6_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
264 {
265 	struct icmpv6_msg *msg = (struct icmpv6_msg *) from;
266 	struct sk_buff *org_skb = msg->skb;
267 	__u32 csum = 0;
268 
269 	csum = skb_copy_and_csum_bits(org_skb, msg->offset + offset,
270 				      to, len, csum);
271 	skb->csum = csum_block_add(skb->csum, csum, odd);
272 	if (!(msg->type & ICMPV6_INFOMSG_MASK))
273 		nf_ct_attach(skb, org_skb);
274 	return 0;
275 }
276 
277 #ifdef CONFIG_IPV6_MIP6
278 static void mip6_addr_swap(struct sk_buff *skb)
279 {
280 	struct ipv6hdr *iph = skb->nh.ipv6h;
281 	struct inet6_skb_parm *opt = IP6CB(skb);
282 	struct ipv6_destopt_hao *hao;
283 	struct in6_addr tmp;
284 	int off;
285 
286 	if (opt->dsthao) {
287 		off = ipv6_find_tlv(skb, opt->dsthao, IPV6_TLV_HAO);
288 		if (likely(off >= 0)) {
289 			hao = (struct ipv6_destopt_hao *)(skb->nh.raw + off);
290 			ipv6_addr_copy(&tmp, &iph->saddr);
291 			ipv6_addr_copy(&iph->saddr, &hao->addr);
292 			ipv6_addr_copy(&hao->addr, &tmp);
293 		}
294 	}
295 }
296 #else
297 static inline void mip6_addr_swap(struct sk_buff *skb) {}
298 #endif
299 
300 /*
301  *	Send an ICMP message in response to a packet in error
302  */
303 void icmpv6_send(struct sk_buff *skb, int type, int code, __u32 info,
304 		 struct net_device *dev)
305 {
306 	struct inet6_dev *idev = NULL;
307 	struct ipv6hdr *hdr = skb->nh.ipv6h;
308 	struct sock *sk;
309 	struct ipv6_pinfo *np;
310 	struct in6_addr *saddr = NULL;
311 	struct dst_entry *dst;
312 	struct icmp6hdr tmp_hdr;
313 	struct flowi fl;
314 	struct icmpv6_msg msg;
315 	int iif = 0;
316 	int addr_type = 0;
317 	int len;
318 	int hlimit, tclass;
319 	int err = 0;
320 
321 	if ((u8*)hdr < skb->head || (u8*)(hdr+1) > skb->tail)
322 		return;
323 
324 	/*
325 	 *	Make sure we respect the rules
326 	 *	i.e. RFC 1885 2.4(e)
327 	 *	Rule (e.1) is enforced by not using icmpv6_send
328 	 *	in any code that processes icmp errors.
329 	 */
330 	addr_type = ipv6_addr_type(&hdr->daddr);
331 
332 	if (ipv6_chk_addr(&hdr->daddr, skb->dev, 0))
333 		saddr = &hdr->daddr;
334 
335 	/*
336 	 *	Dest addr check
337 	 */
338 
339 	if ((addr_type & IPV6_ADDR_MULTICAST || skb->pkt_type != PACKET_HOST)) {
340 		if (type != ICMPV6_PKT_TOOBIG &&
341 		    !(type == ICMPV6_PARAMPROB &&
342 		      code == ICMPV6_UNK_OPTION &&
343 		      (opt_unrec(skb, info))))
344 			return;
345 
346 		saddr = NULL;
347 	}
348 
349 	addr_type = ipv6_addr_type(&hdr->saddr);
350 
351 	/*
352 	 *	Source addr check
353 	 */
354 
355 	if (addr_type & IPV6_ADDR_LINKLOCAL)
356 		iif = skb->dev->ifindex;
357 
358 	/*
359 	 *	Must not send error if the source does not uniquely
360 	 *	identify a single node (RFC2463 Section 2.4).
361 	 *	We check unspecified / multicast addresses here,
362 	 *	and anycast addresses will be checked later.
