xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 366e41d9774d7010cb63112b6db2fce6dc7809c0)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/net/ipv4/ip_output.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  *
15  *	Changes:
16  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
17  *				extension headers are implemented.
18  *				route changes now work.
19  *				ip6_forward does not confuse sniffers.
20  *				etc.
21  *
22  *      H. von Brand    :       Added missing #include <linux/string.h>
23  *	Imran Patel	:	frag id should be in NBO
24  *      Kazunori MIYAZAWA @USAGI
25  *			:       add ip6_append_data and related functions
26  *				for datagram xmit
27  */
28 
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
41 
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
44 
45 #include <net/sock.h>
46 #include <net/snmp.h>
47 
48 #include <net/ipv6.h>
49 #include <net/ndisc.h>
50 #include <net/protocol.h>
51 #include <net/ip6_route.h>
52 #include <net/addrconf.h>
53 #include <net/rawv6.h>
54 #include <net/icmp.h>
55 #include <net/xfrm.h>
56 #include <net/checksum.h>
57 #include <linux/mroute6.h>
58 
59 static int ip6_finish_output2(struct sk_buff *skb)
60 {
61 	struct dst_entry *dst = skb_dst(skb);
62 	struct net_device *dev = dst->dev;
63 	struct neighbour *neigh;
64 	struct in6_addr *nexthop;
65 	int ret;
66 
67 	skb->protocol = htons(ETH_P_IPV6);
68 	skb->dev = dev;
69 
70 	if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
72 
73 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
74 		    ((mroute6_socket(dev_net(dev), skb) &&
75 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 		     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 					 &ipv6_hdr(skb)->saddr))) {
78 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
79 
80 			/* Do not check for IFF_ALLMULTI; multicast routing
81 			   is not supported in any case.
82 			 */
83 			if (newskb)
84 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 					newskb, NULL, newskb->dev,
86 					dev_loopback_xmit);
87 
88 			if (ipv6_hdr(skb)->hop_limit == 0) {
89 				IP6_INC_STATS(dev_net(dev), idev,
90 					      IPSTATS_MIB_OUTDISCARDS);
91 				kfree_skb(skb);
92 				return 0;
93 			}
94 		}
95 
96 		IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
97 				skb->len);
98 
99 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 		    IPV6_ADDR_SCOPE_NODELOCAL &&
101 		    !(dev->flags & IFF_LOOPBACK)) {
102 			kfree_skb(skb);
103 			return 0;
104 		}
105 	}
106 
107 	rcu_read_lock_bh();
108 	nexthop = rt6_nexthop((struct rt6_info *)dst);
109 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 	if (unlikely(!neigh))
111 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 	if (!IS_ERR(neigh)) {
113 		ret = dst_neigh_output(dst, neigh, skb);
114 		rcu_read_unlock_bh();
115 		return ret;
116 	}
117 	rcu_read_unlock_bh();
118 
119 	IP6_INC_STATS(dev_net(dst->dev),
120 		      ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
121 	kfree_skb(skb);
122 	return -EINVAL;
123 }
124 
125 static int ip6_finish_output(struct sk_buff *skb)
126 {
127 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 	    dst_allfrag(skb_dst(skb)) ||
129 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 		return ip6_fragment(skb, ip6_finish_output2);
131 	else
132 		return ip6_finish_output2(skb);
133 }
134 
135 int ip6_output(struct sock *sk, struct sk_buff *skb)
136 {
137 	struct net_device *dev = skb_dst(skb)->dev;
138 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 	if (unlikely(idev->cnf.disable_ipv6)) {
140 		IP6_INC_STATS(dev_net(dev), idev,
141 			      IPSTATS_MIB_OUTDISCARDS);
142 		kfree_skb(skb);
143 		return 0;
144 	}
145 
146 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
147 			    ip6_finish_output,
148 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
149 }
150 
151 /*
152  *	xmit an sk_buff (used by TCP, SCTP and DCCP)
153  */
154 
155 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
156 	     struct ipv6_txoptions *opt, int tclass)
157 {
158 	struct net *net = sock_net(sk);
159 	struct ipv6_pinfo *np = inet6_sk(sk);
160 	struct in6_addr *first_hop = &fl6->daddr;
161 	struct dst_entry *dst = skb_dst(skb);
162 	struct ipv6hdr *hdr;
163 	u8  proto = fl6->flowi6_proto;
164 	int seg_len = skb->len;
165 	int hlimit = -1;
166 	u32 mtu;
167 
168 	if (opt) {
169 		unsigned int head_room;
170 
171 		/* First: exthdrs may take lots of space (~8K for now)
