xref: /linux/net/ipv6/ip6_output.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/net/ipv4/ip_output.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  *
15  *	Changes:
16  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
17  *				extension headers are implemented.
18  *				route changes now work.
19  *				ip6_forward does not confuse sniffers.
20  *				etc.
21  *
22  *      H. von Brand    :       Added missing #include <linux/string.h>
23  *	Imran Patel	:	frag id should be in NBO
24  *      Kazunori MIYAZAWA @USAGI
25  *			:       add ip6_append_data and related functions
26  *				for datagram xmit
27  */
28 
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
41 
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
44 
45 #include <net/sock.h>
46 #include <net/snmp.h>
47 
48 #include <net/ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
54 #include <net/icmp.h>
55 #include <net/xfrm.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
58 
59 static int ip6_finish_output2(struct sk_buff *skb)
60 {
61 	struct dst_entry *dst = skb_dst(skb);
62 	struct net_device *dev = dst->dev;
63 	struct neighbour *neigh;
64 	struct in6_addr *nexthop;
65 	int ret;
66 
67 	skb->protocol = htons(ETH_P_IPV6);
68 	skb->dev = dev;
69 
70 	if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
72 
73 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
74 		    ((mroute6_socket(dev_net(dev), skb) &&
75 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 		     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 					 &ipv6_hdr(skb)->saddr))) {
78 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
79 
80 			/* Do not check for IFF_ALLMULTI; multicast routing
81 			   is not supported in any case.
82 			 */
83 			if (newskb)
84 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 					newskb, NULL, newskb->dev,
86 					dev_loopback_xmit);
87 
88 			if (ipv6_hdr(skb)->hop_limit == 0) {
89 				IP6_INC_STATS(dev_net(dev), idev,
90 					      IPSTATS_MIB_OUTDISCARDS);
91 				kfree_skb(skb);
92 				return 0;
93 			}
94 		}
95 
96 		IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
97 				skb->len);
98 
99 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 		    IPV6_ADDR_SCOPE_NODELOCAL &&
101 		    !(dev->flags & IFF_LOOPBACK)) {
102 			kfree_skb(skb);
103 			return 0;
104 		}
105 	}
106 
107 	rcu_read_lock_bh();
108 	nexthop = rt6_nexthop((struct rt6_info *)dst);
109 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 	if (unlikely(!neigh))
111 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 	if (!IS_ERR(neigh)) {
113 		ret = dst_neigh_output(dst, neigh, skb);
114 		rcu_read_unlock_bh();
115 		return ret;
116 	}
117 	rcu_read_unlock_bh();
118 
119 	IP6_INC_STATS(dev_net(dst->dev),
120 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
121 	kfree_skb(skb);
122 	return -EINVAL;
123 }
124 
125 static int ip6_finish_output(struct sk_buff *skb)
126 {
127 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 	    dst_allfrag(skb_dst(skb)) ||
129 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 		return ip6_fragment(skb, ip6_finish_output2);
131 	else
132 		return ip6_finish_output2(skb);
133 }
134 
135 int ip6_output(struct sock *sk, struct sk_buff *skb)
136 {
137 	struct net_device *dev = skb_dst(skb)->dev;
138 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 	if (unlikely(idev->cnf.disable_ipv6)) {
140 		IP6_INC_STATS(dev_net(dev), idev,
141 			      IPSTATS_MIB_OUTDISCARDS);
142 		kfree_skb(skb);
143 		return 0;
144 	}
145 
146 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
147 			    ip6_finish_output,
148 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
149 }
150 
151 /*
152  *	xmit an sk_buff (used by TCP, SCTP and DCCP)
153  */
154 
155 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
156 	     struct ipv6_txoptions *opt, int tclass)
157 {
158 	struct net *net = sock_net(sk);
159 	struct ipv6_pinfo *np = inet6_sk(sk);
160 	struct in6_addr *first_hop = &fl6->daddr;
161 	struct dst_entry *dst = skb_dst(skb);
162 	struct ipv6hdr *hdr;
163 	u8  proto = fl6->flowi6_proto;
164 	int seg_len = skb->len;
165 	int hlimit = -1;
166 	u32 mtu;
167 
168 	if (opt) {
169 		unsigned int head_room;
170 
171 		/* First: exthdrs may take lots of space (~8K for now)
172 		   MAX_HEADER is not enough.
173 		 */
174 		head_room = opt->opt_nflen + opt->opt_flen;
175 		seg_len += head_room;
176 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
177 
178 		if (skb_headroom(skb) < head_room) {
179 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
180 			if (skb2 == NULL) {
181 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
182 					      IPSTATS_MIB_OUTDISCARDS);
183 				kfree_skb(skb);
184 				return -ENOBUFS;
185 			}
186 			consume_skb(skb);
187 			skb = skb2;
188 			skb_set_owner_w(skb, sk);
189 		}
190 		if (opt->opt_flen)
191 			ipv6_push_frag_opts(skb, opt, &proto);
192 		if (opt->opt_nflen)
193 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
194 	}
195 
196 	skb_push(skb, sizeof(struct ipv6hdr));
197 	skb_reset_network_header(skb);
198 	hdr = ipv6_hdr(skb);
199 
200 	/*
201 	 *	Fill in the IPv6 header
202 	 */
203 	if (np)
204 		hlimit = np->hop_limit;
205 	if (hlimit < 0)
206 		hlimit = ip6_dst_hoplimit(dst);
207 
208 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
209 						     np->autoflowlabel));
210 
211 	hdr->payload_len = htons(seg_len);
212 	hdr->nexthdr = proto;
213 	hdr->hop_limit = hlimit;
214 
215 	hdr->saddr = fl6->saddr;
216 	hdr->daddr = *first_hop;
217 
218 	skb->protocol = htons(ETH_P_IPV6);
219 	skb->priority = sk->sk_priority;
220 	skb->mark = sk->sk_mark;
221 
222 	mtu = dst_mtu(dst);
223 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
224 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
225 			      IPSTATS_MIB_OUT, skb->len);
226 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
227 			       dst->dev, dst_output);
228 	}
229 
230 	skb->dev = dst->dev;
231 	ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
232 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
233 	kfree_skb(skb);
234 	return -EMSGSIZE;
235 }
236 EXPORT_SYMBOL(ip6_xmit);
237 
238 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
239 {
240 	struct ip6_ra_chain *ra;
241 	struct sock *last = NULL;
242 
243 	read_lock(&ip6_ra_lock);
244 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
245 		struct sock *sk = ra->sk;
246 		if (sk && ra->sel == sel &&
247 		    (!sk->sk_bound_dev_if ||
248 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
249 			if (last) {
250 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
251 				if (skb2)
252 					rawv6_rcv(last, skb2);
253 			}
254 			last = sk;
255 		}
256 	}
257 
258 	if (last) {
259 		rawv6_rcv(last, skb);
260 		read_unlock(&ip6_ra_lock);
261 		return 1;
262 	}
263 	read_unlock(&ip6_ra_lock);
264 	return 0;
265 }
266 
267 static int ip6_forward_proxy_check(struct sk_buff *skb)
268 {
269 	struct ipv6hdr *hdr = ipv6_hdr(skb);
270 	u8 nexthdr = hdr->nexthdr;
271 	__be16 frag_off;
272 	int offset;
273 
274 	if (ipv6_ext_hdr(nexthdr)) {
275 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
276 		if (offset < 0)
277 			return 0;
278 	} else
279 		offset = sizeof(struct ipv6hdr);
280 
281 	if (nexthdr == IPPROTO_ICMPV6) {
282 		struct icmp6hdr *icmp6;
283 
284 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
285 					 offset + 1 - skb->data)))
286 			return 0;
287 
288 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
289 
290 		switch (icmp6->icmp6_type) {
291 		case NDISC_ROUTER_SOLICITATION:
292 		case NDISC_ROUTER_ADVERTISEMENT:
293 		case NDISC_NEIGHBOUR_SOLICITATION:
294 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
295 		case NDISC_REDIRECT:
296 			/* For reaction involving unicast neighbor discovery
297 			 * message destined to the proxied address, pass it to
298 			 * input function.
