xref: /linux/net/ipv6/ip6_output.c (revision 96890d62523c2cddc2c053ad29de35c4d935cf11)
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/bpf-cgroup.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
61 
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 {
64 	struct dst_entry *dst = skb_dst(skb);
65 	struct net_device *dev = dst->dev;
66 	struct neighbour *neigh;
67 	struct in6_addr *nexthop;
68 	int ret;
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(sk) &&
74 		    ((mroute6_socket(net, 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 					net, sk, newskb, NULL, newskb->dev,
86 					dev_loopback_xmit);
87 
88 			if (ipv6_hdr(skb)->hop_limit == 0) {
89 				IP6_INC_STATS(net, idev,
90 					      IPSTATS_MIB_OUTDISCARDS);
91 				kfree_skb(skb);
92 				return 0;
93 			}
94 		}
95 
96 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
97 
98 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
99 		    IPV6_ADDR_SCOPE_NODELOCAL &&
100 		    !(dev->flags & IFF_LOOPBACK)) {
101 			kfree_skb(skb);
102 			return 0;
103 		}
104 	}
105 
106 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
107 		int res = lwtunnel_xmit(skb);
108 
109 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
110 			return res;
111 	}
112 
113 	rcu_read_lock_bh();
114 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
115 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
116 	if (unlikely(!neigh))
117 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
118 	if (!IS_ERR(neigh)) {
119 		sock_confirm_neigh(skb, neigh);
120 		ret = neigh_output(neigh, skb);
121 		rcu_read_unlock_bh();
122 		return ret;
123 	}
124 	rcu_read_unlock_bh();
125 
126 	IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
127 	kfree_skb(skb);
128 	return -EINVAL;
129 }
130 
131 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
132 {
133 	int ret;
134 
135 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
136 	if (ret) {
137 		kfree_skb(skb);
138 		return ret;
139 	}
140 
141 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
142 	/* Policy lookup after SNAT yielded a new policy */
143 	if (skb_dst(skb)->xfrm) {
144 		IPCB(skb)->flags |= IPSKB_REROUTED;
145 		return dst_output(net, sk, skb);
146 	}
147 #endif
148 
149 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
150 	    dst_allfrag(skb_dst(skb)) ||
151 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
152 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
153 	else
154 		return ip6_finish_output2(net, sk, skb);
155 }
156 
157 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
158 {
159 	struct net_device *dev = skb_dst(skb)->dev;
160 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
161 
162 	skb->protocol = htons(ETH_P_IPV6);
163 	skb->dev = dev;
164 
165 	if (unlikely(idev->cnf.disable_ipv6)) {
166 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
167 		kfree_skb(skb);
168 		return 0;
169 	}
170 
171 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
172 			    net, sk, skb, NULL, dev,
173 			    ip6_finish_output,
174 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
175 }
176 
177 static bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
178 {
179 	if (!np->autoflowlabel_set)
180 		return ip6_default_np_autolabel(net);
181 	else
182 		return np->autoflowlabel;
183 }
184 
185 /*
186  * xmit an sk_buff (used by TCP, SCTP and DCCP)
187  * Note : socket lock is not held for SYNACK packets, but might be modified
188  * by calls to skb_set_owner_w() and ipv6_local_error(),
189  * which are using proper atomic operations or spinlocks.
190  */
191 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
192 	     __u32 mark, struct ipv6_txoptions *opt, int tclass)
193 {
194 	struct net *net = sock_net(sk);
195 	const struct ipv6_pinfo *np = inet6_sk(sk);
196 	struct in6_addr *first_hop = &fl6->daddr;
197 	struct dst_entry *dst = skb_dst(skb);
198 	struct ipv6hdr *hdr;
199 	u8  proto = fl6->flowi6_proto;
200 	int seg_len = skb->len;
201 	int hlimit = -1;
202 	u32 mtu;
203 
204 	if (opt) {
205 		unsigned int head_room;
206 
207 		/* First: exthdrs may take lots of space (~8K for now)
208 		   MAX_HEADER is not enough.
209 		 */
210 		head_room = opt->opt_nflen + opt->opt_flen;
211 		seg_len += head_room;
212 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
213 
214 		if (skb_headroom(skb) < head_room) {
215 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
216 			if (!skb2) {
217 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 					      IPSTATS_MIB_OUTDISCARDS);
219 				kfree_skb(skb);
220 				return -ENOBUFS;
221 			}
222 			consume_skb(skb);
223 			skb = skb2;
224 			/* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
225 			 * it is safe to call in our context (socket lock not held)
226 			 */
227 			skb_set_owner_w(skb, (struct sock *)sk);
228 		}
229 		if (opt->opt_flen)
230 			ipv6_push_frag_opts(skb, opt, &proto);
231 		if (opt->opt_nflen)
232 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
233 					     &fl6->saddr);
234 	}
235 
236 	skb_push(skb, sizeof(struct ipv6hdr));
237 	skb_reset_network_header(skb);
238 	hdr = ipv6_hdr(skb);
239 
240 	/*
241 	 *	Fill in the IPv6 header
242 	 */
243 	if (np)
244 		hlimit = np->hop_limit;
245 	if (hlimit < 0)
246 		hlimit = ip6_dst_hoplimit(dst);
247 
248 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
249 				ip6_autoflowlabel(net, np), fl6));
250 
251 	hdr->payload_len = htons(seg_len);
252 	hdr->nexthdr = proto;
253 	hdr->hop_limit = hlimit;
254 
255 	hdr->saddr = fl6->saddr;
256 	hdr->daddr = *first_hop;
257 
258 	skb->protocol = htons(ETH_P_IPV6);
259 	skb->priority = sk->sk_priority;
260 	skb->mark = mark;
261 
262 	mtu = dst_mtu(dst);
263 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
264 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
265 			      IPSTATS_MIB_OUT, skb->len);
266 
267 		/* if egress device is enslaved to an L3 master device pass the
268 		 * skb to its handler for processing
269 		 */
270 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
271 		if (unlikely(!skb))
272 			return 0;
273 
274 		/* hooks should never assume socket lock is held.