172 		   MAX_HEADER is not enough.
173 		 */
174 		head_room = opt->opt_nflen + opt->opt_flen;
175 		seg_len += head_room;
176 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
177 
178 		if (skb_headroom(skb) < head_room) {
179 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
180 			if (skb2 == NULL) {
181 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
182 					      IPSTATS_MIB_OUTDISCARDS);
183 				kfree_skb(skb);
184 				return -ENOBUFS;
185 			}
186 			consume_skb(skb);
187 			skb = skb2;
188 			skb_set_owner_w(skb, sk);
189 		}
190 		if (opt->opt_flen)
191 			ipv6_push_frag_opts(skb, opt, &proto);
192 		if (opt->opt_nflen)
193 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
194 	}
195 
196 	skb_push(skb, sizeof(struct ipv6hdr));
197 	skb_reset_network_header(skb);
198 	hdr = ipv6_hdr(skb);
199 
200 	/*
201 	 *	Fill in the IPv6 header
202 	 */
203 	if (np)
204 		hlimit = np->hop_limit;
205 	if (hlimit < 0)
206 		hlimit = ip6_dst_hoplimit(dst);
207 
208 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
209 						     np->autoflowlabel));
210 
211 	hdr->payload_len = htons(seg_len);
212 	hdr->nexthdr = proto;
213 	hdr->hop_limit = hlimit;
214 
215 	hdr->saddr = fl6->saddr;
216 	hdr->daddr = *first_hop;
217 
218 	skb->protocol = htons(ETH_P_IPV6);
219 	skb->priority = sk->sk_priority;
220 	skb->mark = sk->sk_mark;
221 
222 	mtu = dst_mtu(dst);
223 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
224 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
225 			      IPSTATS_MIB_OUT, skb->len);
226 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
227 			       dst->dev, dst_output);
228 	}
229 
230 	skb->dev = dst->dev;
231 	ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
232 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
233 	kfree_skb(skb);
234 	return -EMSGSIZE;
235 }
236 EXPORT_SYMBOL(ip6_xmit);
237 
238 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
239 {
240 	struct ip6_ra_chain *ra;
241 	struct sock *last = NULL;
242 
243 	read_lock(&ip6_ra_lock);
244 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
245 		struct sock *sk = ra->sk;
246 		if (sk && ra->sel == sel &&
247 		    (!sk->sk_bound_dev_if ||
248 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
249 			if (last) {
250 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
251 				if (skb2)
252 					rawv6_rcv(last, skb2);
253 			}
254 			last = sk;
255 		}
256 	}
257 
258 	if (last) {
259 		rawv6_rcv(last, skb);
260 		read_unlock(&ip6_ra_lock);
261 		return 1;
262 	}
263 	read_unlock(&ip6_ra_lock);
264 	return 0;
265 }
266 
267 static int ip6_forward_proxy_check(struct sk_buff *skb)
268 {
269 	struct ipv6hdr *hdr = ipv6_hdr(skb);
270 	u8 nexthdr = hdr->nexthdr;
271 	__be16 frag_off;
272 	int offset;
273 
274 	if (ipv6_ext_hdr(nexthdr)) {
275 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
276 		if (offset < 0)
277 			return 0;
278 	} else
279 		offset = sizeof(struct ipv6hdr);
280 
281 	if (nexthdr == IPPROTO_ICMPV6) {
282 		struct icmp6hdr *icmp6;
283 
284 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
285 					 offset + 1 - skb->data)))
286 			return 0;
287 
288 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
289 
290 		switch (icmp6->icmp6_type) {
291 		case NDISC_ROUTER_SOLICITATION:
292 		case NDISC_ROUTER_ADVERTISEMENT:
293 		case NDISC_NEIGHBOUR_SOLICITATION:
294 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
295 		case NDISC_REDIRECT:
296 			/* For reaction involving unicast neighbor discovery
297 			 * message destined to the proxied address, pass it to
298 			 * input function.