299 			 */
300 			return 1;
301 		default:
302 			break;
303 		}
304 	}
305 
306 	/*
307 	 * The proxying router can't forward traffic sent to a link-local
308 	 * address, so signal the sender and discard the packet. This
309 	 * behavior is clarified by the MIPv6 specification.
310 	 */
311 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
312 		dst_link_failure(skb);
313 		return -1;
314 	}
315 
316 	return 0;
317 }
318 
319 static inline int ip6_forward_finish(struct sk_buff *skb)
320 {
321 	skb_sender_cpu_clear(skb);
322 	return dst_output(skb);
323 }
324 
325 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
326 {
327 	unsigned int mtu;
328 	struct inet6_dev *idev;
329 
330 	if (dst_metric_locked(dst, RTAX_MTU)) {
331 		mtu = dst_metric_raw(dst, RTAX_MTU);
332 		if (mtu)
333 			return mtu;
334 	}
335 
336 	mtu = IPV6_MIN_MTU;
337 	rcu_read_lock();
338 	idev = __in6_dev_get(dst->dev);
339 	if (idev)
340 		mtu = idev->cnf.mtu6;
341 	rcu_read_unlock();
342 
343 	return mtu;
344 }
345 
346 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
347 {
348 	if (skb->len <= mtu)
349 		return false;
350 
351 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
352 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
353 		return true;
354 
355 	if (skb->ignore_df)
356 		return false;
357 
358 	if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
359 		return false;
360 
361 	return true;
362 }
363 
364 int ip6_forward(struct sk_buff *skb)
365 {
366 	struct dst_entry *dst = skb_dst(skb);
367 	struct ipv6hdr *hdr = ipv6_hdr(skb);
368 	struct inet6_skb_parm *opt = IP6CB(skb);
369 	struct net *net = dev_net(dst->dev);
370 	u32 mtu;
371 
372 	if (net->ipv6.devconf_all->forwarding == 0)
373 		goto error;
374 
375 	if (skb->pkt_type != PACKET_HOST)
376 		goto drop;
377 
378 	if (skb_warn_if_lro(skb))
379 		goto drop;
380 
381 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
382 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
383 				 IPSTATS_MIB_INDISCARDS);
384 		goto drop;
385 	}
386 
387 	skb_forward_csum(skb);
388 
389 	/*
390 	 *	We DO NOT make any processing on
391 	 *	RA packets, pushing them to user level AS IS
392 	 *	without ane WARRANTY that application will be able
393 	 *	to interpret them. The reason is that we
394 	 *	cannot make anything clever here.
395 	 *
396 	 *	We are not end-node, so that if packet contains
397 	 *	AH/ESP, we cannot make anything.
398 	 *	Defragmentation also would be mistake, RA packets
399 	 *	cannot be fragmented, because there is no warranty
400 	 *	that different fragments will go along one path. --ANK
401 	 */
402 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
403 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
404 			return 0;
405 	}
406 
407 	/*
408 	 *	check and decrement ttl
409 	 */
410 	if (hdr->hop_limit <= 1) {
411 		/* Force OUTPUT device used as source address */
412 		skb->dev = dst->dev;
413 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
414 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
415 				 IPSTATS_MIB_INHDRERRORS);
416 
417 		kfree_skb(skb);
418 		return -ETIMEDOUT;
419 	}
420 
421 	/* XXX: idev->cnf.proxy_ndp? */
422 	if (net->ipv6.devconf_all->proxy_ndp &&
423 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
424 		int proxied = ip6_forward_proxy_check(skb);
425 		if (proxied > 0)
426 			return ip6_input(skb);
427 		else if (proxied < 0) {
428 			IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
429 					 IPSTATS_MIB_INDISCARDS);
430 			goto drop;
431 		}
432 	}
433 
434 	if (!xfrm6_route_forward(skb)) {
435 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
436 				 IPSTATS_MIB_INDISCARDS);
437 		goto drop;
438 	}
439 	dst = skb_dst(skb);
440 
441 	/* IPv6 specs say nothing about it, but it is clear that we cannot
442 	   send redirects to source routed frames.
443 	   We don't send redirects to frames decapsulated from IPsec.
444 	 */
445 	if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
446 		struct in6_addr *target = NULL;
447 		struct inet_peer *peer;
448 		struct rt6_info *rt;
449 
450 		/*
451 		 *	incoming and outgoing devices are the same
452 		 *	send a redirect.
