xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 588b48ca)
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 
237 EXPORT_SYMBOL(ip6_xmit);
238 
239 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
240 {
241 	struct ip6_ra_chain *ra;
242 	struct sock *last = NULL;
243 
244 	read_lock(&ip6_ra_lock);
245 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
246 		struct sock *sk = ra->sk;
247 		if (sk && ra->sel == sel &&
248 		    (!sk->sk_bound_dev_if ||
249 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
250 			if (last) {
251 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
252 				if (skb2)
253 					rawv6_rcv(last, skb2);
254 			}
255 			last = sk;
256 		}
257 	}
258 
259 	if (last) {
260 		rawv6_rcv(last, skb);
261 		read_unlock(&ip6_ra_lock);
262 		return 1;
263 	}
264 	read_unlock(&ip6_ra_lock);
265 	return 0;
266 }
267 
268 static int ip6_forward_proxy_check(struct sk_buff *skb)
269 {
270 	struct ipv6hdr *hdr = ipv6_hdr(skb);
271 	u8 nexthdr = hdr->nexthdr;
272 	__be16 frag_off;
273 	int offset;
274 
275 	if (ipv6_ext_hdr(nexthdr)) {
276 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
277 		if (offset < 0)
278 			return 0;
279 	} else
280 		offset = sizeof(struct ipv6hdr);
281 
282 	if (nexthdr == IPPROTO_ICMPV6) {
283 		struct icmp6hdr *icmp6;
284 
285 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
286 					 offset + 1 - skb->data)))
287 			return 0;
288 
289 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
290 
291 		switch (icmp6->icmp6_type) {
292 		case NDISC_ROUTER_SOLICITATION:
293 		case NDISC_ROUTER_ADVERTISEMENT:
294 		case NDISC_NEIGHBOUR_SOLICITATION:
295 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
296 		case NDISC_REDIRECT:
297 			/* For reaction involving unicast neighbor discovery
298 			 * message destined to the proxied address, pass it to
299 			 * input function.
300 			 */
301 			return 1;
302 		default:
303 			break;
304 		}
305 	}
306 
307 	/*
308 	 * The proxying router can't forward traffic sent to a link-local
309 	 * address, so signal the sender and discard the packet. This
310 	 * behavior is clarified by the MIPv6 specification.
311 	 */
312 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
313 		dst_link_failure(skb);
314 		return -1;
315 	}
316 
317 	return 0;
318 }
319 
320 static inline int ip6_forward_finish(struct sk_buff *skb)
321 {
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 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
542 {
543 	static u32 ip6_idents_hashrnd __read_mostly;
544 	u32 hash, id;
545 
546 	net_get_random_once(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
547 
548 	hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
549 	hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
550 
551 	id = ip_idents_reserve(hash, 1);
552 	fhdr->identification = htonl(id);
553 }
554 
555 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
556 {
557 	struct sk_buff *frag;
558 	struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
559 	struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
560 	struct ipv6hdr *tmp_hdr;
561 	struct frag_hdr *fh;
562 	unsigned int mtu, hlen, left, len;
563 	int hroom, troom;
564 	__be32 frag_id = 0;
565 	int ptr, offset = 0, err=0;
566 	u8 *prevhdr, nexthdr = 0;
567 	struct net *net = dev_net(skb_dst(skb)->dev);
568 
569 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
570 	nexthdr = *prevhdr;
571 
572 	mtu = ip6_skb_dst_mtu(skb);
573 
574 	/* We must not fragment if the socket is set to force MTU discovery
575 	 * or if the skb it not generated by a local socket.