363 	 */
364 	if ((addr_type == IPV6_ADDR_ANY) || (addr_type & IPV6_ADDR_MULTICAST)) {
365 		LIMIT_NETDEBUG(KERN_DEBUG "icmpv6_send: addr_any/mcast source\n");
366 		return;
367 	}
368 
369 	/*
370 	 *	Never answer to a ICMP packet.
371 	 */
372 	if (is_ineligible(skb)) {
373 		LIMIT_NETDEBUG(KERN_DEBUG "icmpv6_send: no reply to icmp error\n");
374 		return;
375 	}
376 
377 	mip6_addr_swap(skb);
378 
379 	memset(&fl, 0, sizeof(fl));
380 	fl.proto = IPPROTO_ICMPV6;
381 	ipv6_addr_copy(&fl.fl6_dst, &hdr->saddr);
382 	if (saddr)
383 		ipv6_addr_copy(&fl.fl6_src, saddr);
384 	fl.oif = iif;
385 	fl.fl_icmp_type = type;
386 	fl.fl_icmp_code = code;
387 	security_skb_classify_flow(skb, &fl);
388 
389 	if (icmpv6_xmit_lock())
390 		return;
391 
392 	sk = icmpv6_socket->sk;
393 	np = inet6_sk(sk);
394 
395 	if (!icmpv6_xrlim_allow(sk, type, &fl))
396 		goto out;
397 
398 	tmp_hdr.icmp6_type = type;
399 	tmp_hdr.icmp6_code = code;
400 	tmp_hdr.icmp6_cksum = 0;
401 	tmp_hdr.icmp6_pointer = htonl(info);
402 
403 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst))
404 		fl.oif = np->mcast_oif;
405 
406 	err = ip6_dst_lookup(sk, &dst, &fl);
407 	if (err)
408 		goto out;
409 
410 	/*
411 	 * We won't send icmp if the destination is known
412 	 * anycast.
413 	 */
414 	if (((struct rt6_info *)dst)->rt6i_flags & RTF_ANYCAST) {
415 		LIMIT_NETDEBUG(KERN_DEBUG "icmpv6_send: acast source\n");
416 		goto out_dst_release;
417 	}
418 
419 	if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
420 		goto out;
421 
422 	if (ipv6_addr_is_multicast(&fl.fl6_dst))
423 		hlimit = np->mcast_hops;
424 	else
425 		hlimit = np->hop_limit;
426 	if (hlimit < 0)
427 		hlimit = dst_metric(dst, RTAX_HOPLIMIT);
428 	if (hlimit < 0)
429 		hlimit = ipv6_get_hoplimit(dst->dev);
430 
431 	tclass = np->tclass;
432 	if (tclass < 0)
433 		tclass = 0;
434 
435 	msg.skb = skb;
436 	msg.offset = skb->nh.raw - skb->data;
437 	msg.type = type;
438 
439 	len = skb->len - msg.offset;
440 	len = min_t(unsigned int, len, IPV6_MIN_MTU - sizeof(struct ipv6hdr) -sizeof(struct icmp6hdr));
441 	if (len < 0) {
442 		LIMIT_NETDEBUG(KERN_DEBUG "icmp: len problem\n");
443 		goto out_dst_release;
444 	}
445 
446 	idev = in6_dev_get(skb->dev);
447 
448 	err = ip6_append_data(sk, icmpv6_getfrag, &msg,
449 			      len + sizeof(struct icmp6hdr),
450 			      sizeof(struct icmp6hdr),
451 			      hlimit, tclass, NULL, &fl, (struct rt6_info*)dst,
452 			      MSG_DONTWAIT);
453 	if (err) {
454 		ip6_flush_pending_frames(sk);
455 		goto out_put;
456 	}
457 	err = icmpv6_push_pending_frames(sk, &fl, &tmp_hdr, len + sizeof(struct icmp6hdr));
458 
459 	if (type >= ICMPV6_DEST_UNREACH && type <= ICMPV6_PARAMPROB)
460 		ICMP6_INC_STATS_OFFSET_BH(idev, ICMP6_MIB_OUTDESTUNREACHS, type - ICMPV6_DEST_UNREACH);
461 	ICMP6_INC_STATS_BH(idev, ICMP6_MIB_OUTMSGS);
462 
463 out_put:
464 	if (likely(idev != NULL))
465 		in6_dev_put(idev);
466 out_dst_release:
467 	dst_release(dst);
468 out:
469 	icmpv6_xmit_unlock();
470 }
471 
472 static void icmpv6_echo_reply(struct sk_buff *skb)
473 {
474 	struct sock *sk;
475 	struct inet6_dev *idev;
476 	struct ipv6_pinfo *np;