453 		 */
454 
455 		rt = (struct rt6_info *) dst;
456 		if (rt->rt6i_flags & RTF_GATEWAY)
457 			target = &rt->rt6i_gateway;
458 		else
459 			target = &hdr->daddr;
460 
461 		peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
462 
463 		/* Limit redirects both by destination (here)
464 		   and by source (inside ndisc_send_redirect)
465 		 */
466 		if (inet_peer_xrlim_allow(peer, 1*HZ))
467 			ndisc_send_redirect(skb, target);
468 		if (peer)
469 			inet_putpeer(peer);
470 	} else {
471 		int addrtype = ipv6_addr_type(&hdr->saddr);
472 
473 		/* This check is security critical. */
474 		if (addrtype == IPV6_ADDR_ANY ||
475 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
476 			goto error;
477 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
478 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
479 				    ICMPV6_NOT_NEIGHBOUR, 0);
480 			goto error;
481 		}
482 	}
483 
484 	mtu = ip6_dst_mtu_forward(dst);
485 	if (mtu < IPV6_MIN_MTU)
486 		mtu = IPV6_MIN_MTU;
487 
488 	if (ip6_pkt_too_big(skb, mtu)) {
489 		/* Again, force OUTPUT device used as source address */
490 		skb->dev = dst->dev;
491 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
492 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
493 				 IPSTATS_MIB_INTOOBIGERRORS);
494 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
495 				 IPSTATS_MIB_FRAGFAILS);
496 		kfree_skb(skb);
497 		return -EMSGSIZE;
498 	}
499 
500 	if (skb_cow(skb, dst->dev->hard_header_len)) {
501 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
502 				 IPSTATS_MIB_OUTDISCARDS);
503 		goto drop;
504 	}
505 
506 	hdr = ipv6_hdr(skb);
507 
508 	/* Mangling hops number delayed to point after skb COW */
509 
510 	hdr->hop_limit--;
511 
512 	IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
513 	IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
514 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
515 		       ip6_forward_finish);
516 
517 error:
518 	IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
519 drop:
520 	kfree_skb(skb);
521 	return -EINVAL;
522 }
523 
524 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
525 {
526 	to->pkt_type = from->pkt_type;
527 	to->priority = from->priority;
528 	to->protocol = from->protocol;
529 	skb_dst_drop(to);
530 	skb_dst_set(to, dst_clone(skb_dst(from)));
531 	to->dev = from->dev;
532 	to->mark = from->mark;
533 
534 #ifdef CONFIG_NET_SCHED
535 	to->tc_index = from->tc_index;
536 #endif
537 	nf_copy(to, from);
538 	skb_copy_secmark(to, from);
539 }
540 
541 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
542 {
543 	struct sk_buff *frag;
544 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
545 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
546 				inet6_sk(skb->sk) : NULL;
547 	struct ipv6hdr *tmp_hdr;
548 	struct frag_hdr *fh;
549 	unsigned int mtu, hlen, left, len;
550 	int hroom, troom;
551 	__be32 frag_id = 0;
552 	int ptr, offset = 0, err = 0;
553 	u8 *prevhdr, nexthdr = 0;
554 	struct net *net = dev_net(skb_dst(skb)->dev);
555 
556 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
557 	nexthdr = *prevhdr;
558 
559 	mtu = ip6_skb_dst_mtu(skb);
560 
561 	/* We must not fragment if the socket is set to force MTU discovery
562 	 * or if the skb it not generated by a local socket.
563 	 */
564 	if (unlikely(!skb->ignore_df && skb->len > mtu) ||
565 		     (IP6CB(skb)->frag_max_size &&
566 		      IP6CB(skb)->frag_max_size > mtu)) {
567 		if (skb->sk && dst_allfrag(skb_dst(skb)))
568 			sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
569 
570 		skb->dev = skb_dst(skb)->dev;
571 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
572 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
573 			      IPSTATS_MIB_FRAGFAILS);
574 		kfree_skb(skb);
575 		return -EMSGSIZE;
576 	}
577 
578 	if (np && np->frag_size < mtu) {
579 		if (np->frag_size)
580 			mtu = np->frag_size;
581 	}
582 	mtu -= hlen + sizeof(struct frag_hdr);
583 
584 	if (skb_has_frag_list(skb)) {
585 		int first_len = skb_pagelen(skb);
586 		struct sk_buff *frag2;
587 
588 		if (first_len - hlen > mtu ||
589 		    ((first_len - hlen) & 7) ||
590 		    skb_cloned(skb))
591 			goto slow_path;
592 
593 		skb_walk_frags(skb, frag) {
594 			/* Correct geometry. */
595 			if (frag->len > mtu ||
596 			    ((frag->len & 7) && frag->next) ||
597 			    skb_headroom(frag) < hlen)
598 				goto slow_path_clean;
599 
600 			/* Partially cloned skb? */
601 			if (skb_shared(frag))
602 				goto slow_path_clean;
603 
604 			BUG_ON(frag->sk);
605 			if (skb->sk) {
606 				frag->sk = skb->sk;
607 				frag->destructor = sock_wfree;
608 			}
609 			skb->truesize -= frag->truesize;
610 		}
611 
612 		err = 0;
613 		offset = 0;
614 		frag = skb_shinfo(skb)->frag_list;
615 		skb_frag_list_init(skb);
616 		/* BUILD HEADER */
617 
618 		*prevhdr = NEXTHDR_FRAGMENT;
619 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
620 		if (!tmp_hdr) {
621 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
622 				      IPSTATS_MIB_FRAGFAILS);
623 			return -ENOMEM;
624 		}
625 
626 		__skb_pull(skb, hlen);
627 		fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
628 		__skb_push(skb, hlen);
629 		skb_reset_network_header(skb);
630 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
631 
632 		ipv6_select_ident(fh, rt);
633 		fh->nexthdr = nexthdr;
634 		fh->reserved = 0;
635 		fh->frag_off = htons(IP6_MF);
636 		frag_id = fh->identification;
637 
638 		first_len = skb_pagelen(skb);
639 		skb->data_len = first_len - skb_headlen(skb);
640 		skb->len = first_len;
641 		ipv6_hdr(skb)->payload_len = htons(first_len -
642 						   sizeof(struct ipv6hdr));
643 
644 		dst_hold(&rt->dst);
645 
646 		for (;;) {
647 			/* Prepare header of the next frame,
648 			 * before previous one went down. */
649 			if (frag) {
650 				frag->ip_summed = CHECKSUM_NONE;
651 				skb_reset_transport_header(frag);
652 				fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
653 				__skb_push(frag, hlen);
654 				skb_reset_network_header(frag);
655 				memcpy(skb_network_header(frag), tmp_hdr,
656 				       hlen);
657 				offset += skb->len - hlen - sizeof(struct frag_hdr);
658 				fh->nexthdr = nexthdr;
659 				fh->reserved = 0;
660 				fh->frag_off = htons(offset);
661 				if (frag->next != NULL)
662 					fh->frag_off |= htons(IP6_MF);
663 				fh->identification = frag_id;
664 				ipv6_hdr(frag)->payload_len =
665 						htons(frag->len -
666 						      sizeof(struct ipv6hdr));
667 				ip6_copy_metadata(frag, skb);
668 			}
669 
670 			err = output(skb);
671 			if (!err)
672 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
673 					      IPSTATS_MIB_FRAGCREATES);
674 
675 			if (err || !frag)
676 				break;
677 
678 			skb = frag;
679 			frag = skb->next;
680 			skb->next = NULL;
681 		}
682 
683 		kfree(tmp_hdr);
684 
685 		if (err == 0) {
686 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
687 				      IPSTATS_MIB_FRAGOKS);
688 			ip6_rt_put(rt);
689 			return 0;
690 		}
691 
692 		kfree_skb_list(frag);
693 
694 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
695 			      IPSTATS_MIB_FRAGFAILS);
696 		ip6_rt_put(rt);
697 		return err;
698 
699 slow_path_clean:
700 		skb_walk_frags(skb, frag2) {
701 			if (frag2 == frag)
702 				break;
703 			frag2->sk = NULL;
704 			frag2->destructor = NULL;
705 			skb->truesize += frag2->truesize;
706 		}
707 	}
708 
709 slow_path:
710 	if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
711 	    skb_checksum_help(skb))
712 		goto fail;
713 
714 	left = skb->len - hlen;		/* Space per frame */
715 	ptr = hlen;			/* Where to start from */
716 
717 	/*
718 	 *	Fragment the datagram.