275 		 * we promote our socket to non const
276 		 */
277 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
278 			       net, (struct sock *)sk, skb, NULL, dst->dev,
279 			       dst_output);
280 	}
281 
282 	skb->dev = dst->dev;
283 	/* ipv6_local_error() does not require socket lock,
284 	 * we promote our socket to non const
285 	 */
286 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
287 
288 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
289 	kfree_skb(skb);
290 	return -EMSGSIZE;
291 }
292 EXPORT_SYMBOL(ip6_xmit);
293 
294 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
295 {
296 	struct ip6_ra_chain *ra;
297 	struct sock *last = NULL;
298 
299 	read_lock(&ip6_ra_lock);
300 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
301 		struct sock *sk = ra->sk;
302 		if (sk && ra->sel == sel &&
303 		    (!sk->sk_bound_dev_if ||
304 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
305 			if (last) {
306 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
307 				if (skb2)
308 					rawv6_rcv(last, skb2);
309 			}
310 			last = sk;
311 		}
312 	}
313 
314 	if (last) {
315 		rawv6_rcv(last, skb);
316 		read_unlock(&ip6_ra_lock);
317 		return 1;
318 	}
319 	read_unlock(&ip6_ra_lock);
320 	return 0;
321 }
322 
323 static int ip6_forward_proxy_check(struct sk_buff *skb)
324 {
325 	struct ipv6hdr *hdr = ipv6_hdr(skb);
326 	u8 nexthdr = hdr->nexthdr;
327 	__be16 frag_off;
328 	int offset;
329 
330 	if (ipv6_ext_hdr(nexthdr)) {
331 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
332 		if (offset < 0)
333 			return 0;
334 	} else
335 		offset = sizeof(struct ipv6hdr);
336 
337 	if (nexthdr == IPPROTO_ICMPV6) {
338 		struct icmp6hdr *icmp6;
339 
340 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
341 					 offset + 1 - skb->data)))
342 			return 0;
343 
344 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
345 
346 		switch (icmp6->icmp6_type) {
347 		case NDISC_ROUTER_SOLICITATION:
348 		case NDISC_ROUTER_ADVERTISEMENT:
349 		case NDISC_NEIGHBOUR_SOLICITATION:
350 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
351 		case NDISC_REDIRECT:
352 			/* For reaction involving unicast neighbor discovery
353 			 * message destined to the proxied address, pass it to
354 			 * input function.
355 			 */
356 			return 1;
357 		default:
358 			break;
359 		}
360 	}
361 
362 	/*
363 	 * The proxying router can't forward traffic sent to a link-local
364 	 * address, so signal the sender and discard the packet. This
365 	 * behavior is clarified by the MIPv6 specification.
366 	 */
367 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
368 		dst_link_failure(skb);
369 		return -1;
370 	}
371 
372 	return 0;
373 }
374 
375 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
376 				     struct sk_buff *skb)
377 {
378 	return dst_output(net, sk, skb);
379 }
380 
381 unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
382 {
383 	unsigned int mtu;
384 	struct inet6_dev *idev;
385 
386 	if (dst_metric_locked(dst, RTAX_MTU)) {
387 		mtu = dst_metric_raw(dst, RTAX_MTU);
388 		if (mtu)
389 			return mtu;
390 	}
391 
392 	mtu = IPV6_MIN_MTU;
393 	rcu_read_lock();
394 	idev = __in6_dev_get(dst->dev);
395 	if (idev)
396 		mtu = idev->cnf.mtu6;
397 	rcu_read_unlock();
398 
399 	return mtu;
400 }
401 EXPORT_SYMBOL_GPL(ip6_dst_mtu_forward);
402 
403 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
404 {
405 	if (skb->len <= mtu)
406 		return false;
407 
408 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
409 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
410 		return true;
411 
412 	if (skb->ignore_df)
413 		return false;
414 
415 	if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
416 		return false;
417 
418 	return true;
419 }
420 
421 int ip6_forward(struct sk_buff *skb)
422 {
423 	struct dst_entry *dst = skb_dst(skb);
424 	struct ipv6hdr *hdr = ipv6_hdr(skb);
425 	struct inet6_skb_parm *opt = IP6CB(skb);
426 	struct net *net = dev_net(dst->dev);
427 	u32 mtu;
428 
429 	if (net->ipv6.devconf_all->forwarding == 0)
430 		goto error;
431 
432 	if (skb->pkt_type != PACKET_HOST)
433 		goto drop;
434 
435 	if (unlikely(skb->sk))
436 		goto drop;
437 
438 	if (skb_warn_if_lro(skb))
439 		goto drop;
440 
441 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
442 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
443 				IPSTATS_MIB_INDISCARDS);
444 		goto drop;
445 	}
446 
447 	skb_forward_csum(skb);
448 
449 	/*
450 	 *	We DO NOT make any processing on
451 	 *	RA packets, pushing them to user level AS IS
452 	 *	without ane WARRANTY that application will be able
453 	 *	to interpret them. The reason is that we
454 	 *	cannot make anything clever here.
455 	 *
456 	 *	We are not end-node, so that if packet contains
457 	 *	AH/ESP, we cannot make anything.
458 	 *	Defragmentation also would be mistake, RA packets
459 	 *	cannot be fragmented, because there is no warranty
460 	 *	that different fragments will go along one path. --ANK
461 	 */
462 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
463 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
464 			return 0;
465 	}
466 
467 	/*
468 	 *	check and decrement ttl
469 	 */
470 	if (hdr->hop_limit <= 1) {
471 		/* Force OUTPUT device used as source address */
472 		skb->dev = dst->dev;
473 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
474 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
475 				IPSTATS_MIB_INHDRERRORS);
476 
477 		kfree_skb(skb);
478 		return -ETIMEDOUT;
479 	}
480 
481 	/* XXX: idev->cnf.proxy_ndp? */
482 	if (net->ipv6.devconf_all->proxy_ndp &&
483 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
484 		int proxied = ip6_forward_proxy_check(skb);
485 		if (proxied > 0)
486 			return ip6_input(skb);
487 		else if (proxied < 0) {
488 			__IP6_INC_STATS(net, ip6_dst_idev(dst),
489 					IPSTATS_MIB_INDISCARDS);
490 			goto drop;
491 		}
492 	}
493 
494 	if (!xfrm6_route_forward(skb)) {
495 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
496 				IPSTATS_MIB_INDISCARDS);
497 		goto drop;
498 	}
499 	dst = skb_dst(skb);
500 
501 	/* IPv6 specs say nothing about it, but it is clear that we cannot
502 	   send redirects to source routed frames.
503 	   We don't send redirects to frames decapsulated from IPsec.
504 	 */
505 	if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
506 		struct in6_addr *target = NULL;
507 		struct inet_peer *peer;
508 		struct rt6_info *rt;
509 
510 		/*
511 		 *	incoming and outgoing devices are the same
512 		 *	send a redirect.