576 	 */
577 	if (unlikely(!skb->ignore_df && skb->len > mtu) ||
578 		     (IP6CB(skb)->frag_max_size &&
579 		      IP6CB(skb)->frag_max_size > mtu)) {
580 		if (skb->sk && dst_allfrag(skb_dst(skb)))
581 			sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
582 
583 		skb->dev = skb_dst(skb)->dev;
584 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
585 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
586 			      IPSTATS_MIB_FRAGFAILS);
587 		kfree_skb(skb);
588 		return -EMSGSIZE;
589 	}
590 
591 	if (np && np->frag_size < mtu) {
592 		if (np->frag_size)
593 			mtu = np->frag_size;
594 	}
595 	mtu -= hlen + sizeof(struct frag_hdr);
596 
597 	if (skb_has_frag_list(skb)) {
598 		int first_len = skb_pagelen(skb);
599 		struct sk_buff *frag2;
600 
601 		if (first_len - hlen > mtu ||
602 		    ((first_len - hlen) & 7) ||
603 		    skb_cloned(skb))
604 			goto slow_path;
605 
606 		skb_walk_frags(skb, frag) {
607 			/* Correct geometry. */
608 			if (frag->len > mtu ||
609 			    ((frag->len & 7) && frag->next) ||
610 			    skb_headroom(frag) < hlen)
611 				goto slow_path_clean;
612 
613 			/* Partially cloned skb? */
614 			if (skb_shared(frag))
615 				goto slow_path_clean;
616 
617 			BUG_ON(frag->sk);
618 			if (skb->sk) {
619 				frag->sk = skb->sk;
620 				frag->destructor = sock_wfree;
621 			}
622 			skb->truesize -= frag->truesize;
623 		}
624 
625 		err = 0;
626 		offset = 0;
627 		frag = skb_shinfo(skb)->frag_list;
628 		skb_frag_list_init(skb);
629 		/* BUILD HEADER */
630 
631 		*prevhdr = NEXTHDR_FRAGMENT;
632 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
633 		if (!tmp_hdr) {
634 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
635 				      IPSTATS_MIB_FRAGFAILS);
636 			return -ENOMEM;
637 		}
638 
639 		__skb_pull(skb, hlen);
640 		fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
641 		__skb_push(skb, hlen);
642 		skb_reset_network_header(skb);
643 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
644 
645 		ipv6_select_ident(fh, rt);
646 		fh->nexthdr = nexthdr;
647 		fh->reserved = 0;
648 		fh->frag_off = htons(IP6_MF);
649 		frag_id = fh->identification;
650 
651 		first_len = skb_pagelen(skb);
652 		skb->data_len = first_len - skb_headlen(skb);
653 		skb->len = first_len;
654 		ipv6_hdr(skb)->payload_len = htons(first_len -
655 						   sizeof(struct ipv6hdr));
656 
657 		dst_hold(&rt->dst);
658 
659 		for (;;) {
660 			/* Prepare header of the next frame,
661 			 * before previous one went down. */
662 			if (frag) {
663 				frag->ip_summed = CHECKSUM_NONE;
664 				skb_reset_transport_header(frag);
665 				fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
666 				__skb_push(frag, hlen);
667 				skb_reset_network_header(frag);
668 				memcpy(skb_network_header(frag), tmp_hdr,
669 				       hlen);
670 				offset += skb->len - hlen - sizeof(struct frag_hdr);
671 				fh->nexthdr = nexthdr;
672 				fh->reserved = 0;
673 				fh->frag_off = htons(offset);
674 				if (frag->next != NULL)
675 					fh->frag_off |= htons(IP6_MF);
676 				fh->identification = frag_id;
677 				ipv6_hdr(frag)->payload_len =
678 						htons(frag->len -
679 						      sizeof(struct ipv6hdr));
680 				ip6_copy_metadata(frag, skb);
681 			}
682 
683 			err = output(skb);
684 			if(!err)
685 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
686 					      IPSTATS_MIB_FRAGCREATES);
687 
688 			if (err || !frag)
689 				break;
690 
691 			skb = frag;
692 			frag = skb->next;
693 			skb->next = NULL;
694 		}
695 
696 		kfree(tmp_hdr);
697 
698 		if (err == 0) {
699 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
700 				      IPSTATS_MIB_FRAGOKS);
701 			ip6_rt_put(rt);
702 			return 0;
703 		}
704 
705 		while (frag) {
706 			skb = frag->next;
707 			kfree_skb(frag);
708 			frag = skb;
709 		}
710 
711 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
712 			      IPSTATS_MIB_FRAGFAILS);
713 		ip6_rt_put(rt);
714 		return err;
715 
716 slow_path_clean:
717 		skb_walk_frags(skb, frag2) {
718 			if (frag2 == frag)
719 				break;
720 			frag2->sk = NULL;
721 			frag2->destructor = NULL;
722 			skb->truesize += frag2->truesize;
723 		}
724 	}
725 
726 slow_path:
727 	if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
728 	    skb_checksum_help(skb))
729 		goto fail;
730 
731 	left = skb->len - hlen;		/* Space per frame */
732 	ptr = hlen;			/* Where to start from */
733 
734 	/*
735 	 *	Fragment the datagram.