477 	struct in6_addr *saddr = NULL;
478 	struct icmp6hdr *icmph = (struct icmp6hdr *) skb->h.raw;
479 	struct icmp6hdr tmp_hdr;
480 	struct flowi fl;
481 	struct icmpv6_msg msg;
482 	struct dst_entry *dst;
483 	int err = 0;
484 	int hlimit;
485 	int tclass;
486 
487 	saddr = &skb->nh.ipv6h->daddr;
488 
489 	if (!ipv6_unicast_destination(skb))
490 		saddr = NULL;
491 
492 	memcpy(&tmp_hdr, icmph, sizeof(tmp_hdr));
493 	tmp_hdr.icmp6_type = ICMPV6_ECHO_REPLY;
494 
495 	memset(&fl, 0, sizeof(fl));
496 	fl.proto = IPPROTO_ICMPV6;
497 	ipv6_addr_copy(&fl.fl6_dst, &skb->nh.ipv6h->saddr);
498 	if (saddr)
499 		ipv6_addr_copy(&fl.fl6_src, saddr);
500 	fl.oif = skb->dev->ifindex;
501 	fl.fl_icmp_type = ICMPV6_ECHO_REPLY;
502 	security_skb_classify_flow(skb, &fl);
503 
504 	if (icmpv6_xmit_lock())
505 		return;
506 
507 	sk = icmpv6_socket->sk;
508 	np = inet6_sk(sk);
509 
510 	if (!fl.oif && ipv6_addr_is_multicast(&fl.fl6_dst))
511 		fl.oif = np->mcast_oif;
512 
513 	err = ip6_dst_lookup(sk, &dst, &fl);
514 	if (err)
515 		goto out;
516 	if ((err = xfrm_lookup(&dst, &fl, sk, 0)) < 0)
517 		goto out;
518 
519 	if (ipv6_addr_is_multicast(&fl.fl6_dst))
520 		hlimit = np->mcast_hops;
521 	else
522 		hlimit = np->hop_limit;
523 	if (hlimit < 0)
524 		hlimit = dst_metric(dst, RTAX_HOPLIMIT);
525 	if (hlimit < 0)
526 		hlimit = ipv6_get_hoplimit(dst->dev);
527 
528 	tclass = np->tclass;
529 	if (tclass < 0)
530 		tclass = 0;
531 
532 	idev = in6_dev_get(skb->dev);
533 
534 	msg.skb = skb;
535 	msg.offset = 0;
536 	msg.type = ICMPV6_ECHO_REPLY;
537 
538 	err = ip6_append_data(sk, icmpv6_getfrag, &msg, skb->len + sizeof(struct icmp6hdr),
539 				sizeof(struct icmp6hdr), hlimit, tclass, NULL, &fl,
540 				(struct rt6_info*)dst, MSG_DONTWAIT);
541 
542 	if (err) {
543 		ip6_flush_pending_frames(sk);
544 		goto out_put;
545 	}
546 	err = icmpv6_push_pending_frames(sk, &fl, &tmp_hdr, skb->len + sizeof(struct icmp6hdr));
547 
548         ICMP6_INC_STATS_BH(idev, ICMP6_MIB_OUTECHOREPLIES);
549         ICMP6_INC_STATS_BH(idev, ICMP6_MIB_OUTMSGS);
550 
551 out_put:
552 	if (likely(idev != NULL))
553 		in6_dev_put(idev);
554 	dst_release(dst);
555 out:
556 	icmpv6_xmit_unlock();
557 }
558 
559 static void icmpv6_notify(struct sk_buff *skb, int type, int code, __be32 info)
560 {
561 	struct in6_addr *saddr, *daddr;
562 	struct inet6_protocol *ipprot;
563 	struct sock *sk;
564 	int inner_offset;
565 	int hash;
566 	u8 nexthdr;
567 
568 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
569 		return;
570 
571 	nexthdr = ((struct ipv6hdr *)skb->data)->nexthdr;
572 	if (ipv6_ext_hdr(nexthdr)) {
573 		/* now skip over extension headers */
574 		inner_offset = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
575 		if (inner_offset<0)
576 			return;
577 	} else {
578 		inner_offset = sizeof(struct ipv6hdr);
579 	}
580 
581 	/* Checkin header including 8 bytes of inner protocol header. */
582 	if (!pskb_may_pull(skb, inner_offset+8))
583 		return;
584 
585 	saddr = &skb->nh.ipv6h->saddr;
586 	daddr = &skb->nh.ipv6h->daddr;
587 
588 	/* BUGGG_FUTURE: we should try to parse exthdrs in this packet.