513 		 */
514 
515 		rt = (struct rt6_info *) dst;
516 		if (rt->rt6i_flags & RTF_GATEWAY)
517 			target = &rt->rt6i_gateway;
518 		else
519 			target = &hdr->daddr;
520 
521 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
522 
523 		/* Limit redirects both by destination (here)
524 		   and by source (inside ndisc_send_redirect)
525 		 */
526 		if (inet_peer_xrlim_allow(peer, 1*HZ))
527 			ndisc_send_redirect(skb, target);
528 		if (peer)
529 			inet_putpeer(peer);
530 	} else {
531 		int addrtype = ipv6_addr_type(&hdr->saddr);
532 
533 		/* This check is security critical. */
534 		if (addrtype == IPV6_ADDR_ANY ||
535 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
536 			goto error;
537 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
538 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
539 				    ICMPV6_NOT_NEIGHBOUR, 0);
540 			goto error;
541 		}
542 	}
543 
544 	mtu = ip6_dst_mtu_forward(dst);
545 	if (mtu < IPV6_MIN_MTU)
546 		mtu = IPV6_MIN_MTU;
547 
548 	if (ip6_pkt_too_big(skb, mtu)) {
549 		/* Again, force OUTPUT device used as source address */
550 		skb->dev = dst->dev;
551 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
552 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
553 				IPSTATS_MIB_INTOOBIGERRORS);
554 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
555 				IPSTATS_MIB_FRAGFAILS);
556 		kfree_skb(skb);
557 		return -EMSGSIZE;
558 	}
559 
560 	if (skb_cow(skb, dst->dev->hard_header_len)) {
561 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
562 				IPSTATS_MIB_OUTDISCARDS);
563 		goto drop;
564 	}
565 
566 	hdr = ipv6_hdr(skb);
567 
568 	/* Mangling hops number delayed to point after skb COW */
569 
570 	hdr->hop_limit--;
571 
572 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
573 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
574 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
575 		       net, NULL, skb, skb->dev, dst->dev,
576 		       ip6_forward_finish);
577 
578 error:
579 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
580 drop:
581 	kfree_skb(skb);
582 	return -EINVAL;
583 }
584 
585 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
586 {
587 	to->pkt_type = from->pkt_type;
588 	to->priority = from->priority;
589 	to->protocol = from->protocol;
590 	skb_dst_drop(to);
591 	skb_dst_set(to, dst_clone(skb_dst(from)));
592 	to->dev = from->dev;
593 	to->mark = from->mark;
594 
595 #ifdef CONFIG_NET_SCHED
596 	to->tc_index = from->tc_index;
597 #endif
598 	nf_copy(to, from);
599 	skb_copy_secmark(to, from);
600 }
601 
602 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
603 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
604 {
605 	struct sk_buff *frag;
606 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
607 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
608 				inet6_sk(skb->sk) : NULL;
609 	struct ipv6hdr *tmp_hdr;
610 	struct frag_hdr *fh;
611 	unsigned int mtu, hlen, left, len;
612 	int hroom, troom;
613 	__be32 frag_id;
614 	int ptr, offset = 0, err = 0;
615 	u8 *prevhdr, nexthdr = 0;
616 
617 	err = ip6_find_1stfragopt(skb, &prevhdr);
618 	if (err < 0)
619 		goto fail;
620 	hlen = err;
621 	nexthdr = *prevhdr;
622 
623 	mtu = ip6_skb_dst_mtu(skb);
624 
625 	/* We must not fragment if the socket is set to force MTU discovery
626 	 * or if the skb it not generated by a local socket.
627 	 */
628 	if (unlikely(!skb->ignore_df && skb->len > mtu))
629 		goto fail_toobig;
630 
631 	if (IP6CB(skb)->frag_max_size) {
632 		if (IP6CB(skb)->frag_max_size > mtu)
633 			goto fail_toobig;
634 
635 		/* don't send fragments larger than what we received */
636 		mtu = IP6CB(skb)->frag_max_size;
637 		if (mtu < IPV6_MIN_MTU)
638 			mtu = IPV6_MIN_MTU;
639 	}
640 
641 	if (np && np->frag_size < mtu) {
642 		if (np->frag_size)
643 			mtu = np->frag_size;
644 	}
645 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
646 		goto fail_toobig;
647 	mtu -= hlen + sizeof(struct frag_hdr);
648 
649 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
650 				    &ipv6_hdr(skb)->saddr);
651 
652 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
653 	    (err = skb_checksum_help(skb)))
654 		goto fail;
655 
656 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
657 	if (skb_has_frag_list(skb)) {
658 		unsigned int first_len = skb_pagelen(skb);
659 		struct sk_buff *frag2;
660 
661 		if (first_len - hlen > mtu ||
662 		    ((first_len - hlen) & 7) ||
663 		    skb_cloned(skb) ||
664 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
665 			goto slow_path;
666 
667 		skb_walk_frags(skb, frag) {
668 			/* Correct geometry. */
669 			if (frag->len > mtu ||
670 			    ((frag->len & 7) && frag->next) ||
671 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
672 				goto slow_path_clean;
673 
674 			/* Partially cloned skb? */
675 			if (skb_shared(frag))
676 				goto slow_path_clean;
677 
678 			BUG_ON(frag->sk);
679 			if (skb->sk) {
680 				frag->sk = skb->sk;
681 				frag->destructor = sock_wfree;
682 			}
683 			skb->truesize -= frag->truesize;
684 		}
685 
686 		err = 0;
687 		offset = 0;
688 		/* BUILD HEADER */
689 
690 		*prevhdr = NEXTHDR_FRAGMENT;
691 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
692 		if (!tmp_hdr) {
693 			err = -ENOMEM;
694 			goto fail;
695 		}
696 		frag = skb_shinfo(skb)->frag_list;
697 		skb_frag_list_init(skb);
698 
699 		__skb_pull(skb, hlen);
700 		fh = __skb_push(skb, sizeof(struct frag_hdr));
701 		__skb_push(skb, hlen);
702 		skb_reset_network_header(skb);
703 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
704 
705 		fh->nexthdr = nexthdr;
706 		fh->reserved = 0;
707 		fh->frag_off = htons(IP6_MF);
708 		fh->identification = frag_id;
709 
710 		first_len = skb_pagelen(skb);
711 		skb->data_len = first_len - skb_headlen(skb);
712 		skb->len = first_len;
713 		ipv6_hdr(skb)->payload_len = htons(first_len -
714 						   sizeof(struct ipv6hdr));
715 
716 		for (;;) {
717 			/* Prepare header of the next frame,
718 			 * before previous one went down. */
719 			if (frag) {
720 				frag->ip_summed = CHECKSUM_NONE;
721 				skb_reset_transport_header(frag);
722 				fh = __skb_push(frag, sizeof(struct frag_hdr));
723 				__skb_push(frag, hlen);
724 				skb_reset_network_header(frag);
725 				memcpy(skb_network_header(frag), tmp_hdr,
726 				       hlen);
727 				offset += skb->len - hlen - sizeof(struct frag_hdr);
728 				fh->nexthdr = nexthdr;
729 				fh->reserved = 0;
730 				fh->frag_off = htons(offset);
731 				if (frag->next)
732 					fh->frag_off |= htons(IP6_MF);
733 				fh->identification = frag_id;
734 				ipv6_hdr(frag)->payload_len =
735 						htons(frag->len -
736 						      sizeof(struct ipv6hdr));
737 				ip6_copy_metadata(frag, skb);
738 			}
739 
740 			err = output(net, sk, skb);
741 			if (!err)
742 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
743 					      IPSTATS_MIB_FRAGCREATES);
744 
745 			if (err || !frag)
746 				break;
747 
748 			skb = frag;
749 			frag = skb->next;
750 			skb->next = NULL;
751 		}
752 
753 		kfree(tmp_hdr);
754 
755 		if (err == 0) {
756 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
757 				      IPSTATS_MIB_FRAGOKS);
758 			return 0;
759 		}
760 
761 		kfree_skb_list(frag);
762 
763 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
764 			      IPSTATS_MIB_FRAGFAILS);
765 		return err;
766 
767 slow_path_clean:
768 		skb_walk_frags(skb, frag2) {
769 			if (frag2 == frag)
770 				break;
771 			frag2->sk = NULL;
772 			frag2->destructor = NULL;
773 			skb->truesize += frag2->truesize;
774 		}
775 	}
776 
777 slow_path:
778 	left = skb->len - hlen;		/* Space per frame */
779 	ptr = hlen;			/* Where to start from */
780 
781 	/*
782 	 *	Fragment the datagram.