736 	 */
737 
738 	*prevhdr = NEXTHDR_FRAGMENT;
739 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
740 	troom = rt->dst.dev->needed_tailroom;
741 
742 	/*
743 	 *	Keep copying data until we run out.
744 	 */
745 	while(left > 0)	{
746 		len = left;
747 		/* IF: it doesn't fit, use 'mtu' - the data space left */
748 		if (len > mtu)
749 			len = mtu;
750 		/* IF: we are not sending up to and including the packet end
751 		   then align the next start on an eight byte boundary */
752 		if (len < left)	{
753 			len &= ~7;
754 		}
755 		/*
756 		 *	Allocate buffer.
757 		 */
758 
759 		if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
760 				      hroom + troom, GFP_ATOMIC)) == NULL) {
761 			NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
762 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
763 				      IPSTATS_MIB_FRAGFAILS);
764 			err = -ENOMEM;
765 			goto fail;
766 		}
767 
768 		/*
769 		 *	Set up data on packet
770 		 */
771 
772 		ip6_copy_metadata(frag, skb);
773 		skb_reserve(frag, hroom);
774 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
775 		skb_reset_network_header(frag);
776 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
777 		frag->transport_header = (frag->network_header + hlen +
778 					  sizeof(struct frag_hdr));
779 
780 		/*
781 		 *	Charge the memory for the fragment to any owner
782 		 *	it might possess
783 		 */
784 		if (skb->sk)
785 			skb_set_owner_w(frag, skb->sk);
786 
787 		/*
788 		 *	Copy the packet header into the new buffer.
789 		 */
790 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
791 
792 		/*
793 		 *	Build fragment header.
794 		 */
795 		fh->nexthdr = nexthdr;
796 		fh->reserved = 0;
797 		if (!frag_id) {
798 			ipv6_select_ident(fh, rt);
799 			frag_id = fh->identification;
800 		} else
801 			fh->identification = frag_id;
802 
803 		/*
804 		 *	Copy a block of the IP datagram.
805 		 */
806 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
807 				     len));
808 		left -= len;
809 
810 		fh->frag_off = htons(offset);
811 		if (left > 0)
812 			fh->frag_off |= htons(IP6_MF);
813 		ipv6_hdr(frag)->payload_len = htons(frag->len -
814 						    sizeof(struct ipv6hdr));
815 
816 		ptr += len;
817 		offset += len;
818 
819 		/*
820 		 *	Put this fragment into the sending queue.
821 		 */
822 		err = output(frag);
823 		if (err)
824 			goto fail;
825 
826 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
827 			      IPSTATS_MIB_FRAGCREATES);
828 	}
829 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
830 		      IPSTATS_MIB_FRAGOKS);
831 	consume_skb(skb);
832 	return err;
833 
834 fail:
835 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
836 		      IPSTATS_MIB_FRAGFAILS);
837 	kfree_skb(skb);
838 	return err;
839 }
840 
841 static inline int ip6_rt_check(const struct rt6key *rt_key,
842 			       const struct in6_addr *fl_addr,
843 			       const struct in6_addr *addr_cache)
844 {
845 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
846 		(addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
847 }
848 
849 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
850 					  struct dst_entry *dst,
851 					  const struct flowi6 *fl6)
852 {
853 	struct ipv6_pinfo *np = inet6_sk(sk);
854 	struct rt6_info *rt;
855 
856 	if (!dst)
857 		goto out;
858 
859 	if (dst->ops->family != AF_INET6) {
860 		dst_release(dst);
861 		return NULL;
862 	}
863 
864 	rt = (struct rt6_info *)dst;
865 	/* Yes, checking route validity in not connected
866 	 * case is not very simple. Take into account,
867 	 * that we do not support routing by source, TOS,
868 	 * and MSG_DONTROUTE 		--ANK (980726)
869 	 *
870 	 * 1. ip6_rt_check(): If route was host route,
871 	 *    check that cached destination is current.