589 	   Without this we will not able f.e. to make source routed
590 	   pmtu discovery.
591 	   Corresponding argument (opt) to notifiers is already added.
592 	   --ANK (980726)
593 	 */
594 
595 	hash = nexthdr & (MAX_INET_PROTOS - 1);
596 
597 	rcu_read_lock();
598 	ipprot = rcu_dereference(inet6_protos[hash]);
599 	if (ipprot && ipprot->err_handler)
600 		ipprot->err_handler(skb, NULL, type, code, inner_offset, info);
601 	rcu_read_unlock();
602 
603 	read_lock(&raw_v6_lock);
604 	if ((sk = sk_head(&raw_v6_htable[hash])) != NULL) {
605 		while((sk = __raw_v6_lookup(sk, nexthdr, daddr, saddr,
606 					    IP6CB(skb)->iif))) {
607 			rawv6_err(sk, skb, NULL, type, code, inner_offset, info);
608 			sk = sk_next(sk);
609 		}
610 	}
611 	read_unlock(&raw_v6_lock);
612 }
613 
614 /*
615  *	Handle icmp messages
616  */
617 
618 static int icmpv6_rcv(struct sk_buff **pskb)
619 {
620 	struct sk_buff *skb = *pskb;
621 	struct net_device *dev = skb->dev;
622 	struct inet6_dev *idev = __in6_dev_get(dev);
623 	struct in6_addr *saddr, *daddr;
624 	struct ipv6hdr *orig_hdr;
625 	struct icmp6hdr *hdr;
626 	int type;
627 
628 	ICMP6_INC_STATS_BH(idev, ICMP6_MIB_INMSGS);
629 
630 	saddr = &skb->nh.ipv6h->saddr;
631 	daddr = &skb->nh.ipv6h->daddr;
632 
633 	/* Perform checksum. */
634 	switch (skb->ip_summed) {
635 	case CHECKSUM_COMPLETE:
636 		if (!csum_ipv6_magic(saddr, daddr, skb->len, IPPROTO_ICMPV6,
637 				     skb->csum))
638 			break;
639 		/* fall through */
640 	case CHECKSUM_NONE:
641 		skb->csum = ~csum_ipv6_magic(saddr, daddr, skb->len,
642 					     IPPROTO_ICMPV6, 0);
643 		if (__skb_checksum_complete(skb)) {
644 			LIMIT_NETDEBUG(KERN_DEBUG "ICMPv6 checksum failed [" NIP6_FMT " > " NIP6_FMT "]\n",
645 				       NIP6(*saddr), NIP6(*daddr));
646 			goto discard_it;
647 		}
648 	}
649 
650 	if (!pskb_pull(skb, sizeof(struct icmp6hdr)))
651 		goto discard_it;
652 
653 	hdr = (struct icmp6hdr *) skb->h.raw;
654 
655 	type = hdr->icmp6_type;
656 
657 	if (type >= ICMPV6_DEST_UNREACH && type <= ICMPV6_PARAMPROB)
658 		ICMP6_INC_STATS_OFFSET_BH(idev, ICMP6_MIB_INDESTUNREACHS, type - ICMPV6_DEST_UNREACH);
659 	else if (type >= ICMPV6_ECHO_REQUEST && type <= NDISC_REDIRECT)
660 		ICMP6_INC_STATS_OFFSET_BH(idev, ICMP6_MIB_INECHOS, type - ICMPV6_ECHO_REQUEST);
661 
662 	switch (type) {
663 	case ICMPV6_ECHO_REQUEST:
664 		icmpv6_echo_reply(skb);
665 		break;
666 
667 	case ICMPV6_ECHO_REPLY:
668 		/* we couldn't care less */
669 		break;
670 
671 	case ICMPV6_PKT_TOOBIG:
672 		/* BUGGG_FUTURE: if packet contains rthdr, we cannot update
673 		   standard destination cache. Seems, only "advanced"