299 			 */
300 			return 1;
301 		default:
302 			break;
303 		}
304 	}
305 
306 	/*
307 	 * The proxying router can't forward traffic sent to a link-local
308 	 * address, so signal the sender and discard the packet. This
309 	 * behavior is clarified by the MIPv6 specification.
310 	 */
311 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
312 		dst_link_failure(skb);
313 		return -1;
314 	}
315 
316 	return 0;
317 }
318 
319 static inline int ip6_forward_finish(struct sk_buff *skb)
320 {
321 	return dst_output(skb);
322 }
323 
324 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
325 {
326 	unsigned int mtu;
327 	struct inet6_dev *idev;
328 
329 	if (dst_metric_locked(dst, RTAX_MTU)) {
330 		mtu = dst_metric_raw(dst, RTAX_MTU);
331 		if (mtu)
332 			return mtu;
333 	}
334 
335 	mtu = IPV6_MIN_MTU;
336 	rcu_read_lock();
337 	idev = __in6_dev_get(dst->dev);
338 	if (idev)
339 		mtu = idev->cnf.mtu6;
340 	rcu_read_unlock();
341 
342 	return mtu;
343 }
344 
345 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
346 {
347 	if (skb->len <= mtu)
348 		return false;
349 
350 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
351 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
352 		return true;
353 
354 	if (skb->ignore_df)
355 		return false;
356 
357 	if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
358 		return false;
359 
360 	return true;
361 }
362 
363 int ip6_forward(struct sk_buff *skb)
364 {
365 	struct dst_entry *dst = skb_dst(skb);
366 	struct ipv6hdr *hdr = ipv6_hdr(skb);
367 	struct inet6_skb_parm *opt = IP6CB(skb);
368 	struct net *net = dev_net(dst->dev);
369 	u32 mtu;
370 
371 	if (net->ipv6.devconf_all->forwarding == 0)
372 		goto error;
373 
374 	if (skb->pkt_type != PACKET_HOST)
375 		goto drop;
376 
377 	if (skb_warn_if_lro(skb))
378 		goto drop;
379 
380 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
382 				 IPSTATS_MIB_INDISCARDS);
383 		goto drop;
384 	}
385 
386 	skb_forward_csum(skb);
387 
388 	/*
389 	 *	We DO NOT make any processing on
390 	 *	RA packets, pushing them to user level AS IS
391 	 *	without ane WARRANTY that application will be able
392 	 *	to interpret them. The reason is that we
393 	 *	cannot make anything clever here.
394 	 *
395 	 *	We are not end-node, so that if packet contains
396 	 *	AH/ESP, we cannot make anything.
397 	 *	Defragmentation also would be mistake, RA packets
398 	 *	cannot be fragmented, because there is no warranty
399 	 *	that different fragments will go along one path. --ANK
400 	 */
401 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
402 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
403 			return 0;
404 	}
405 
406 	/*
407 	 *	check and decrement ttl
408 	 */
409 	if (hdr->hop_limit <= 1) {
410 		/* Force OUTPUT device used as source address */
411 		skb->dev = dst->dev;
412 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
413 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
414 				 IPSTATS_MIB_INHDRERRORS);
415 
416 		kfree_skb(skb);
417 		return -ETIMEDOUT;
418 	}
419 
420 	/* XXX: idev->cnf.proxy_ndp? */
421 	if (net->ipv6.devconf_all->proxy_ndp &&
422 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
423 		int proxied = ip6_forward_proxy_check(skb);
424 		if (proxied > 0)
425 			return ip6_input(skb);
426 		else if (proxied < 0) {
427 			IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
428 					 IPSTATS_MIB_INDISCARDS);
429 			goto drop;
430 		}
431 	}
432 
433 	if (!xfrm6_route_forward(skb)) {
434 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
435 				 IPSTATS_MIB_INDISCARDS);
436 		goto drop;
437 	}
438 	dst = skb_dst(skb);
439 
440 	/* IPv6 specs say nothing about it, but it is clear that we cannot
441 	   send redirects to source routed frames.
442 	   We don't send redirects to frames decapsulated from IPsec.
443 	 */
444 	if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
445 		struct in6_addr *target = NULL;
446 		struct inet_peer *peer;
447 		struct rt6_info *rt;
448 
449 		/*
450 		 *	incoming and outgoing devices are the same
451 		 *	send a redirect.