872 	 *    If it is network route, we still may
873 	 *    check its validity using saved pointer
874 	 *    to the last used address: daddr_cache.
875 	 *    We do not want to save whole address now,
876 	 *    (because main consumer of this service
877 	 *    is tcp, which has not this problem),
878 	 *    so that the last trick works only on connected
879 	 *    sockets.
880 	 * 2. oif also should be the same.
881 	 */
882 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
883 #ifdef CONFIG_IPV6_SUBTREES
884 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
885 #endif
886 	    (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
887 		dst_release(dst);
888 		dst = NULL;
889 	}
890 
891 out:
892 	return dst;
893 }
894 
895 static int ip6_dst_lookup_tail(struct sock *sk,
896 			       struct dst_entry **dst, struct flowi6 *fl6)
897 {
898 	struct net *net = sock_net(sk);
899 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
900 	struct neighbour *n;
901 	struct rt6_info *rt;
902 #endif
903 	int err;
904 
905 	if (*dst == NULL)
906 		*dst = ip6_route_output(net, sk, fl6);
907 
908 	if ((err = (*dst)->error))
909 		goto out_err_release;
910 
911 	if (ipv6_addr_any(&fl6->saddr)) {
912 		struct rt6_info *rt = (struct rt6_info *) *dst;
913 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
914 					  sk ? inet6_sk(sk)->srcprefs : 0,
915 					  &fl6->saddr);
916 		if (err)
917 			goto out_err_release;
918 	}
919 
920 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
921 	/*
922 	 * Here if the dst entry we've looked up
923 	 * has a neighbour entry that is in the INCOMPLETE
924 	 * state and the src address from the flow is
925 	 * marked as OPTIMISTIC, we release the found
926 	 * dst entry and replace it instead with the
927 	 * dst entry of the nexthop router
928 	 */
929 	rt = (struct rt6_info *) *dst;
930 	rcu_read_lock_bh();
931 	n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
932 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
933 	rcu_read_unlock_bh();
934 
935 	if (err) {
936 		struct inet6_ifaddr *ifp;
937 		struct flowi6 fl_gw6;
938 		int redirect;
939 
940 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
941 				      (*dst)->dev, 1);
942 
943 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
944 		if (ifp)
945 			in6_ifa_put(ifp);
946 
947 		if (redirect) {
948 			/*
949 			 * We need to get the dst entry for the
950 			 * default router instead
951 			 */
952 			dst_release(*dst);
953 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
954 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
955 			*dst = ip6_route_output(net, sk, &fl_gw6);
956 			if ((err = (*dst)->error))
957 				goto out_err_release;
958 		}
959 	}
960 #endif
961 
962 	return 0;
963 
964 out_err_release:
965 	if (err == -ENETUNREACH)
966 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
967 	dst_release(*dst);
968 	*dst = NULL;
969 	return err;
970 }
971 
972 /**
973  *	ip6_dst_lookup - perform route lookup on flow
974  *	@sk: socket which provides route info
975  *	@dst: pointer to dst_entry * for result
976  *	@fl6: flow to lookup
977  *
978  *	This function performs a route lookup on the given flow.
979  *
980  *	It returns zero on success, or a standard errno code on error.