783 	 */
784 
785 	troom = rt->dst.dev->needed_tailroom;
786 
787 	/*
788 	 *	Keep copying data until we run out.
789 	 */
790 	while (left > 0)	{
791 		u8 *fragnexthdr_offset;
792 
793 		len = left;
794 		/* IF: it doesn't fit, use 'mtu' - the data space left */
795 		if (len > mtu)
796 			len = mtu;
797 		/* IF: we are not sending up to and including the packet end
798 		   then align the next start on an eight byte boundary */
799 		if (len < left)	{
800 			len &= ~7;
801 		}
802 
803 		/* Allocate buffer */
804 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
805 				 hroom + troom, GFP_ATOMIC);
806 		if (!frag) {
807 			err = -ENOMEM;
808 			goto fail;
809 		}
810 
811 		/*
812 		 *	Set up data on packet
813 		 */
814 
815 		ip6_copy_metadata(frag, skb);
816 		skb_reserve(frag, hroom);
817 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
818 		skb_reset_network_header(frag);
819 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
820 		frag->transport_header = (frag->network_header + hlen +
821 					  sizeof(struct frag_hdr));
822 
823 		/*
824 		 *	Charge the memory for the fragment to any owner
825 		 *	it might possess
826 		 */
827 		if (skb->sk)
828 			skb_set_owner_w(frag, skb->sk);
829 
830 		/*
831 		 *	Copy the packet header into the new buffer.
832 		 */
833 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
834 
835 		fragnexthdr_offset = skb_network_header(frag);
836 		fragnexthdr_offset += prevhdr - skb_network_header(skb);
837 		*fragnexthdr_offset = NEXTHDR_FRAGMENT;
838 
839 		/*
840 		 *	Build fragment header.
841 		 */
842 		fh->nexthdr = nexthdr;
843 		fh->reserved = 0;
844 		fh->identification = frag_id;
845 
846 		/*
847 		 *	Copy a block of the IP datagram.
848 		 */
849 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
850 				     len));
851 		left -= len;
852 
853 		fh->frag_off = htons(offset);
854 		if (left > 0)
855 			fh->frag_off |= htons(IP6_MF);
856 		ipv6_hdr(frag)->payload_len = htons(frag->len -
857 						    sizeof(struct ipv6hdr));
858 
859 		ptr += len;
860 		offset += len;
861 
862 		/*
863 		 *	Put this fragment into the sending queue.
864 		 */
865 		err = output(net, sk, frag);
866 		if (err)
867 			goto fail;
868 
869 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
870 			      IPSTATS_MIB_FRAGCREATES);
871 	}
872 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
873 		      IPSTATS_MIB_FRAGOKS);
874 	consume_skb(skb);
875 	return err;
876 
877 fail_toobig:
878 	if (skb->sk && dst_allfrag(skb_dst(skb)))
879 		sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
880 
881 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
882 	err = -EMSGSIZE;
883 
884 fail:
885 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
886 		      IPSTATS_MIB_FRAGFAILS);
887 	kfree_skb(skb);
888 	return err;
889 }
890 
891 static inline int ip6_rt_check(const struct rt6key *rt_key,
892 			       const struct in6_addr *fl_addr,
893 			       const struct in6_addr *addr_cache)
894 {
895 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
896 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
897 }
898 
899 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
900 					  struct dst_entry *dst,
901 					  const struct flowi6 *fl6)
902 {
903 	struct ipv6_pinfo *np = inet6_sk(sk);
904 	struct rt6_info *rt;
905 
906 	if (!dst)
907 		goto out;
908 
909 	if (dst->ops->family != AF_INET6) {
910 		dst_release(dst);
911 		return NULL;
912 	}
913 
914 	rt = (struct rt6_info *)dst;
915 	/* Yes, checking route validity in not connected
916 	 * case is not very simple. Take into account,
917 	 * that we do not support routing by source, TOS,
918 	 * and MSG_DONTROUTE		--ANK (980726)
919 	 *
920 	 * 1. ip6_rt_check(): If route was host route,
921 	 *    check that cached destination is current.
922 	 *    If it is network route, we still may
923 	 *    check its validity using saved pointer
924 	 *    to the last used address: daddr_cache.
925 	 *    We do not want to save whole address now,
926 	 *    (because main consumer of this service
927 	 *    is tcp, which has not this problem),
928 	 *    so that the last trick works only on connected
929 	 *    sockets.
930 	 * 2. oif also should be the same.
931 	 */
932 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
933 #ifdef CONFIG_IPV6_SUBTREES
934 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
935 #endif
936 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
937 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
938 		dst_release(dst);
939 		dst = NULL;
940 	}
941 
942 out:
943 	return dst;
944 }
945 
946 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
947 			       struct dst_entry **dst, struct flowi6 *fl6)
948 {
949 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
950 	struct neighbour *n;
951 	struct rt6_info *rt;
952 #endif
953 	int err;
954 	int flags = 0;
955 
956 	/* The correct way to handle this would be to do
957 	 * ip6_route_get_saddr, and then ip6_route_output; however,
958 	 * the route-specific preferred source forces the
959 	 * ip6_route_output call _before_ ip6_route_get_saddr.