452 		 */
453 
454 		rt = (struct rt6_info *) dst;
455 		if (rt->rt6i_flags & RTF_GATEWAY)
456 			target = &rt->rt6i_gateway;
457 		else
458 			target = &hdr->daddr;
459 
460 		peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
461 
462 		/* Limit redirects both by destination (here)
463 		   and by source (inside ndisc_send_redirect)
464 		 */
465 		if (inet_peer_xrlim_allow(peer, 1*HZ))
466 			ndisc_send_redirect(skb, target);
467 		if (peer)
468 			inet_putpeer(peer);
469 	} else {
470 		int addrtype = ipv6_addr_type(&hdr->saddr);
471 
472 		/* This check is security critical. */
473 		if (addrtype == IPV6_ADDR_ANY ||
474 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
475 			goto error;
476 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
477 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
478 				    ICMPV6_NOT_NEIGHBOUR, 0);
479 			goto error;
480 		}
481 	}
482 
483 	mtu = ip6_dst_mtu_forward(dst);
484 	if (mtu < IPV6_MIN_MTU)
485 		mtu = IPV6_MIN_MTU;
486 
487 	if (ip6_pkt_too_big(skb, mtu)) {
488 		/* Again, force OUTPUT device used as source address */
489 		skb->dev = dst->dev;
490 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
491 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
492 				 IPSTATS_MIB_INTOOBIGERRORS);
493 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
494 				 IPSTATS_MIB_FRAGFAILS);
495 		kfree_skb(skb);
496 		return -EMSGSIZE;
497 	}
498 
499 	if (skb_cow(skb, dst->dev->hard_header_len)) {
500 		IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
501 				 IPSTATS_MIB_OUTDISCARDS);
502 		goto drop;
503 	}
504 
505 	hdr = ipv6_hdr(skb);
506 
507 	/* Mangling hops number delayed to point after skb COW */
508 
509 	hdr->hop_limit--;
510 
511 	IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 	IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
514 		       ip6_forward_finish);
515 
516 error:
517 	IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
518 drop:
519 	kfree_skb(skb);
520 	return -EINVAL;
521 }
522 
523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
524 {
525 	to->pkt_type = from->pkt_type;
526 	to->priority = from->priority;
527 	to->protocol = from->protocol;
528 	skb_dst_drop(to);
529 	skb_dst_set(to, dst_clone(skb_dst(from)));
530 	to->dev = from->dev;
531 	to->mark = from->mark;
532 
533 #ifdef CONFIG_NET_SCHED
534 	to->tc_index = from->tc_index;
535 #endif
536 	nf_copy(to, from);
537 	skb_copy_secmark(to, from);
538 }
539 
540 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
541 {
542 	static u32 ip6_idents_hashrnd __read_mostly;
543 	u32 hash, id;
544 
545 	net_get_random_once(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
546 
547 	hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
548 	hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
549 
550 	id = ip_idents_reserve(hash, 1);
551 	fhdr->identification = htonl(id);
552 }
553 
554 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
555 {
556 	struct sk_buff *frag;
557 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
558 	struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
559 	struct ipv6hdr *tmp_hdr;
560 	struct frag_hdr *fh;
561 	unsigned int mtu, hlen, left, len;
562 	int hroom, troom;
563 	__be32 frag_id = 0;
564 	int ptr, offset = 0, err = 0;
565 	u8 *prevhdr, nexthdr = 0;
566 	struct net *net = dev_net(skb_dst(skb)->dev);
567 
568 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
569 	nexthdr = *prevhdr;
570 
571 	mtu = ip6_skb_dst_mtu(skb);
572 
573 	/* We must not fragment if the socket is set to force MTU discovery
574 	 * or if the skb it not generated by a local socket.