981  */
982 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
983 {
984 	*dst = NULL;
985 	return ip6_dst_lookup_tail(sk, dst, fl6);
986 }
987 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
988 
989 /**
990  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
991  *	@sk: socket which provides route info
992  *	@fl6: flow to lookup
993  *	@final_dst: final destination address for ipsec lookup
994  *
995  *	This function performs a route lookup on the given flow.
996  *
997  *	It returns a valid dst pointer on success, or a pointer encoded
998  *	error code.
999  */
1000 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1001 				      const struct in6_addr *final_dst)
1002 {
1003 	struct dst_entry *dst = NULL;
1004 	int err;
1005 
1006 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
1007 	if (err)
1008 		return ERR_PTR(err);
1009 	if (final_dst)
1010 		fl6->daddr = *final_dst;
1011 
1012 	return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1013 }
1014 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1015 
1016 /**
1017  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1018  *	@sk: socket which provides the dst cache and route info
1019  *	@fl6: flow to lookup
1020  *	@final_dst: final destination address for ipsec lookup
1021  *
1022  *	This function performs a route lookup on the given flow with the
1023  *	possibility of using the cached route in the socket if it is valid.
1024  *	It will take the socket dst lock when operating on the dst cache.
1025  *	As a result, this function can only be used in process context.
1026  *
1027  *	It returns a valid dst pointer on success, or a pointer encoded
1028  *	error code.
1029  */
1030 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1031 					 const struct in6_addr *final_dst)
1032 {
1033 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1034 	int err;
1035 
1036 	dst = ip6_sk_dst_check(sk, dst, fl6);
1037 
1038 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
1039 	if (err)
1040 		return ERR_PTR(err);
1041 	if (final_dst)
1042 		fl6->daddr = *final_dst;
1043 
1044 	return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1045 }
1046 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1047 
1048 static inline int ip6_ufo_append_data(struct sock *sk,
1049 			int getfrag(void *from, char *to, int offset, int len,
1050 			int odd, struct sk_buff *skb),
1051 			void *from, int length, int hh_len, int fragheaderlen,
1052 			int transhdrlen, int mtu,unsigned int flags,
1053 			struct rt6_info *rt)
1054 
1055 {
1056 	struct sk_buff *skb;
1057 	struct frag_hdr fhdr;
1058 	int err;
1059 
1060 	/* There is support for UDP large send offload by network
1061 	 * device, so create one single skb packet containing complete
1062 	 * udp datagram
1063 	 */
1064 	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1065 		skb = sock_alloc_send_skb(sk,
1066 			hh_len + fragheaderlen + transhdrlen + 20,
1067 			(flags & MSG_DONTWAIT), &err);
1068 		if (skb == NULL)
1069 			return err;
1070 
1071 		/* reserve space for Hardware header */
1072 		skb_reserve(skb, hh_len);
1073 
1074 		/* create space for UDP/IP header */
1075 		skb_put(skb,fragheaderlen + transhdrlen);
1076 
1077 		/* initialize network header pointer */
1078 		skb_reset_network_header(skb);
1079 
1080 		/* initialize protocol header pointer */
1081 		skb->transport_header = skb->network_header + fragheaderlen;
1082 
1083 		skb->protocol = htons(ETH_P_IPV6);
1084 		skb->csum = 0;
1085 
1086 		__skb_queue_tail(&sk->sk_write_queue, skb);
1087 	} else if (skb_is_gso(skb)) {
1088 		goto append;
1089 	}
1090 
1091 	skb->ip_summed = CHECKSUM_PARTIAL;
1092 	/* Specify the length of each IPv6 datagram fragment.
1093 	 * It has to be a multiple of 8.