960 	 *
961 	 * In source specific routing (no src=any default route),
962 	 * ip6_route_output will fail given src=any saddr, though, so
963 	 * that's why we try it again later.
964 	 */
965 	if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
966 		struct rt6_info *rt;
967 		bool had_dst = *dst != NULL;
968 
969 		if (!had_dst)
970 			*dst = ip6_route_output(net, sk, fl6);
971 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
972 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
973 					  sk ? inet6_sk(sk)->srcprefs : 0,
974 					  &fl6->saddr);
975 		if (err)
976 			goto out_err_release;
977 
978 		/* If we had an erroneous initial result, pretend it
979 		 * never existed and let the SA-enabled version take
980 		 * over.
981 		 */
982 		if (!had_dst && (*dst)->error) {
983 			dst_release(*dst);
984 			*dst = NULL;
985 		}
986 
987 		if (fl6->flowi6_oif)
988 			flags |= RT6_LOOKUP_F_IFACE;
989 	}
990 
991 	if (!*dst)
992 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
993 
994 	err = (*dst)->error;
995 	if (err)
996 		goto out_err_release;
997 
998 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
999 	/*
1000 	 * Here if the dst entry we've looked up
1001 	 * has a neighbour entry that is in the INCOMPLETE
1002 	 * state and the src address from the flow is
1003 	 * marked as OPTIMISTIC, we release the found
1004 	 * dst entry and replace it instead with the
1005 	 * dst entry of the nexthop router
1006 	 */
1007 	rt = (struct rt6_info *) *dst;
1008 	rcu_read_lock_bh();
1009 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1010 				      rt6_nexthop(rt, &fl6->daddr));
1011 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1012 	rcu_read_unlock_bh();
1013 
1014 	if (err) {
1015 		struct inet6_ifaddr *ifp;
1016 		struct flowi6 fl_gw6;
1017 		int redirect;
1018 
1019 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1020 				      (*dst)->dev, 1);
1021 
1022 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1023 		if (ifp)
1024 			in6_ifa_put(ifp);
1025 
1026 		if (redirect) {
1027 			/*
1028 			 * We need to get the dst entry for the
1029 			 * default router instead
1030 			 */
1031 			dst_release(*dst);
1032 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1033 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1034 			*dst = ip6_route_output(net, sk, &fl_gw6);
1035 			err = (*dst)->error;
1036 			if (err)
1037 				goto out_err_release;
1038 		}
1039 	}
1040 #endif
1041 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1042 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1043 		err = -EAFNOSUPPORT;
1044 		goto out_err_release;
1045 	}
1046 
1047 	return 0;
1048 
1049 out_err_release:
1050 	dst_release(*dst);
1051 	*dst = NULL;
1052 
1053 	if (err == -ENETUNREACH)
1054 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1055 	return err;
1056 }
1057 
1058 /**
1059  *	ip6_dst_lookup - perform route lookup on flow
1060  *	@sk: socket which provides route info
1061  *	@dst: pointer to dst_entry * for result
1062  *	@fl6: flow to lookup
1063  *
1064  *	This function performs a route lookup on the given flow.
1065  *
1066  *	It returns zero on success, or a standard errno code on error.
1067  */
1068 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1069 		   struct flowi6 *fl6)
1070 {
1071 	*dst = NULL;
1072 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1073 }
1074 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1075 
1076 /**
1077  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1078  *	@sk: socket which provides route info
1079  *	@fl6: flow to lookup
1080  *	@final_dst: final destination address for ipsec lookup
1081  *
1082  *	This function performs a route lookup on the given flow.
1083  *
1084  *	It returns a valid dst pointer on success, or a pointer encoded
1085  *	error code.
1086  */
1087 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1088 				      const struct in6_addr *final_dst)
1089 {
1090 	struct dst_entry *dst = NULL;
1091 	int err;
1092 
1093 	err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1094 	if (err)
1095 		return ERR_PTR(err);
1096 	if (final_dst)
1097 		fl6->daddr = *final_dst;
1098 
1099 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1100 }
1101 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1102 
1103 /**
1104  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1105  *	@sk: socket which provides the dst cache and route info
1106  *	@fl6: flow to lookup
1107  *	@final_dst: final destination address for ipsec lookup
1108  *
1109  *	This function performs a route lookup on the given flow with the
1110  *	possibility of using the cached route in the socket if it is valid.
1111  *	It will take the socket dst lock when operating on the dst cache.
1112  *	As a result, this function can only be used in process context.
1113  *
1114  *	It returns a valid dst pointer on success, or a pointer encoded
1115  *	error code.
1116  */
1117 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1118 					 const struct in6_addr *final_dst)
1119 {
1120 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1121 
1122 	dst = ip6_sk_dst_check(sk, dst, fl6);
1123 	if (!dst)
1124 		dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1125 
1126 	return dst;
1127 }
1128 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1129 
1130 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1131 					       gfp_t gfp)
1132 {
1133 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1134 }
1135 
1136 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1137 						gfp_t gfp)
1138 {
1139 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1140 }
1141 
1142 static void ip6_append_data_mtu(unsigned int *mtu,
1143 				int *maxfraglen,
1144 				unsigned int fragheaderlen,
1145 				struct sk_buff *skb,
1146 				struct rt6_info *rt,
1147 				unsigned int orig_mtu)
1148 {
1149 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1150 		if (!skb) {
1151 			/* first fragment, reserve header_len */
1152 			*mtu = orig_mtu - rt->dst.header_len;
1153 
1154 		} else {
1155 			/*
1156 			 * this fragment is not first, the headers
1157 			 * space is regarded as data space.