719 	 */
720 
721 	*prevhdr = NEXTHDR_FRAGMENT;
722 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
723 	troom = rt->dst.dev->needed_tailroom;
724 
725 	/*
726 	 *	Keep copying data until we run out.
727 	 */
728 	while (left > 0)	{
729 		len = left;
730 		/* IF: it doesn't fit, use 'mtu' - the data space left */
731 		if (len > mtu)
732 			len = mtu;
733 		/* IF: we are not sending up to and including the packet end
734 		   then align the next start on an eight byte boundary */
735 		if (len < left)	{
736 			len &= ~7;
737 		}
738 
739 		/* Allocate buffer */
740 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
741 				 hroom + troom, GFP_ATOMIC);
742 		if (!frag) {
743 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
744 				      IPSTATS_MIB_FRAGFAILS);
745 			err = -ENOMEM;
746 			goto fail;
747 		}
748 
749 		/*
750 		 *	Set up data on packet
751 		 */
752 
753 		ip6_copy_metadata(frag, skb);
754 		skb_reserve(frag, hroom);
755 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
756 		skb_reset_network_header(frag);
757 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
758 		frag->transport_header = (frag->network_header + hlen +
759 					  sizeof(struct frag_hdr));
760 
761 		/*
762 		 *	Charge the memory for the fragment to any owner
763 		 *	it might possess
764 		 */
765 		if (skb->sk)
766 			skb_set_owner_w(frag, skb->sk);
767 
768 		/*
769 		 *	Copy the packet header into the new buffer.
770 		 */
771 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
772 
773 		/*
774 		 *	Build fragment header.
775 		 */
776 		fh->nexthdr = nexthdr;
777 		fh->reserved = 0;
778 		if (!frag_id) {
779 			ipv6_select_ident(fh, rt);
780 			frag_id = fh->identification;
781 		} else
782 			fh->identification = frag_id;
783 
784 		/*
785 		 *	Copy a block of the IP datagram.
786 		 */
787 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
788 				     len));
789 		left -= len;
790 
791 		fh->frag_off = htons(offset);
792 		if (left > 0)
793 			fh->frag_off |= htons(IP6_MF);
794 		ipv6_hdr(frag)->payload_len = htons(frag->len -
795 						    sizeof(struct ipv6hdr));
796 
797 		ptr += len;
798 		offset += len;
799 
800 		/*
801 		 *	Put this fragment into the sending queue.
802 		 */
803 		err = output(frag);
804 		if (err)
805 			goto fail;
806 
807 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
808 			      IPSTATS_MIB_FRAGCREATES);
809 	}
810 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
811 		      IPSTATS_MIB_FRAGOKS);
812 	consume_skb(skb);
813 	return err;
814 
815 fail:
816 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
817 		      IPSTATS_MIB_FRAGFAILS);
818 	kfree_skb(skb);
819 	return err;
820 }
821 
822 static inline int ip6_rt_check(const struct rt6key *rt_key,
823 			       const struct in6_addr *fl_addr,
824 			       const struct in6_addr *addr_cache)
825 {
826 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
827 		(addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
828 }
829 
830 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
831 					  struct dst_entry *dst,
832 					  const struct flowi6 *fl6)
833 {
834 	struct ipv6_pinfo *np = inet6_sk(sk);
835 	struct rt6_info *rt;
836 
837 	if (!dst)
838 		goto out;
839 
840 	if (dst->ops->family != AF_INET6) {
841 		dst_release(dst);
842 		return NULL;
843 	}
844 
845 	rt = (struct rt6_info *)dst;
846 	/* Yes, checking route validity in not connected
847 	 * case is not very simple. Take into account,
848 	 * that we do not support routing by source, TOS,
849 	 * and MSG_DONTROUTE		--ANK (980726)
850 	 *
851 	 * 1. ip6_rt_check(): If route was host route,
852 	 *    check that cached destination is current.
853 	 *    If it is network route, we still may
854 	 *    check its validity using saved pointer
855 	 *    to the last used address: daddr_cache.
856 	 *    We do not want to save whole address now,
857 	 *    (because main consumer of this service
858 	 *    is tcp, which has not this problem),
859 	 *    so that the last trick works only on connected
860 	 *    sockets.
861 	 * 2. oif also should be the same.
862 	 */
863 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
864 #ifdef CONFIG_IPV6_SUBTREES
865 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
866 #endif
867 	    (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
868 		dst_release(dst);
869 		dst = NULL;
870 	}
871 
872 out:
873 	return dst;
874 }
875 
876 static int ip6_dst_lookup_tail(struct sock *sk,
877 			       struct dst_entry **dst, struct flowi6 *fl6)
878 {
879 	struct net *net = sock_net(sk);
880 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
881 	struct neighbour *n;
882 	struct rt6_info *rt;
883 #endif
884 	int err;
885 
886 	if (*dst == NULL)
887 		*dst = ip6_route_output(net, sk, fl6);
888 
889 	err = (*dst)->error;
890 	if (err)
891 		goto out_err_release;
892 
893 	if (ipv6_addr_any(&fl6->saddr)) {
894 		struct rt6_info *rt = (struct rt6_info *) *dst;
895 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
896 					  sk ? inet6_sk(sk)->srcprefs : 0,
897 					  &fl6->saddr);
898 		if (err)
899 			goto out_err_release;
900 	}
901 
902 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
903 	/*
904 	 * Here if the dst entry we've looked up
905 	 * has a neighbour entry that is in the INCOMPLETE
906 	 * state and the src address from the flow is
907 	 * marked as OPTIMISTIC, we release the found
908 	 * dst entry and replace it instead with the
909 	 * dst entry of the nexthop router
910 	 */
911 	rt = (struct rt6_info *) *dst;
912 	rcu_read_lock_bh();
913 	n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
914 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
915 	rcu_read_unlock_bh();
916 
917 	if (err) {
918 		struct inet6_ifaddr *ifp;
919 		struct flowi6 fl_gw6;
920 		int redirect;
921 
922 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
923 				      (*dst)->dev, 1);
924 
925 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
926 		if (ifp)
927 			in6_ifa_put(ifp);
928 
929 		if (redirect) {
930 			/*
931 			 * We need to get the dst entry for the
932 			 * default router instead
933 			 */
934 			dst_release(*dst);
935 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
936 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
937 			*dst = ip6_route_output(net, sk, &fl_gw6);
938 			err = (*dst)->error;
939 			if (err)
940 				goto out_err_release;
941 		}
942 	}
943 #endif
944 
945 	return 0;
946 
947 out_err_release:
948 	if (err == -ENETUNREACH)
949 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
950 	dst_release(*dst);
951 	*dst = NULL;
952 	return err;
953 }
954 
955 /**
956  *	ip6_dst_lookup - perform route lookup on flow
957  *	@sk: socket which provides route info
958  *	@dst: pointer to dst_entry * for result
959  *	@fl6: flow to lookup
960  *
961  *	This function performs a route lookup on the given flow.