1094 	 */
1095 	skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1096 				     sizeof(struct frag_hdr)) & ~7;
1097 	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1098 	ipv6_select_ident(&fhdr, rt);
1099 	skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1100 
1101 append:
1102 	return skb_append_datato_frags(sk, skb, getfrag, from,
1103 				       (length - transhdrlen));
1104 }
1105 
1106 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1107 					       gfp_t gfp)
1108 {
1109 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1110 }
1111 
1112 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1113 						gfp_t gfp)
1114 {
1115 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1116 }
1117 
1118 static void ip6_append_data_mtu(unsigned int *mtu,
1119 				int *maxfraglen,
1120 				unsigned int fragheaderlen,
1121 				struct sk_buff *skb,
1122 				struct rt6_info *rt,
1123 				unsigned int orig_mtu)
1124 {
1125 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1126 		if (skb == NULL) {
1127 			/* first fragment, reserve header_len */
1128 			*mtu = orig_mtu - rt->dst.header_len;
1129 
1130 		} else {
1131 			/*
1132 			 * this fragment is not first, the headers
1133 			 * space is regarded as data space.
1134 			 */
1135 			*mtu = orig_mtu;
1136 		}
1137 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1138 			      + fragheaderlen - sizeof(struct frag_hdr);
1139 	}
1140 }
1141 
1142 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1143 	int offset, int len, int odd, struct sk_buff *skb),
1144 	void *from, int length, int transhdrlen,
1145 	int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1146 	struct rt6_info *rt, unsigned int flags, int dontfrag)
1147 {
1148 	struct inet_sock *inet = inet_sk(sk);
1149 	struct ipv6_pinfo *np = inet6_sk(sk);
1150 	struct inet_cork *cork;
1151 	struct sk_buff *skb, *skb_prev = NULL;
1152 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1153 	int exthdrlen;
1154 	int dst_exthdrlen;
1155 	int hh_len;
1156 	int copy;
1157 	int err;
1158 	int offset = 0;
1159 	__u8 tx_flags = 0;
1160 	u32 tskey = 0;
1161 
1162 	if (flags&MSG_PROBE)
1163 		return 0;
1164 	cork = &inet->cork.base;
1165 	if (skb_queue_empty(&sk->sk_write_queue)) {
1166 		/*
1167 		 * setup for corking
1168 		 */
1169 		if (opt) {
1170 			if (WARN_ON(np->cork.opt))
1171 				return -EINVAL;
1172 
1173 			np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1174 			if (unlikely(np->cork.opt == NULL))
1175 				return -ENOBUFS;
1176 
1177 			np->cork.opt->tot_len = opt->tot_len;
1178 			np->cork.opt->opt_flen = opt->opt_flen;
1179 			np->cork.opt->opt_nflen = opt->opt_nflen;
1180 
1181 			np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1182 							    sk->sk_allocation);
1183 			if (opt->dst0opt && !np->cork.opt->dst0opt)
1184 				return -ENOBUFS;
1185 
1186 			np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1187 							    sk->sk_allocation);
1188 			if (opt->dst1opt && !np->cork.opt->dst1opt)
1189 				return -ENOBUFS;
1190 
1191 			np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1192 							   sk->sk_allocation);
1193 			if (opt->hopopt && !np->cork.opt->hopopt)
1194 				return -ENOBUFS;
1195 
1196 			np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1197 							    sk->sk_allocation);
1198 			if (opt->srcrt && !np->cork.opt->srcrt)
1199 				return -ENOBUFS;
1200 
1201 			/* need source address above miyazawa*/
1202 		}
1203 		dst_hold(&rt->dst);
1204 		cork->dst = &rt->dst;
1205 		inet->cork.fl.u.ip6 = *fl6;
1206 		np->cork.hop_limit = hlimit;
1207 		np->cork.tclass = tclass;
1208 		if (rt->dst.flags & DST_XFRM_TUNNEL)
1209 			mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1210 			      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1211 		else
1212 			mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1213 			      rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1214 		if (np->frag_size < mtu) {
1215 			if (np->frag_size)
1216 				mtu = np->frag_size;
1217 		}
1218 		cork->fragsize = mtu;
1219 		if (dst_allfrag(rt->dst.path))
1220 			cork->flags |= IPCORK_ALLFRAG;
1221 		cork->length = 0;
1222 		exthdrlen = (opt ? opt->opt_flen : 0);
1223 		length += exthdrlen;
1224 		transhdrlen += exthdrlen;
1225 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1226 	} else {
1227 		rt = (struct rt6_info *)cork->dst;
1228 		fl6 = &inet->cork.fl.u.ip6;
1229 		opt = np->cork.opt;
1230 		transhdrlen = 0;
1231 		exthdrlen = 0;
1232 		dst_exthdrlen = 0;
1233 		mtu = cork->fragsize;
1234 	}
1235 	orig_mtu = mtu;
1236 
1237 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1238 
1239 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1240 			(opt ? opt->opt_nflen : 0);
1241 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1242 		     sizeof(struct frag_hdr);
1243 
1244 	if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1245 		unsigned int maxnonfragsize, headersize;
1246 
1247 		headersize = sizeof(struct ipv6hdr) +
1248 			     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1249 			     (dst_allfrag(&rt->dst) ?