575 	 */
576 	if (unlikely(!skb->ignore_df && skb->len > mtu) ||
577 		     (IP6CB(skb)->frag_max_size &&
578 		      IP6CB(skb)->frag_max_size > mtu)) {
579 		if (skb->sk && dst_allfrag(skb_dst(skb)))
580 			sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
581 
582 		skb->dev = skb_dst(skb)->dev;
583 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
584 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
585 			      IPSTATS_MIB_FRAGFAILS);
586 		kfree_skb(skb);
587 		return -EMSGSIZE;
588 	}
589 
590 	if (np && np->frag_size < mtu) {
591 		if (np->frag_size)
592 			mtu = np->frag_size;
593 	}
594 	mtu -= hlen + sizeof(struct frag_hdr);
595 
596 	if (skb_has_frag_list(skb)) {
597 		int first_len = skb_pagelen(skb);
598 		struct sk_buff *frag2;
599 
600 		if (first_len - hlen > mtu ||
601 		    ((first_len - hlen) & 7) ||
602 		    skb_cloned(skb))
603 			goto slow_path;
604 
605 		skb_walk_frags(skb, frag) {
606 			/* Correct geometry. */
607 			if (frag->len > mtu ||
608 			    ((frag->len & 7) && frag->next) ||
609 			    skb_headroom(frag) < hlen)
610 				goto slow_path_clean;
611 
612 			/* Partially cloned skb? */
613 			if (skb_shared(frag))
614 				goto slow_path_clean;
615 
616 			BUG_ON(frag->sk);
617 			if (skb->sk) {
618 				frag->sk = skb->sk;
619 				frag->destructor = sock_wfree;
620 			}
621 			skb->truesize -= frag->truesize;
622 		}
623 
624 		err = 0;
625 		offset = 0;
626 		frag = skb_shinfo(skb)->frag_list;
627 		skb_frag_list_init(skb);
628 		/* BUILD HEADER */
629 
630 		*prevhdr = NEXTHDR_FRAGMENT;
631 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
632 		if (!tmp_hdr) {
633 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
634 				      IPSTATS_MIB_FRAGFAILS);
635 			return -ENOMEM;
636 		}
637 
638 		__skb_pull(skb, hlen);
639 		fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
640 		__skb_push(skb, hlen);
641 		skb_reset_network_header(skb);
642 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
643 
644 		ipv6_select_ident(fh, rt);
645 		fh->nexthdr = nexthdr;
646 		fh->reserved = 0;
647 		fh->frag_off = htons(IP6_MF);
648 		frag_id = fh->identification;
649 
650 		first_len = skb_pagelen(skb);
651 		skb->data_len = first_len - skb_headlen(skb);
652 		skb->len = first_len;
653 		ipv6_hdr(skb)->payload_len = htons(first_len -
654 						   sizeof(struct ipv6hdr));
655 
656 		dst_hold(&rt->dst);
657 
658 		for (;;) {
659 			/* Prepare header of the next frame,
660 			 * before previous one went down. */
661 			if (frag) {
662 				frag->ip_summed = CHECKSUM_NONE;
663 				skb_reset_transport_header(frag);
664 				fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
665 				__skb_push(frag, hlen);
666 				skb_reset_network_header(frag);
667 				memcpy(skb_network_header(frag), tmp_hdr,
668 				       hlen);
669 				offset += skb->len - hlen - sizeof(struct frag_hdr);
670 				fh->nexthdr = nexthdr;
671 				fh->reserved = 0;
672 				fh->frag_off = htons(offset);
673 				if (frag->next != NULL)
674 					fh->frag_off |= htons(IP6_MF);
675 				fh->identification = frag_id;
676 				ipv6_hdr(frag)->payload_len =
677 						htons(frag->len -
678 						      sizeof(struct ipv6hdr));
679 				ip6_copy_metadata(frag, skb);
680 			}
681 
682 			err = output(skb);
683 			if (!err)
684 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
685 					      IPSTATS_MIB_FRAGCREATES);
686 
687 			if (err || !frag)
688 				break;
689 
690 			skb = frag;
691 			frag = skb->next;
692 			skb->next = NULL;
693 		}
694 
695 		kfree(tmp_hdr);
696 
697 		if (err == 0) {
698 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
699 				      IPSTATS_MIB_FRAGOKS);
700 			ip6_rt_put(rt);
701 			return 0;
702 		}
703 
704 		kfree_skb_list(frag);
705 
706 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
707 			      IPSTATS_MIB_FRAGFAILS);
708 		ip6_rt_put(rt);
709 		return err;
710 
711 slow_path_clean:
712 		skb_walk_frags(skb, frag2) {
713 			if (frag2 == frag)
714 				break;
715 			frag2->sk = NULL;
716 			frag2->destructor = NULL;
717 			skb->truesize += frag2->truesize;
718 		}
719 	}
720 
721 slow_path:
722 	if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
723 	    skb_checksum_help(skb))
724 		goto fail;
725 
726 	left = skb->len - hlen;		/* Space per frame */
727 	ptr = hlen;			/* Where to start from */
728 
729 	/*
730 	 *	Fragment the datagram.