962  *
963  *	It returns zero on success, or a standard errno code on error.
964  */
965 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
966 {
967 	*dst = NULL;
968 	return ip6_dst_lookup_tail(sk, dst, fl6);
969 }
970 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
971 
972 /**
973  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
974  *	@sk: socket which provides route info
975  *	@fl6: flow to lookup
976  *	@final_dst: final destination address for ipsec lookup
977  *
978  *	This function performs a route lookup on the given flow.
979  *
980  *	It returns a valid dst pointer on success, or a pointer encoded
981  *	error code.
982  */
983 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
984 				      const struct in6_addr *final_dst)
985 {
986 	struct dst_entry *dst = NULL;
987 	int err;
988 
989 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
990 	if (err)
991 		return ERR_PTR(err);
992 	if (final_dst)
993 		fl6->daddr = *final_dst;
994 
995 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
996 }
997 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
998 
999 /**
1000  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1001  *	@sk: socket which provides the dst cache and route info
1002  *	@fl6: flow to lookup
1003  *	@final_dst: final destination address for ipsec lookup
1004  *
1005  *	This function performs a route lookup on the given flow with the
1006  *	possibility of using the cached route in the socket if it is valid.
1007  *	It will take the socket dst lock when operating on the dst cache.
1008  *	As a result, this function can only be used in process context.
1009  *
1010  *	It returns a valid dst pointer on success, or a pointer encoded
1011  *	error code.
1012  */
1013 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1014 					 const struct in6_addr *final_dst)
1015 {
1016 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1017 	int err;
1018 
1019 	dst = ip6_sk_dst_check(sk, dst, fl6);
1020 
1021 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
1022 	if (err)
1023 		return ERR_PTR(err);
1024 	if (final_dst)
1025 		fl6->daddr = *final_dst;
1026 
1027 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1028 }
1029 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1030 
1031 static inline int ip6_ufo_append_data(struct sock *sk,
1032 			struct sk_buff_head *queue,
1033 			int getfrag(void *from, char *to, int offset, int len,
1034 			int odd, struct sk_buff *skb),
1035 			void *from, int length, int hh_len, int fragheaderlen,
1036 			int transhdrlen, int mtu, unsigned int flags,
1037 			struct rt6_info *rt)
1038 
1039 {
1040 	struct sk_buff *skb;
1041 	struct frag_hdr fhdr;
1042 	int err;
1043 
1044 	/* There is support for UDP large send offload by network
1045 	 * device, so create one single skb packet containing complete
1046 	 * udp datagram
1047 	 */
1048 	skb = skb_peek_tail(queue);
1049 	if (skb == NULL) {
1050 		skb = sock_alloc_send_skb(sk,
1051 			hh_len + fragheaderlen + transhdrlen + 20,
1052 			(flags & MSG_DONTWAIT), &err);
1053 		if (skb == NULL)
1054 			return err;
1055 
1056 		/* reserve space for Hardware header */
1057 		skb_reserve(skb, hh_len);
1058 
1059 		/* create space for UDP/IP header */
1060 		skb_put(skb, fragheaderlen + transhdrlen);
1061 
1062 		/* initialize network header pointer */
1063 		skb_reset_network_header(skb);
1064 
1065 		/* initialize protocol header pointer */
1066 		skb->transport_header = skb->network_header + fragheaderlen;
1067 
1068 		skb->protocol = htons(ETH_P_IPV6);
1069 		skb->csum = 0;
1070 
1071 		__skb_queue_tail(queue, skb);
1072 	} else if (skb_is_gso(skb)) {
1073 		goto append;
1074 	}
1075 
1076 	skb->ip_summed = CHECKSUM_PARTIAL;
1077 	/* Specify the length of each IPv6 datagram fragment.
1078 	 * It has to be a multiple of 8.
1079 	 */
1080 	skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1081 				     sizeof(struct frag_hdr)) & ~7;
1082 	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1083 	ipv6_select_ident(&fhdr, rt);
1084 	skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1085 
1086 append:
1087 	return skb_append_datato_frags(sk, skb, getfrag, from,
1088 				       (length - transhdrlen));
1089 }
1090 
1091 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1092 					       gfp_t gfp)
1093 {
1094 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1095 }
1096 
1097 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1098 						gfp_t gfp)
1099 {
1100 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1101 }
1102 
1103 static void ip6_append_data_mtu(unsigned int *mtu,
1104 				int *maxfraglen,
1105 				unsigned int fragheaderlen,
1106 				struct sk_buff *skb,
1107 				struct rt6_info *rt,
1108 				unsigned int orig_mtu)
1109 {
1110 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1111 		if (skb == NULL) {
1112 			/* first fragment, reserve header_len */
1113 			*mtu = orig_mtu - rt->dst.header_len;
1114 
1115 		} else {
1116 			/*
1117 			 * this fragment is not first, the headers
1118 			 * space is regarded as data space.