1158 			 */
1159 			*mtu = orig_mtu;
1160 		}
1161 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1162 			      + fragheaderlen - sizeof(struct frag_hdr);
1163 	}
1164 }
1165 
1166 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1167 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1168 			  struct rt6_info *rt, struct flowi6 *fl6)
1169 {
1170 	struct ipv6_pinfo *np = inet6_sk(sk);
1171 	unsigned int mtu;
1172 	struct ipv6_txoptions *opt = ipc6->opt;
1173 
1174 	/*
1175 	 * setup for corking
1176 	 */
1177 	if (opt) {
1178 		if (WARN_ON(v6_cork->opt))
1179 			return -EINVAL;
1180 
1181 		v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1182 		if (unlikely(!v6_cork->opt))
1183 			return -ENOBUFS;
1184 
1185 		v6_cork->opt->tot_len = sizeof(*opt);
1186 		v6_cork->opt->opt_flen = opt->opt_flen;
1187 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1188 
1189 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1190 						    sk->sk_allocation);
1191 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1192 			return -ENOBUFS;
1193 
1194 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1195 						    sk->sk_allocation);
1196 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1197 			return -ENOBUFS;
1198 
1199 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1200 						   sk->sk_allocation);
1201 		if (opt->hopopt && !v6_cork->opt->hopopt)
1202 			return -ENOBUFS;
1203 
1204 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1205 						    sk->sk_allocation);
1206 		if (opt->srcrt && !v6_cork->opt->srcrt)
1207 			return -ENOBUFS;
1208 
1209 		/* need source address above miyazawa*/
1210 	}
1211 	dst_hold(&rt->dst);
1212 	cork->base.dst = &rt->dst;
1213 	cork->fl.u.ip6 = *fl6;
1214 	v6_cork->hop_limit = ipc6->hlimit;
1215 	v6_cork->tclass = ipc6->tclass;
1216 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1217 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1218 		      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1219 	else
1220 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1221 		      rt->dst.dev->mtu : dst_mtu(xfrm_dst_path(&rt->dst));
1222 	if (np->frag_size < mtu) {
1223 		if (np->frag_size)
1224 			mtu = np->frag_size;
1225 	}
1226 	cork->base.fragsize = mtu;
1227 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1228 		cork->base.flags |= IPCORK_ALLFRAG;
1229 	cork->base.length = 0;
1230 
1231 	return 0;
1232 }
1233 
1234 static int __ip6_append_data(struct sock *sk,
1235 			     struct flowi6 *fl6,
1236 			     struct sk_buff_head *queue,
1237 			     struct inet_cork *cork,
1238 			     struct inet6_cork *v6_cork,
1239 			     struct page_frag *pfrag,
1240 			     int getfrag(void *from, char *to, int offset,
1241 					 int len, int odd, struct sk_buff *skb),
1242 			     void *from, int length, int transhdrlen,
1243 			     unsigned int flags, struct ipcm6_cookie *ipc6,
1244 			     const struct sockcm_cookie *sockc)
1245 {
1246 	struct sk_buff *skb, *skb_prev = NULL;
1247 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1248 	int exthdrlen = 0;
1249 	int dst_exthdrlen = 0;
1250 	int hh_len;
1251 	int copy;
1252 	int err;
1253 	int offset = 0;
1254 	__u8 tx_flags = 0;
1255 	u32 tskey = 0;
1256 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1257 	struct ipv6_txoptions *opt = v6_cork->opt;
1258 	int csummode = CHECKSUM_NONE;
1259 	unsigned int maxnonfragsize, headersize;
1260 
1261 	skb = skb_peek_tail(queue);
1262 	if (!skb) {
1263 		exthdrlen = opt ? opt->opt_flen : 0;
1264 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1265 	}
1266 
1267 	mtu = cork->fragsize;
1268 	orig_mtu = mtu;
1269 
1270 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1271 
1272 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1273 			(opt ? opt->opt_nflen : 0);
1274 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1275 		     sizeof(struct frag_hdr);
1276 
1277 	headersize = sizeof(struct ipv6hdr) +
1278 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1279 		     (dst_allfrag(&rt->dst) ?
1280 		      sizeof(struct frag_hdr) : 0) +
1281 		     rt->rt6i_nfheader_len;
1282 
1283 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1284 	    (sk->sk_protocol == IPPROTO_UDP ||
1285 	     sk->sk_protocol == IPPROTO_RAW)) {
1286 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1287 				sizeof(struct ipv6hdr));
1288 		goto emsgsize;
1289 	}
1290 
1291 	if (ip6_sk_ignore_df(sk))
1292 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1293 	else
1294 		maxnonfragsize = mtu;
1295 
1296 	if (cork->length + length > maxnonfragsize - headersize) {
1297 emsgsize:
1298 		ipv6_local_error(sk, EMSGSIZE, fl6,
1299 				 mtu - headersize +
1300 				 sizeof(struct ipv6hdr));
1301 		return -EMSGSIZE;
1302 	}
1303 
1304 	/* CHECKSUM_PARTIAL only with no extension headers and when
1305 	 * we are not going to fragment
1306 	 */
1307 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1308 	    headersize == sizeof(struct ipv6hdr) &&
1309 	    length <= mtu - headersize &&
1310 	    !(flags & MSG_MORE) &&
1311 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1312 		csummode = CHECKSUM_PARTIAL;
1313 
1314 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1315 		sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1316 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1317 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1318 			tskey = sk->sk_tskey++;
1319 	}
1320 
1321 	/*
1322 	 * Let's try using as much space as possible.
1323 	 * Use MTU if total length of the message fits into the MTU.
1324 	 * Otherwise, we need to reserve fragment header and
1325 	 * fragment alignment (= 8-15 octects, in total).
1326 	 *
1327 	 * Note that we may need to "move" the data from the tail of
1328 	 * of the buffer to the new fragment when we split
1329 	 * the message.
1330 	 *
1331 	 * FIXME: It may be fragmented into multiple chunks
1332 	 *        at once if non-fragmentable extension headers
1333 	 *        are too large.
1334 	 * --yoshfuji
1335 	 */
1336 
1337 	cork->length += length;
1338 	if (!skb)
1339 		goto alloc_new_skb;
1340 
1341 	while (length > 0) {
1342 		/* Check if the remaining data fits into current packet. */
1343 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1344 		if (copy < length)
1345 			copy = maxfraglen - skb->len;
1346 
1347 		if (copy <= 0) {
1348 			char *data;
1349 			unsigned int datalen;
1350 			unsigned int fraglen;
1351 			unsigned int fraggap;
1352 			unsigned int alloclen;
1353 alloc_new_skb:
1354 			/* There's no room in the current skb */
1355 			if (skb)
1356 				fraggap = skb->len - maxfraglen;
1357 			else
1358 				fraggap = 0;
1359 			/* update mtu and maxfraglen if necessary */
1360 			if (!skb || !skb_prev)
1361 				ip6_append_data_mtu(&mtu, &maxfraglen,
1362 						    fragheaderlen, skb, rt,
1363 						    orig_mtu);
1364 
1365 			skb_prev = skb;
1366 
1367 			/*
1368 			 * If remaining data exceeds the mtu,
1369 			 * we know we need more fragment(s).
1370 			 */
1371 			datalen = length + fraggap;
1372 
1373 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1374 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1375 			if ((flags & MSG_MORE) &&
1376 			    !(rt->dst.dev->features&NETIF_F_SG))
1377 				alloclen = mtu;
1378 			else
1379 				alloclen = datalen + fragheaderlen;
1380 
1381 			alloclen += dst_exthdrlen;
1382 
1383 			if (datalen != length + fraggap) {
1384 				/*
1385 				 * this is not the last fragment, the trailer
1386 				 * space is regarded as data space.