731 	 */
732 
733 	*prevhdr = NEXTHDR_FRAGMENT;
734 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
735 	troom = rt->dst.dev->needed_tailroom;
736 
737 	/*
738 	 *	Keep copying data until we run out.
739 	 */
740 	while (left > 0)	{
741 		len = left;
742 		/* IF: it doesn't fit, use 'mtu' - the data space left */
743 		if (len > mtu)
744 			len = mtu;
745 		/* IF: we are not sending up to and including the packet end
746 		   then align the next start on an eight byte boundary */
747 		if (len < left)	{
748 			len &= ~7;
749 		}
750 
751 		/* Allocate buffer */
752 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
753 				 hroom + troom, GFP_ATOMIC);
754 		if (!frag) {
755 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
756 				      IPSTATS_MIB_FRAGFAILS);
757 			err = -ENOMEM;
758 			goto fail;
759 		}
760 
761 		/*
762 		 *	Set up data on packet
763 		 */
764 
765 		ip6_copy_metadata(frag, skb);
766 		skb_reserve(frag, hroom);
767 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
768 		skb_reset_network_header(frag);
769 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
770 		frag->transport_header = (frag->network_header + hlen +
771 					  sizeof(struct frag_hdr));
772 
773 		/*
774 		 *	Charge the memory for the fragment to any owner
775 		 *	it might possess
776 		 */
777 		if (skb->sk)
778 			skb_set_owner_w(frag, skb->sk);
779 
780 		/*
781 		 *	Copy the packet header into the new buffer.
782 		 */
783 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
784 
785 		/*
786 		 *	Build fragment header.
787 		 */
788 		fh->nexthdr = nexthdr;
789 		fh->reserved = 0;
790 		if (!frag_id) {
791 			ipv6_select_ident(fh, rt);
792 			frag_id = fh->identification;
793 		} else
794 			fh->identification = frag_id;
795 
796 		/*
797 		 *	Copy a block of the IP datagram.
798 		 */
799 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
800 				     len));
801 		left -= len;
802 
803 		fh->frag_off = htons(offset);
804 		if (left > 0)
805 			fh->frag_off |= htons(IP6_MF);
806 		ipv6_hdr(frag)->payload_len = htons(frag->len -
807 						    sizeof(struct ipv6hdr));
808 
809 		ptr += len;
810 		offset += len;
811 
812 		/*
813 		 *	Put this fragment into the sending queue.
814 		 */
815 		err = output(frag);
816 		if (err)
817 			goto fail;
818 
819 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
820 			      IPSTATS_MIB_FRAGCREATES);
821 	}
822 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
823 		      IPSTATS_MIB_FRAGOKS);
824 	consume_skb(skb);
825 	return err;
826 
827 fail:
828 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
829 		      IPSTATS_MIB_FRAGFAILS);
830 	kfree_skb(skb);
831 	return err;
832 }
833 
834 static inline int ip6_rt_check(const struct rt6key *rt_key,
835 			       const struct in6_addr *fl_addr,
836 			       const struct in6_addr *addr_cache)
837 {
838 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
839 		(addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
840 }
841 
842 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
843 					  struct dst_entry *dst,
844 					  const struct flowi6 *fl6)
845 {
846 	struct ipv6_pinfo *np = inet6_sk(sk);
847 	struct rt6_info *rt;
848 
849 	if (!dst)
850 		goto out;
851 
852 	if (dst->ops->family != AF_INET6) {
853 		dst_release(dst);
854 		return NULL;
855 	}
856 
857 	rt = (struct rt6_info *)dst;
858 	/* Yes, checking route validity in not connected
859 	 * case is not very simple. Take into account,
860 	 * that we do not support routing by source, TOS,
861 	 * and MSG_DONTROUTE		--ANK (980726)
862 	 *
863 	 * 1. ip6_rt_check(): If route was host route,
864 	 *    check that cached destination is current.