1250 			      sizeof(struct frag_hdr) : 0) +
1251 			     rt->rt6i_nfheader_len;
1252 
1253 		if (ip6_sk_ignore_df(sk))
1254 			maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1255 		else
1256 			maxnonfragsize = mtu;
1257 
1258 		/* dontfrag active */
1259 		if ((cork->length + length > mtu - headersize) && dontfrag &&
1260 		    (sk->sk_protocol == IPPROTO_UDP ||
1261 		     sk->sk_protocol == IPPROTO_RAW)) {
1262 			ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1263 						   sizeof(struct ipv6hdr));
1264 			goto emsgsize;
1265 		}
1266 
1267 		if (cork->length + length > maxnonfragsize - headersize) {
1268 emsgsize:
1269 			ipv6_local_error(sk, EMSGSIZE, fl6,
1270 					 mtu - headersize +
1271 					 sizeof(struct ipv6hdr));
1272 			return -EMSGSIZE;
1273 		}
1274 	}
1275 
1276 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1277 		sock_tx_timestamp(sk, &tx_flags);
1278 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1279 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1280 			tskey = sk->sk_tskey++;
1281 	}
1282 
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 	skb = skb_peek_tail(&sk->sk_write_queue);
1300 	cork->length += length;
1301 	if (((length > mtu) ||
1302 	     (skb && skb_is_gso(skb))) &&
1303 	    (sk->sk_protocol == IPPROTO_UDP) &&
1304 	    (rt->dst.dev->features & NETIF_F_UFO)) {
1305 		err = ip6_ufo_append_data(sk, 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 = CHECKSUM_NONE;
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(&sk->sk_write_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 			struct page_frag *pfrag = sk_page_frag(sk);
1466 
1467 			err = -ENOMEM;
1468 			if (!sk_page_frag_refill(sk, pfrag))
1469 				goto error;
1470 
1471 			if (!skb_can_coalesce(skb, i, pfrag->page,
1472 					      pfrag->offset)) {
1473 				err = -EMSGSIZE;
1474 				if (i == MAX_SKB_FRAGS)
1475 					goto error;
1476 
1477 				__skb_fill_page_desc(skb, i, pfrag->page,
1478 						     pfrag->offset, 0);
1479 				skb_shinfo(skb)->nr_frags = ++i;
1480 				get_page(pfrag->page);
1481 			}
1482 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1483 			if (getfrag(from,
1484 				    page_address(pfrag->page) + pfrag->offset,
1485 				    offset, copy, skb->len, skb) < 0)
1486 				goto error_efault;
1487 
1488 			pfrag->offset += copy;
1489 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1490 			skb->len += copy;
1491 			skb->data_len += copy;
1492 			skb->truesize += copy;
1493 			atomic_add(copy, &sk->sk_wmem_alloc);
1494 		}
1495 		offset += copy;
1496 		length -= copy;
1497 	}
1498 
1499 	return 0;
1500 
1501 error_efault:
1502 	err = -EFAULT;
1503 error:
1504 	cork->length -= length;
1505 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1506 	return err;
1507 }
1508 EXPORT_SYMBOL_GPL(ip6_append_data);
1509 
1510 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1511 {
1512 	if (np->cork.opt) {
1513 		kfree(np->cork.opt->dst0opt);
1514 		kfree(np->cork.opt->dst1opt);
1515 		kfree(np->cork.opt->hopopt);
1516 		kfree(np->cork.opt->srcrt);
1517 		kfree(np->cork.opt);
1518 		np->cork.opt = NULL;
1519 	}
1520 
1521 	if (inet->cork.base.dst) {
1522 		dst_release(inet->cork.base.dst);
1523 		inet->cork.base.dst = NULL;
1524 		inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1525 	}