1387 				 */
1388 				datalen += rt->dst.trailer_len;
1389 			}
1390 
1391 			alloclen += rt->dst.trailer_len;
1392 			fraglen = datalen + fragheaderlen;
1393 
1394 			/*
1395 			 * We just reserve space for fragment header.
1396 			 * Note: this may be overallocation if the message
1397 			 * (without MSG_MORE) fits into the MTU.
1398 			 */
1399 			alloclen += sizeof(struct frag_hdr);
1400 
1401 			copy = datalen - transhdrlen - fraggap;
1402 			if (copy < 0) {
1403 				err = -EINVAL;
1404 				goto error;
1405 			}
1406 			if (transhdrlen) {
1407 				skb = sock_alloc_send_skb(sk,
1408 						alloclen + hh_len,
1409 						(flags & MSG_DONTWAIT), &err);
1410 			} else {
1411 				skb = NULL;
1412 				if (refcount_read(&sk->sk_wmem_alloc) <=
1413 				    2 * sk->sk_sndbuf)
1414 					skb = sock_wmalloc(sk,
1415 							   alloclen + hh_len, 1,
1416 							   sk->sk_allocation);
1417 				if (unlikely(!skb))
1418 					err = -ENOBUFS;
1419 			}
1420 			if (!skb)
1421 				goto error;
1422 			/*
1423 			 *	Fill in the control structures
1424 			 */
1425 			skb->protocol = htons(ETH_P_IPV6);
1426 			skb->ip_summed = csummode;
1427 			skb->csum = 0;
1428 			/* reserve for fragmentation and ipsec header */
1429 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1430 				    dst_exthdrlen);
1431 
1432 			/* Only the initial fragment is time stamped */
1433 			skb_shinfo(skb)->tx_flags = tx_flags;
1434 			tx_flags = 0;
1435 			skb_shinfo(skb)->tskey = tskey;
1436 			tskey = 0;
1437 
1438 			/*
1439 			 *	Find where to start putting bytes
1440 			 */
1441 			data = skb_put(skb, fraglen);
1442 			skb_set_network_header(skb, exthdrlen);
1443 			data += fragheaderlen;
1444 			skb->transport_header = (skb->network_header +
1445 						 fragheaderlen);
1446 			if (fraggap) {
1447 				skb->csum = skb_copy_and_csum_bits(
1448 					skb_prev, maxfraglen,
1449 					data + transhdrlen, fraggap, 0);
1450 				skb_prev->csum = csum_sub(skb_prev->csum,
1451 							  skb->csum);
1452 				data += fraggap;
1453 				pskb_trim_unique(skb_prev, maxfraglen);
1454 			}
1455 			if (copy > 0 &&
1456 			    getfrag(from, data + transhdrlen, offset,
1457 				    copy, fraggap, skb) < 0) {
1458 				err = -EFAULT;
1459 				kfree_skb(skb);
1460 				goto error;
1461 			}
1462 
1463 			offset += copy;
1464 			length -= datalen - fraggap;
1465 			transhdrlen = 0;
1466 			exthdrlen = 0;
1467 			dst_exthdrlen = 0;
1468 
1469 			if ((flags & MSG_CONFIRM) && !skb_prev)
1470 				skb_set_dst_pending_confirm(skb, 1);
1471 
1472 			/*
1473 			 * Put the packet on the pending queue
1474 			 */
1475 			__skb_queue_tail(queue, skb);
1476 			continue;
1477 		}
1478 
1479 		if (copy > length)
1480 			copy = length;
1481 
1482 		if (!(rt->dst.dev->features&NETIF_F_SG)) {
1483 			unsigned int off;
1484 
1485 			off = skb->len;
1486 			if (getfrag(from, skb_put(skb, copy),
1487 						offset, copy, off, skb) < 0) {
1488 				__skb_trim(skb, off);
1489 				err = -EFAULT;
1490 				goto error;
1491 			}
1492 		} else {
1493 			int i = skb_shinfo(skb)->nr_frags;
1494 
1495 			err = -ENOMEM;
1496 			if (!sk_page_frag_refill(sk, pfrag))
1497 				goto error;
1498 
1499 			if (!skb_can_coalesce(skb, i, pfrag->page,
1500 					      pfrag->offset)) {
1501 				err = -EMSGSIZE;
1502 				if (i == MAX_SKB_FRAGS)
1503 					goto error;
1504 
1505 				__skb_fill_page_desc(skb, i, pfrag->page,
1506 						     pfrag->offset, 0);
1507 				skb_shinfo(skb)->nr_frags = ++i;
1508 				get_page(pfrag->page);
1509 			}
1510 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1511 			if (getfrag(from,
1512 				    page_address(pfrag->page) + pfrag->offset,
1513 				    offset, copy, skb->len, skb) < 0)
1514 				goto error_efault;
1515 
1516 			pfrag->offset += copy;
1517 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1518 			skb->len += copy;
1519 			skb->data_len += copy;
1520 			skb->truesize += copy;
1521 			refcount_add(copy, &sk->sk_wmem_alloc);
1522 		}
1523 		offset += copy;
1524 		length -= copy;
1525 	}
1526 
1527 	return 0;
1528 
1529 error_efault:
1530 	err = -EFAULT;
1531 error:
1532 	cork->length -= length;
1533 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1534 	return err;
1535 }
1536 
1537 int ip6_append_data(struct sock *sk,
1538 		    int getfrag(void *from, char *to, int offset, int len,
1539 				int odd, struct sk_buff *skb),
1540 		    void *from, int length, int transhdrlen,
1541 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1542 		    struct rt6_info *rt, unsigned int flags,
1543 		    const struct sockcm_cookie *sockc)
1544 {
1545 	struct inet_sock *inet = inet_sk(sk);
1546 	struct ipv6_pinfo *np = inet6_sk(sk);
1547 	int exthdrlen;
1548 	int err;
1549 
1550 	if (flags&MSG_PROBE)
1551 		return 0;
1552 	if (skb_queue_empty(&sk->sk_write_queue)) {
1553 		/*
1554 		 * setup for corking
1555 		 */
1556 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1557 				     ipc6, rt, fl6);
1558 		if (err)
1559 			return err;
1560 
1561 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1562 		length += exthdrlen;
1563 		transhdrlen += exthdrlen;
1564 	} else {
1565 		fl6 = &inet->cork.fl.u.ip6;
1566 		transhdrlen = 0;
1567 	}
1568 
1569 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1570 				 &np->cork, sk_page_frag(sk), getfrag,
1571 				 from, length, transhdrlen, flags, ipc6, sockc);
1572 }
1573 EXPORT_SYMBOL_GPL(ip6_append_data);
1574 
1575 static void ip6_cork_release(struct inet_cork_full *cork,
1576 			     struct inet6_cork *v6_cork)
1577 {
1578 	if (v6_cork->opt) {
1579 		kfree(v6_cork->opt->dst0opt);
1580 		kfree(v6_cork->opt->dst1opt);
1581 		kfree(v6_cork->opt->hopopt);
1582 		kfree(v6_cork->opt->srcrt);