865 	 *    If it is network route, we still may
866 	 *    check its validity using saved pointer
867 	 *    to the last used address: daddr_cache.
868 	 *    We do not want to save whole address now,
869 	 *    (because main consumer of this service
870 	 *    is tcp, which has not this problem),
871 	 *    so that the last trick works only on connected
872 	 *    sockets.
873 	 * 2. oif also should be the same.
874 	 */
875 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
876 #ifdef CONFIG_IPV6_SUBTREES
877 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
878 #endif
879 	    (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
880 		dst_release(dst);
881 		dst = NULL;
882 	}
883 
884 out:
885 	return dst;
886 }
887 
888 static int ip6_dst_lookup_tail(struct sock *sk,
889 			       struct dst_entry **dst, struct flowi6 *fl6)
890 {
891 	struct net *net = sock_net(sk);
892 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
893 	struct neighbour *n;
894 	struct rt6_info *rt;
895 #endif
896 	int err;
897 
898 	if (*dst == NULL)
899 		*dst = ip6_route_output(net, sk, fl6);
900 
901 	err = (*dst)->error;
902 	if (err)
903 		goto out_err_release;
904 
905 	if (ipv6_addr_any(&fl6->saddr)) {
906 		struct rt6_info *rt = (struct rt6_info *) *dst;
907 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
908 					  sk ? inet6_sk(sk)->srcprefs : 0,
909 					  &fl6->saddr);
910 		if (err)
911 			goto out_err_release;
912 	}
913 
914 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
915 	/*
916 	 * Here if the dst entry we've looked up
917 	 * has a neighbour entry that is in the INCOMPLETE
918 	 * state and the src address from the flow is
919 	 * marked as OPTIMISTIC, we release the found
920 	 * dst entry and replace it instead with the
921 	 * dst entry of the nexthop router
922 	 */
923 	rt = (struct rt6_info *) *dst;
924 	rcu_read_lock_bh();
925 	n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
926 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
927 	rcu_read_unlock_bh();
928 
929 	if (err) {
930 		struct inet6_ifaddr *ifp;
931 		struct flowi6 fl_gw6;
932 		int redirect;
933 
934 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
935 				      (*dst)->dev, 1);
936 
937 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
938 		if (ifp)
939 			in6_ifa_put(ifp);
940 
941 		if (redirect) {
942 			/*
943 			 * We need to get the dst entry for the
944 			 * default router instead
945 			 */
946 			dst_release(*dst);
947 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
948 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
949 			*dst = ip6_route_output(net, sk, &fl_gw6);
950 			err = (*dst)->error;
951 			if (err)
952 				goto out_err_release;
953 		}
954 	}
955 #endif
956 
957 	return 0;
958 
959 out_err_release:
960 	if (err == -ENETUNREACH)
961 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
962 	dst_release(*dst);
963 	*dst = NULL;
964 	return err;
965 }
966 
967 /**
968  *	ip6_dst_lookup - perform route lookup on flow
969  *	@sk: socket which provides route info
970  *	@dst: pointer to dst_entry * for result
971  *	@fl6: flow to lookup
972  *
973  *	This function performs a route lookup on the given flow.
974  *
975  *	It returns zero on success, or a standard errno code on error.
976  */
977 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
978 {
979 	*dst = NULL;
980 	return ip6_dst_lookup_tail(sk, dst, fl6);
981 }
982 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
983 
984 /**
985  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
986  *	@sk: socket which provides route info
987  *	@fl6: flow to lookup
988  *	@final_dst: final destination address for ipsec lookup
989  *
990  *	This function performs a route lookup on the given flow.
991  *
992  *	It returns a valid dst pointer on success, or a pointer encoded
993  *	error code.
994  */
995 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
996 				      const struct in6_addr *final_dst)
997 {
998 	struct dst_entry *dst = NULL;
999 	int err;
1000 
1001 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
1002 	if (err)
1003 		return ERR_PTR(err);
1004 	if (final_dst)
1005 		fl6->daddr = *final_dst;
1006 
1007 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1008 }
1009 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1010 
1011 /**
1012  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1013  *	@sk: socket which provides the dst cache and route info
1014  *	@fl6: flow to lookup
1015  *	@final_dst: final destination address for ipsec lookup
1016  *
1017  *	This function performs a route lookup on the given flow with the
1018  *	possibility of using the cached route in the socket if it is valid.