674 		   destination cache will allow to solve this problem
675 		   --ANK (980726)
676 		 */
677 		if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
678 			goto discard_it;
679 		hdr = (struct icmp6hdr *) skb->h.raw;
680 		orig_hdr = (struct ipv6hdr *) (hdr + 1);
681 		rt6_pmtu_discovery(&orig_hdr->daddr, &orig_hdr->saddr, dev,
682 				   ntohl(hdr->icmp6_mtu));
683 
684 		/*
685 		 *	Drop through to notify
686 		 */
687 
688 	case ICMPV6_DEST_UNREACH:
689 	case ICMPV6_TIME_EXCEED:
690 	case ICMPV6_PARAMPROB:
691 		icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
692 		break;
693 
694 	case NDISC_ROUTER_SOLICITATION:
695 	case NDISC_ROUTER_ADVERTISEMENT:
696 	case NDISC_NEIGHBOUR_SOLICITATION:
697 	case NDISC_NEIGHBOUR_ADVERTISEMENT:
698 	case NDISC_REDIRECT:
699 		ndisc_rcv(skb);
700 		break;
701 
702 	case ICMPV6_MGM_QUERY:
703 		igmp6_event_query(skb);
704 		break;
705 
706 	case ICMPV6_MGM_REPORT:
707 		igmp6_event_report(skb);
708 		break;
709 
710 	case ICMPV6_MGM_REDUCTION:
711 	case ICMPV6_NI_QUERY:
712 	case ICMPV6_NI_REPLY:
713 	case ICMPV6_MLD2_REPORT:
714 	case ICMPV6_DHAAD_REQUEST:
715 	case ICMPV6_DHAAD_REPLY:
716 	case ICMPV6_MOBILE_PREFIX_SOL:
717 	case ICMPV6_MOBILE_PREFIX_ADV:
718 		break;
719 
720 	default:
721 		LIMIT_NETDEBUG(KERN_DEBUG "icmpv6: msg of unknown type\n");
722 
723 		/* informational */
724 		if (type & ICMPV6_INFOMSG_MASK)
725 			break;
726 
727 		/*
728 		 * error of unknown type.
729 		 * must pass to upper level
730 		 */
731 
732 		icmpv6_notify(skb, type, hdr->icmp6_code, hdr->icmp6_mtu);
733 	};
734 	kfree_skb(skb);
735 	return 0;
736 
737 discard_it:
738 	ICMP6_INC_STATS_BH(idev, ICMP6_MIB_INERRORS);
739 	kfree_skb(skb);
740 	return 0;
741 }
742 
743 /*
744  * Special lock-class for __icmpv6_socket:
745  */
746 static struct lock_class_key icmpv6_socket_sk_dst_lock_key;
747 
748 int __init icmpv6_init(struct net_proto_family *ops)
749 {
750 	struct sock *sk;
751 	int err, i, j;
752 
753 	for_each_possible_cpu(i) {
754 		err = sock_create_kern(PF_INET6, SOCK_RAW, IPPROTO_ICMPV6,
755 				       &per_cpu(__icmpv6_socket, i));
756 		if (err < 0) {
757 			printk(KERN_ERR
758 			       "Failed to initialize the ICMP6 control socket "
759 			       "(err %d).\n",
760 			       err);
761 			goto fail;
762 		}
763 
764 		sk = per_cpu(__icmpv6_socket, i)->sk;
765 		sk->sk_allocation = GFP_ATOMIC;
766 		/*
767 		 * Split off their lock-class, because sk->sk_dst_lock
768 		 * gets used from softirqs, which is safe for
769 		 * __icmpv6_socket (because those never get directly used
770 		 * via userspace syscalls), but unsafe for normal sockets.