1583 		kfree(v6_cork->opt);
1584 		v6_cork->opt = NULL;
1585 	}
1586 
1587 	if (cork->base.dst) {
1588 		dst_release(cork->base.dst);
1589 		cork->base.dst = NULL;
1590 		cork->base.flags &= ~IPCORK_ALLFRAG;
1591 	}
1592 	memset(&cork->fl, 0, sizeof(cork->fl));
1593 }
1594 
1595 struct sk_buff *__ip6_make_skb(struct sock *sk,
1596 			       struct sk_buff_head *queue,
1597 			       struct inet_cork_full *cork,
1598 			       struct inet6_cork *v6_cork)
1599 {
1600 	struct sk_buff *skb, *tmp_skb;
1601 	struct sk_buff **tail_skb;
1602 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1603 	struct ipv6_pinfo *np = inet6_sk(sk);
1604 	struct net *net = sock_net(sk);
1605 	struct ipv6hdr *hdr;
1606 	struct ipv6_txoptions *opt = v6_cork->opt;
1607 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1608 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1609 	unsigned char proto = fl6->flowi6_proto;
1610 
1611 	skb = __skb_dequeue(queue);
1612 	if (!skb)
1613 		goto out;
1614 	tail_skb = &(skb_shinfo(skb)->frag_list);
1615 
1616 	/* move skb->data to ip header from ext header */
1617 	if (skb->data < skb_network_header(skb))
1618 		__skb_pull(skb, skb_network_offset(skb));
1619 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1620 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1621 		*tail_skb = tmp_skb;
1622 		tail_skb = &(tmp_skb->next);
1623 		skb->len += tmp_skb->len;
1624 		skb->data_len += tmp_skb->len;
1625 		skb->truesize += tmp_skb->truesize;
1626 		tmp_skb->destructor = NULL;
1627 		tmp_skb->sk = NULL;
1628 	}
1629 
1630 	/* Allow local fragmentation. */
1631 	skb->ignore_df = ip6_sk_ignore_df(sk);
1632 
1633 	*final_dst = fl6->daddr;
1634 	__skb_pull(skb, skb_network_header_len(skb));
1635 	if (opt && opt->opt_flen)
1636 		ipv6_push_frag_opts(skb, opt, &proto);
1637 	if (opt && opt->opt_nflen)
1638 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1639 
1640 	skb_push(skb, sizeof(struct ipv6hdr));
1641 	skb_reset_network_header(skb);
1642 	hdr = ipv6_hdr(skb);
1643 
1644 	ip6_flow_hdr(hdr, v6_cork->tclass,
1645 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1646 					ip6_autoflowlabel(net, np), fl6));
1647 	hdr->hop_limit = v6_cork->hop_limit;
1648 	hdr->nexthdr = proto;
1649 	hdr->saddr = fl6->saddr;
1650 	hdr->daddr = *final_dst;
1651 
1652 	skb->priority = sk->sk_priority;
1653 	skb->mark = sk->sk_mark;
1654 
1655 	skb_dst_set(skb, dst_clone(&rt->dst));
1656 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1657 	if (proto == IPPROTO_ICMPV6) {
1658 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1659 
1660 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1661 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1662 	}
1663 
1664 	ip6_cork_release(cork, v6_cork);
1665 out:
1666 	return skb;
1667 }
1668 
1669 int ip6_send_skb(struct sk_buff *skb)
1670 {
1671 	struct net *net = sock_net(skb->sk);
1672 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1673 	int err;
1674 
1675 	err = ip6_local_out(net, skb->sk, skb);
1676 	if (err) {
1677 		if (err > 0)
1678 			err = net_xmit_errno(err);
1679 		if (err)
1680 			IP6_INC_STATS(net, rt->rt6i_idev,
1681 				      IPSTATS_MIB_OUTDISCARDS);
1682 	}
1683 
1684 	return err;
1685 }
1686 
1687 int ip6_push_pending_frames(struct sock *sk)
1688 {
1689 	struct sk_buff *skb;
1690 
1691 	skb = ip6_finish_skb(sk);
1692 	if (!skb)
1693 		return 0;
1694 
1695 	return ip6_send_skb(skb);
1696 }
1697 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1698 
1699 static void __ip6_flush_pending_frames(struct sock *sk,
1700 				       struct sk_buff_head *queue,
1701 				       struct inet_cork_full *cork,
1702 				       struct inet6_cork *v6_cork)
1703 {
1704 	struct sk_buff *skb;
1705 
1706 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1707 		if (skb_dst(skb))
1708 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1709 				      IPSTATS_MIB_OUTDISCARDS);
1710 		kfree_skb(skb);
1711 	}
1712 
1713 	ip6_cork_release(cork, v6_cork);
1714 }
1715 
1716 void ip6_flush_pending_frames(struct sock *sk)
1717 {
1718 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1719 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1720 }
1721 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1722 
1723 struct sk_buff *ip6_make_skb(struct sock *sk,
1724 			     int getfrag(void *from, char *to, int offset,
1725 					 int len, int odd, struct sk_buff *skb),
1726 			     void *from, int length, int transhdrlen,
1727 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1728 			     struct rt6_info *rt, unsigned int flags,
1729 			     const struct sockcm_cookie *sockc)
1730 {
1731 	struct inet_cork_full cork;
1732 	struct inet6_cork v6_cork;
1733 	struct sk_buff_head queue;
1734 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1735 	int err;
1736 
1737 	if (flags & MSG_PROBE)
1738 		return NULL;
1739 
1740 	__skb_queue_head_init(&queue);
1741 
1742 	cork.base.flags = 0;
1743 	cork.base.addr = 0;
1744 	cork.base.opt = NULL;
1745 	v6_cork.opt = NULL;
1746 	err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
1747 	if (err) {
1748 		ip6_cork_release(&cork, &v6_cork);
1749 		return ERR_PTR(err);
1750 	}
1751 	if (ipc6->dontfrag < 0)
1752 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1753 
1754 	err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1755 				&current->task_frag, getfrag, from,
1756 				length + exthdrlen, transhdrlen + exthdrlen,
1757 				flags, ipc6, sockc);
1758 	if (err) {
1759 		__ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1760 		return ERR_PTR(err);
1761 	}
1762 
1763 	return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1764 }
1765