1526 	memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1527 }
1528 
1529 int ip6_push_pending_frames(struct sock *sk)
1530 {
1531 	struct sk_buff *skb, *tmp_skb;
1532 	struct sk_buff **tail_skb;
1533 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1534 	struct inet_sock *inet = inet_sk(sk);
1535 	struct ipv6_pinfo *np = inet6_sk(sk);
1536 	struct net *net = sock_net(sk);
1537 	struct ipv6hdr *hdr;
1538 	struct ipv6_txoptions *opt = np->cork.opt;
1539 	struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1540 	struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1541 	unsigned char proto = fl6->flowi6_proto;
1542 	int err = 0;
1543 
1544 	if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1545 		goto out;
1546 	tail_skb = &(skb_shinfo(skb)->frag_list);
1547 
1548 	/* move skb->data to ip header from ext header */
1549 	if (skb->data < skb_network_header(skb))
1550 		__skb_pull(skb, skb_network_offset(skb));
1551 	while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1552 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1553 		*tail_skb = tmp_skb;
1554 		tail_skb = &(tmp_skb->next);
1555 		skb->len += tmp_skb->len;
1556 		skb->data_len += tmp_skb->len;
1557 		skb->truesize += tmp_skb->truesize;
1558 		tmp_skb->destructor = NULL;
1559 		tmp_skb->sk = NULL;
1560 	}
1561 
1562 	/* Allow local fragmentation. */
1563 	skb->ignore_df = ip6_sk_ignore_df(sk);
1564 
1565 	*final_dst = fl6->daddr;
1566 	__skb_pull(skb, skb_network_header_len(skb));
1567 	if (opt && opt->opt_flen)
1568 		ipv6_push_frag_opts(skb, opt, &proto);
1569 	if (opt && opt->opt_nflen)
1570 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1571 
1572 	skb_push(skb, sizeof(struct ipv6hdr));
1573 	skb_reset_network_header(skb);
1574 	hdr = ipv6_hdr(skb);
1575 
1576 	ip6_flow_hdr(hdr, np->cork.tclass,
1577 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1578 					np->autoflowlabel));
1579 	hdr->hop_limit = np->cork.hop_limit;
1580 	hdr->nexthdr = proto;
1581 	hdr->saddr = fl6->saddr;
1582 	hdr->daddr = *final_dst;
1583 
1584 	skb->priority = sk->sk_priority;
1585 	skb->mark = sk->sk_mark;
1586 
1587 	skb_dst_set(skb, dst_clone(&rt->dst));
1588 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1589 	if (proto == IPPROTO_ICMPV6) {
1590 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1591 
1592 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1593 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1594 	}
1595 
1596 	err = ip6_local_out(skb);
1597 	if (err) {
1598 		if (err > 0)
1599 			err = net_xmit_errno(err);
1600 		if (err)
1601 			goto error;
1602 	}
1603 
1604 out:
1605 	ip6_cork_release(inet, np);
1606 	return err;
1607 error:
1608 	IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1609 	goto out;
1610 }
1611 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1612 
1613 void ip6_flush_pending_frames(struct sock *sk)
1614 {
1615 	struct sk_buff *skb;
1616 
1617 	while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1618 		if (skb_dst(skb))
1619 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1620 				      IPSTATS_MIB_OUTDISCARDS);
1621 		kfree_skb(skb);
1622 	}
1623 
1624 	ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1625 }
1626 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1627