1119 			 */
1120 			*mtu = orig_mtu;
1121 		}
1122 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1123 			      + fragheaderlen - sizeof(struct frag_hdr);
1124 	}
1125 }
1126 
1127 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1128 			  struct inet6_cork *v6_cork,
1129 			  int hlimit, int tclass, struct ipv6_txoptions *opt,
1130 			  struct rt6_info *rt, struct flowi6 *fl6)
1131 {
1132 	struct ipv6_pinfo *np = inet6_sk(sk);
1133 	unsigned int mtu;
1134 
1135 	/*
1136 	 * setup for corking
1137 	 */
1138 	if (opt) {
1139 		if (WARN_ON(v6_cork->opt))
1140 			return -EINVAL;
1141 
1142 		v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1143 		if (unlikely(v6_cork->opt == NULL))
1144 			return -ENOBUFS;
1145 
1146 		v6_cork->opt->tot_len = opt->tot_len;
1147 		v6_cork->opt->opt_flen = opt->opt_flen;
1148 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1149 
1150 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1151 						    sk->sk_allocation);
1152 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1153 			return -ENOBUFS;
1154 
1155 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1156 						    sk->sk_allocation);
1157 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1158 			return -ENOBUFS;
1159 
1160 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1161 						   sk->sk_allocation);
1162 		if (opt->hopopt && !v6_cork->opt->hopopt)
1163 			return -ENOBUFS;
1164 
1165 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1166 						    sk->sk_allocation);
1167 		if (opt->srcrt && !v6_cork->opt->srcrt)
1168 			return -ENOBUFS;
1169 
1170 		/* need source address above miyazawa*/
1171 	}
1172 	dst_hold(&rt->dst);
1173 	cork->base.dst = &rt->dst;
1174 	cork->fl.u.ip6 = *fl6;
1175 	v6_cork->hop_limit = hlimit;
1176 	v6_cork->tclass = tclass;
1177 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1178 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1179 		      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1180 	else
1181 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1182 		      rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1183 	if (np->frag_size < mtu) {
1184 		if (np->frag_size)
1185 			mtu = np->frag_size;
1186 	}
1187 	cork->base.fragsize = mtu;
1188 	if (dst_allfrag(rt->dst.path))
1189 		cork->base.flags |= IPCORK_ALLFRAG;
1190 	cork->base.length = 0;
1191 
1192 	return 0;
1193 }
1194 
1195 static int __ip6_append_data(struct sock *sk,
1196 			     struct flowi6 *fl6,
1197 			     struct sk_buff_head *queue,
1198 			     struct inet_cork *cork,
1199 			     struct inet6_cork *v6_cork,
1200 			     struct page_frag *pfrag,
1201 			     int getfrag(void *from, char *to, int offset,
1202 					 int len, int odd, struct sk_buff *skb),
1203 			     void *from, int length, int transhdrlen,
1204 			     unsigned int flags, int dontfrag)
1205 {
1206 	struct sk_buff *skb, *skb_prev = NULL;
1207 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1208 	int exthdrlen = 0;
1209 	int dst_exthdrlen = 0;
1210 	int hh_len;
1211 	int copy;
1212 	int err;
1213 	int offset = 0;
1214 	__u8 tx_flags = 0;
1215 	u32 tskey = 0;
1216 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1217 	struct ipv6_txoptions *opt = v6_cork->opt;
1218 	int csummode = CHECKSUM_NONE;
1219 
1220 	skb = skb_peek_tail(queue);
1221 	if (!skb) {
1222 		exthdrlen = opt ? opt->opt_flen : 0;
1223 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1224 	}
1225 
1226 	mtu = cork->fragsize;
1227 	orig_mtu = mtu;
1228 
1229 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1230 
1231 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1232 			(opt ? opt->opt_nflen : 0);
1233 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1234 		     sizeof(struct frag_hdr);
1235 
1236 	if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1237 		unsigned int maxnonfragsize, headersize;
1238 
1239 		headersize = sizeof(struct ipv6hdr) +
1240 			     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1241 			     (dst_allfrag(&rt->dst) ?
1242 			      sizeof(struct frag_hdr) : 0) +
1243 			     rt->rt6i_nfheader_len;
1244 
1245 		if (ip6_sk_ignore_df(sk))
1246 			maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1247 		else
1248 			maxnonfragsize = mtu;
1249 
1250 		/* dontfrag active */
1251 		if ((cork->length + length > mtu - headersize) && dontfrag &&
1252 		    (sk->sk_protocol == IPPROTO_UDP ||
1253 		     sk->sk_protocol == IPPROTO_RAW)) {
1254 			ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1255 						   sizeof(struct ipv6hdr));
1256 			goto emsgsize;
1257 		}
1258 
1259 		if (cork->length + length > maxnonfragsize - headersize) {
1260 emsgsize:
1261 			ipv6_local_error(sk, EMSGSIZE, fl6,
1262 					 mtu - headersize +
1263 					 sizeof(struct ipv6hdr));
1264 			return -EMSGSIZE;
1265 		}
1266 	}
1267 
1268 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1269 		sock_tx_timestamp(sk, &tx_flags);
1270 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1271 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1272 			tskey = sk->sk_tskey++;
1273 	}
1274 
1275 	/* If this is the first and only packet and device
1276 	 * supports checksum offloading, let's use it.
1277 	 */
1278 	if (!skb && sk->sk_protocol == IPPROTO_UDP &&
1279 	    length + fragheaderlen < mtu &&
1280 	    rt->dst.dev->features & NETIF_F_V6_CSUM &&
1281 	    !exthdrlen)
1282 		csummode = CHECKSUM_PARTIAL;
1283 	/*
1284 	 * Let's try using as much space as possible.
1285 	 * Use MTU if total length of the message fits into the MTU.
1286 	 * Otherwise, we need to reserve fragment header and
1287 	 * fragment alignment (= 8-15 octects, in total).
1288 	 *
1289 	 * Note that we may need to "move" the data from the tail of
1290 	 * of the buffer to the new fragment when we split
1291 	 * the message.
1292 	 *
1293 	 * FIXME: It may be fragmented into multiple chunks
1294 	 *        at once if non-fragmentable extension headers
1295 	 *        are too large.
1296 	 * --yoshfuji
1297 	 */
1298 
1299 	cork->length += length;
1300 	if (((length > mtu) ||
1301 	     (skb && skb_is_gso(skb))) &&
1302 	    (sk->sk_protocol == IPPROTO_UDP) &&
1303 	    (rt->dst.dev->features & NETIF_F_UFO) &&
1304 	    (sk->sk_type == SOCK_DGRAM)) {
1305 		err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1306 					  hh_len, fragheaderlen,
1307 					  transhdrlen, mtu, flags, rt);
1308 		if (err)
1309 			goto error;
1310 		return 0;
1311 	}
1312 
1313 	if (!skb)
1314 		goto alloc_new_skb;
1315 
1316 	while (length > 0) {
1317 		/* Check if the remaining data fits into current packet. */
1318 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1319 		if (copy < length)
1320 			copy = maxfraglen - skb->len;
1321 
1322 		if (copy <= 0) {
1323 			char *data;
1324 			unsigned int datalen;
1325 			unsigned int fraglen;
1326 			unsigned int fraggap;
1327 			unsigned int alloclen;
1328 alloc_new_skb:
1329 			/* There's no room in the current skb */
1330 			if (skb)
1331 				fraggap = skb->len - maxfraglen;
1332 			else
1333 				fraggap = 0;
1334 			/* update mtu and maxfraglen if necessary */
1335 			if (skb == NULL || skb_prev == NULL)
1336 				ip6_append_data_mtu(&mtu, &maxfraglen,
1337 						    fragheaderlen, skb, rt,
1338 						    orig_mtu);
1339 
1340 			skb_prev = skb;
1341 
1342 			/*
1343 			 * If remaining data exceeds the mtu,
1344 			 * we know we need more fragment(s).