771 		 */
772 		lockdep_set_class(&sk->sk_dst_lock,
773 				  &icmpv6_socket_sk_dst_lock_key);
774 
775 		/* Enough space for 2 64K ICMP packets, including
776 		 * sk_buff struct overhead.
777 		 */
778 		sk->sk_sndbuf =
779 			(2 * ((64 * 1024) + sizeof(struct sk_buff)));
780 
781 		sk->sk_prot->unhash(sk);
782 	}
783 
784 
785 	if (inet6_add_protocol(&icmpv6_protocol, IPPROTO_ICMPV6) < 0) {
786 		printk(KERN_ERR "Failed to register ICMP6 protocol\n");
787 		err = -EAGAIN;
788 		goto fail;
789 	}
790 
791 	return 0;
792 
793  fail:
794 	for (j = 0; j < i; j++) {
795 		if (!cpu_possible(j))
796 			continue;
797 		sock_release(per_cpu(__icmpv6_socket, j));
798 	}
799 
800 	return err;
801 }
802 
803 void icmpv6_cleanup(void)
804 {
805 	int i;
806 
807 	for_each_possible_cpu(i) {
808 		sock_release(per_cpu(__icmpv6_socket, i));
809 	}
810 	inet6_del_protocol(&icmpv6_protocol, IPPROTO_ICMPV6);
811 }
812 
813 static const struct icmp6_err {
814 	int err;
815 	int fatal;
816 } tab_unreach[] = {
817 	{	/* NOROUTE */
818 		.err	= ENETUNREACH,
819 		.fatal	= 0,
820 	},
821 	{	/* ADM_PROHIBITED */
822 		.err	= EACCES,
823 		.fatal	= 1,
824 	},
825 	{	/* Was NOT_NEIGHBOUR, now reserved */
826 		.err	= EHOSTUNREACH,
827 		.fatal	= 0,
828 	},
829 	{	/* ADDR_UNREACH	*/
830 		.err	= EHOSTUNREACH,
831 		.fatal	= 0,
832 	},
833 	{	/* PORT_UNREACH	*/
834 		.err	= ECONNREFUSED,
835 		.fatal	= 1,
836 	},
837 };
838 
839 int icmpv6_err_convert(int type, int code, int *err)
840 {
841 	int fatal = 0;
842 
843 	*err = EPROTO;
844 
845 	switch (type) {
846 	case ICMPV6_DEST_UNREACH:
847 		fatal = 1;
848 		if (code <= ICMPV6_PORT_UNREACH) {
849 			*err  = tab_unreach[code].err;
850 			fatal = tab_unreach[code].fatal;
851 		}
852 		break;
853 
854 	case ICMPV6_PKT_TOOBIG:
855 		*err = EMSGSIZE;
856 		break;
857 
858 	case ICMPV6_PARAMPROB:
859 		*err = EPROTO;
860 		fatal = 1;
861 		break;
862 
863 	case ICMPV6_TIME_EXCEED:
864 		*err = EHOSTUNREACH;
865 		break;
866 	};
867 
868 	return fatal;
869 }
870 
871 #ifdef CONFIG_SYSCTL
872 ctl_table ipv6_icmp_table[] = {
873 	{
874 		.ctl_name	= NET_IPV6_ICMP_RATELIMIT,
875 		.procname	= "ratelimit",
876 		.data		= &sysctl_icmpv6_time,
877 		.maxlen		= sizeof(int),
878 		.mode		= 0644,
879 		.proc_handler	= &proc_dointvec
880 	},
881 	{ .ctl_name = 0 },
882 };
883 #endif
884 
885