1345 			 */
1346 			datalen = length + fraggap;
1347 
1348 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1349 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1350 			if ((flags & MSG_MORE) &&
1351 			    !(rt->dst.dev->features&NETIF_F_SG))
1352 				alloclen = mtu;
1353 			else
1354 				alloclen = datalen + fragheaderlen;
1355 
1356 			alloclen += dst_exthdrlen;
1357 
1358 			if (datalen != length + fraggap) {
1359 				/*
1360 				 * this is not the last fragment, the trailer
1361 				 * space is regarded as data space.
1362 				 */
1363 				datalen += rt->dst.trailer_len;
1364 			}
1365 
1366 			alloclen += rt->dst.trailer_len;
1367 			fraglen = datalen + fragheaderlen;
1368 
1369 			/*
1370 			 * We just reserve space for fragment header.
1371 			 * Note: this may be overallocation if the message
1372 			 * (without MSG_MORE) fits into the MTU.
1373 			 */
1374 			alloclen += sizeof(struct frag_hdr);
1375 
1376 			if (transhdrlen) {
1377 				skb = sock_alloc_send_skb(sk,
1378 						alloclen + hh_len,
1379 						(flags & MSG_DONTWAIT), &err);
1380 			} else {
1381 				skb = NULL;
1382 				if (atomic_read(&sk->sk_wmem_alloc) <=
1383 				    2 * sk->sk_sndbuf)
1384 					skb = sock_wmalloc(sk,
1385 							   alloclen + hh_len, 1,
1386 							   sk->sk_allocation);
1387 				if (unlikely(skb == NULL))
1388 					err = -ENOBUFS;
1389 			}
1390 			if (skb == NULL)
1391 				goto error;
1392 			/*
1393 			 *	Fill in the control structures
1394 			 */
1395 			skb->protocol = htons(ETH_P_IPV6);
1396 			skb->ip_summed = csummode;
1397 			skb->csum = 0;
1398 			/* reserve for fragmentation and ipsec header */
1399 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1400 				    dst_exthdrlen);
1401 
1402 			/* Only the initial fragment is time stamped */
1403 			skb_shinfo(skb)->tx_flags = tx_flags;
1404 			tx_flags = 0;
1405 			skb_shinfo(skb)->tskey = tskey;
1406 			tskey = 0;
1407 
1408 			/*
1409 			 *	Find where to start putting bytes
1410 			 */
1411 			data = skb_put(skb, fraglen);
1412 			skb_set_network_header(skb, exthdrlen);
1413 			data += fragheaderlen;
1414 			skb->transport_header = (skb->network_header +
1415 						 fragheaderlen);
1416 			if (fraggap) {
1417 				skb->csum = skb_copy_and_csum_bits(
1418 					skb_prev, maxfraglen,
1419 					data + transhdrlen, fraggap, 0);
1420 				skb_prev->csum = csum_sub(skb_prev->csum,
1421 							  skb->csum);
1422 				data += fraggap;
1423 				pskb_trim_unique(skb_prev, maxfraglen);
1424 			}
1425 			copy = datalen - transhdrlen - fraggap;
1426 
1427 			if (copy < 0) {
1428 				err = -EINVAL;
1429 				kfree_skb(skb);
1430 				goto error;
1431 			} else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1432 				err = -EFAULT;
1433 				kfree_skb(skb);
1434 				goto error;
1435 			}
1436 
1437 			offset += copy;
1438 			length -= datalen - fraggap;
1439 			transhdrlen = 0;
1440 			exthdrlen = 0;
1441 			dst_exthdrlen = 0;
1442 
1443 			/*
1444 			 * Put the packet on the pending queue
1445 			 */
1446 			__skb_queue_tail(queue, skb);
1447 			continue;
1448 		}
1449 
1450 		if (copy > length)
1451 			copy = length;
1452 
1453 		if (!(rt->dst.dev->features&NETIF_F_SG)) {
1454 			unsigned int off;
1455 
1456 			off = skb->len;
1457 			if (getfrag(from, skb_put(skb, copy),
1458 						offset, copy, off, skb) < 0) {
1459 				__skb_trim(skb, off);
1460 				err = -EFAULT;
1461 				goto error;
1462 			}
1463 		} else {
1464 			int i = skb_shinfo(skb)->nr_frags;
1465 
1466 			err = -ENOMEM;
1467 			if (!sk_page_frag_refill(sk, pfrag))
1468 				goto error;
1469 
1470 			if (!skb_can_coalesce(skb, i, pfrag->page,
1471 					      pfrag->offset)) {
1472 				err = -EMSGSIZE;
1473 				if (i == MAX_SKB_FRAGS)
1474 					goto error;
1475 
1476 				__skb_fill_page_desc(skb, i, pfrag->page,
1477 						     pfrag->offset, 0);
1478 				skb_shinfo(skb)->nr_frags = ++i;
1479 				get_page(pfrag->page);
1480 			}
1481 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1482 			if (getfrag(from,
1483 				    page_address(pfrag->page) + pfrag->offset,
1484 				    offset, copy, skb->len, skb) < 0)
1485 				goto error_efault;
1486 
1487 			pfrag->offset += copy;
1488 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1489 			skb->len += copy;
1490 			skb->data_len += copy;
1491 			skb->truesize += copy;
1492 			atomic_add(copy, &sk->sk_wmem_alloc);
1493 		}
1494 		offset += copy;
1495 		length -= copy;
1496 	}
1497 
1498 	return 0;
1499 
1500 error_efault:
1501 	err = -EFAULT;
1502 error:
1503 	cork->length -= length;
1504 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1505 	return err;
1506 }
1507 
1508 int ip6_append_data(struct sock *sk,
1509 		    int getfrag(void *from, char *to, int offset, int len,
1510 				int odd, struct sk_buff *skb),
1511 		    void *from, int length, int transhdrlen, int hlimit,
1512 		    int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1513 		    struct rt6_info *rt, unsigned int flags, int dontfrag)
1514 {
1515 	struct inet_sock *inet = inet_sk(sk);
1516 	struct ipv6_pinfo *np = inet6_sk(sk);
1517 	int exthdrlen;
1518 	int err;
1519 
1520 	if (flags&MSG_PROBE)
1521 		return 0;
1522 	if (skb_queue_empty(&sk->sk_write_queue)) {
1523 		/*
1524 		 * setup for corking
1525 		 */
1526 		err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1527 				     tclass, opt, rt, fl6);
1528 		if (err)
1529 			return err;
1530 
1531 		exthdrlen = (opt ? opt->opt_flen : 0);
1532 		length += exthdrlen;
1533 		transhdrlen += exthdrlen;
1534 	} else {
1535 		fl6 = &inet->cork.fl.u.ip6;
1536 		transhdrlen = 0;
1537 	}
1538 
1539 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1540 				 &np->cork, sk_page_frag(sk), getfrag,
1541 				 from, length, transhdrlen, flags, dontfrag);
1542 }
1543 EXPORT_SYMBOL_GPL(ip6_append_data);
1544 
1545 static void ip6_cork_release(struct inet_cork_full *cork,
1546 			     struct inet6_cork *v6_cork)
1547 {
1548 	if (v6_cork->opt) {
1549 		kfree(v6_cork->opt->dst0opt);
1550 		kfree(v6_cork->opt->dst1opt);
1551 		kfree(v6_cork->opt->hopopt);
1552 		kfree(v6_cork->opt->srcrt);
1553 		kfree(v6_cork->opt);
1554 		v6_cork->opt = NULL;
1555 	}
1556 
1557 	if (cork->base.dst) {
1558 		dst_release(cork->base.dst);
1559 		cork->base.dst = NULL;
1560 		cork->base.flags &= ~IPCORK_ALLFRAG;
1561 	}
1562 	memset(&cork->fl, 0, sizeof(cork->fl));
1563 }
1564 
1565 struct sk_buff *__ip6_make_skb(struct sock *sk,
1566 			       struct sk_buff_head *queue,
1567 			       struct inet_cork_full *cork,
1568 			       struct inet6_cork *v6_cork)
1569 {
1570 	struct sk_buff *skb, *tmp_skb;
1571 	struct sk_buff **tail_skb;
1572 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1573 	struct ipv6_pinfo *np = inet6_sk(sk);
1574 	struct net *net = sock_net(sk);
1575 	struct ipv6hdr *hdr;
1576 	struct ipv6_txoptions *opt = v6_cork->opt;
1577 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1578 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1579 	unsigned char proto = fl6->flowi6_proto;
1580 
1581 	skb = __skb_dequeue(queue);
1582 	if (skb == NULL)
1583 		goto out;
1584 	tail_skb = &(skb_shinfo(skb)->frag_list);
1585 
1586 	/* move skb->data to ip header from ext header */
1587 	if (skb->data < skb_network_header(skb))
1588 		__skb_pull(skb, skb_network_offset(skb));
1589 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1590 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1591 		*tail_skb = tmp_skb;
1592 		tail_skb = &(tmp_skb->next);
1593 		skb->len += tmp_skb->len;
1594 		skb->data_len += tmp_skb->len;
1595 		skb->truesize += tmp_skb->truesize;
1596 		tmp_skb->destructor = NULL;
1597 		tmp_skb->sk = NULL;
1598 	}
1599 
1600 	/* Allow local fragmentation. */
1601 	skb->ignore_df = ip6_sk_ignore_df(sk);
1602 
1603 	*final_dst = fl6->daddr;
1604 	__skb_pull(skb, skb_network_header_len(skb));
1605 	if (opt && opt->opt_flen)
1606 		ipv6_push_frag_opts(skb, opt, &proto);
1607 	if (opt && opt->opt_nflen)
1608 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1609 
1610 	skb_push(skb, sizeof(struct ipv6hdr));
1611 	skb_reset_network_header(skb);
1612 	hdr = ipv6_hdr(skb);
1613 
1614 	ip6_flow_hdr(hdr, v6_cork->tclass,
1615 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1616 					np->autoflowlabel));
1617 	hdr->hop_limit = v6_cork->hop_limit;
1618 	hdr->nexthdr = proto;
1619 	hdr->saddr = fl6->saddr;
1620 	hdr->daddr = *final_dst;
1621 
1622 	skb->priority = sk->sk_priority;
1623 	skb->mark = sk->sk_mark;
1624 
1625 	skb_dst_set(skb, dst_clone(&rt->dst));
1626 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1627 	if (proto == IPPROTO_ICMPV6) {
1628 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1629 
1630 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1631 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1632 	}
1633 
1634 	ip6_cork_release(cork, v6_cork);
1635 out:
1636 	return skb;
1637 }
1638 
1639 int ip6_send_skb(struct sk_buff *skb)
1640 {
1641 	struct net *net = sock_net(skb->sk);
1642 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1643 	int err;
1644 
1645 	err = ip6_local_out(skb);
1646 	if (err) {
1647 		if (err > 0)
1648 			err = net_xmit_errno(err);
1649 		if (err)
1650 			IP6_INC_STATS(net, rt->rt6i_idev,
1651 				      IPSTATS_MIB_OUTDISCARDS);
1652 	}
1653 
1654 	return err;
1655 }
1656 
1657 int ip6_push_pending_frames(struct sock *sk)
1658 {
1659 	struct sk_buff *skb;
1660 
1661 	skb = ip6_finish_skb(sk);
1662 	if (!skb)
1663 		return 0;
1664 
1665 	return ip6_send_skb(skb);
1666 }
1667 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1668 
1669 static void __ip6_flush_pending_frames(struct sock *sk,
1670 				       struct sk_buff_head *queue,
1671 				       struct inet_cork_full *cork,
1672 				       struct inet6_cork *v6_cork)
1673 {
1674 	struct sk_buff *skb;
1675 
1676 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1677 		if (skb_dst(skb))
1678 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1679 				      IPSTATS_MIB_OUTDISCARDS);
1680 		kfree_skb(skb);
1681 	}
1682 
1683 	ip6_cork_release(cork, v6_cork);
1684 }
1685 
1686 void ip6_flush_pending_frames(struct sock *sk)
1687 {
1688 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1689 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1690 }
1691 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1692 
1693 struct sk_buff *ip6_make_skb(struct sock *sk,
1694 			     int getfrag(void *from, char *to, int offset,
1695 					 int len, int odd, struct sk_buff *skb),
1696 			     void *from, int length, int transhdrlen,
1697 			     int hlimit, int tclass,
1698 			     struct ipv6_txoptions *opt, struct flowi6 *fl6,
1699 			     struct rt6_info *rt, unsigned int flags,
1700 			     int dontfrag)
1701 {
1702 	struct inet_cork_full cork;
1703 	struct inet6_cork v6_cork;
1704 	struct sk_buff_head queue;
1705 	int exthdrlen = (opt ? opt->opt_flen : 0);
1706 	int err;
1707 
1708 	if (flags & MSG_PROBE)
1709 		return NULL;
1710 
1711 	__skb_queue_head_init(&queue);
1712 
1713 	cork.base.flags = 0;
1714 	cork.base.addr = 0;
1715 	cork.base.opt = NULL;
1716 	v6_cork.opt = NULL;
1717 	err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1718 	if (err)
1719 		return ERR_PTR(err);
1720 
1721 	if (dontfrag < 0)
1722 		dontfrag = inet6_sk(sk)->dontfrag;
1723 
1724 	err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1725 				&current->task_frag, getfrag, from,
1726 				length + exthdrlen, transhdrlen + exthdrlen,
1727 				flags, dontfrag);
1728 	if (err) {
1729 		__ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1730 		return ERR_PTR(err);
1731 	}
1732 
1733 	return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1734 }
1735