1019  *	It will take the socket dst lock when operating on the dst cache.
1020  *	As a result, this function can only be used in process context.
1021  *
1022  *	It returns a valid dst pointer on success, or a pointer encoded
1023  *	error code.
1024  */
1025 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1026 					 const struct in6_addr *final_dst)
1027 {
1028 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1029 	int err;
1030 
1031 	dst = ip6_sk_dst_check(sk, dst, fl6);
1032 
1033 	err = ip6_dst_lookup_tail(sk, &dst, fl6);
1034 	if (err)
1035 		return ERR_PTR(err);
1036 	if (final_dst)
1037 		fl6->daddr = *final_dst;
1038 
1039 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1040 }
1041 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1042 
1043 static inline int ip6_ufo_append_data(struct sock *sk,
1044 			int getfrag(void *from, char *to, int offset, int len,
1045 			int odd, struct sk_buff *skb),
1046 			void *from, int length, int hh_len, int fragheaderlen,
1047 			int transhdrlen, int mtu, unsigned int flags,
1048 			struct rt6_info *rt)
1049 
1050 {
1051 	struct sk_buff *skb;
1052 	struct frag_hdr fhdr;
1053 	int err;
1054 
1055 	/* There is support for UDP large send offload by network
1056 	 * device, so create one single skb packet containing complete
1057 	 * udp datagram
1058 	 */
1059 	skb = skb_peek_tail(&sk->sk_write_queue);
1060 	if (skb == NULL) {
1061 		skb = sock_alloc_send_skb(sk,
1062 			hh_len + fragheaderlen + transhdrlen + 20,
1063 			(flags & MSG_DONTWAIT), &err);
1064 		if (skb == NULL)
1065 			return err;
1066 
1067 		/* reserve space for Hardware header */
1068 		skb_reserve(skb, hh_len);
1069 
1070 		/* create space for UDP/IP header */
1071 		skb_put(skb, fragheaderlen + transhdrlen);
1072 
1073 		/* initialize network header pointer */
1074 		skb_reset_network_header(skb);
1075 
1076 		/* initialize protocol header pointer */
1077 		skb->transport_header = skb->network_header + fragheaderlen;
1078 
1079 		skb->protocol = htons(ETH_P_IPV6);
1080 		skb->csum = 0;
1081 
1082 		__skb_queue_tail(&sk->sk_write_queue, skb);
1083 	} else if (skb_is_gso(skb)) {
1084 		goto append;
1085 	}
1086 
1087 	skb->ip_summed = CHECKSUM_PARTIAL;
1088 	/* Specify the length of each IPv6 datagram fragment.
1089 	 * It has to be a multiple of 8.
1090 	 */
1091 	skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1092 				     sizeof(struct frag_hdr)) & ~7;
1093 	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1094 	ipv6_select_ident(&fhdr, rt);
1095 	skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1096 
1097 append:
1098 	return skb_append_datato_frags(sk, skb, getfrag, from,
1099 				       (length - transhdrlen));
1100 }
1101 
1102 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1103 					       gfp_t gfp)
1104 {
1105 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1106 }
1107 
1108 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1109 						gfp_t gfp)
1110 {
1111 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1112 }
1113 
1114 static void ip6_append_data_mtu(unsigned int *mtu,
1115 				int *maxfraglen,
1116 				unsigned int fragheaderlen,
1117 				struct sk_buff *skb,
1118 				struct rt6_info *rt,
1119 				unsigned int orig_mtu)
1120 {
1121 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1122 		if (skb == NULL) {
1123 			/* first fragment, reserve header_len */
1124 			*mtu = orig_mtu - rt->dst.header_len;
1125 
1126 		} else {
1127 			/*
1128 			 * this fragment is not first, the headers
1129 			 * space is regarded as data space.
1130 			 */
1131 			*mtu = orig_mtu;
1132 		}
1133 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1134 			      + fragheaderlen - sizeof(struct frag_hdr);
1135 	}
1136 }
1137 
1138 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1139 			  struct inet6_cork *v6_cork,
1140 			  int hlimit, int tclass, struct ipv6_txoptions *opt,
1141 			  struct rt6_info *rt, struct flowi6 *fl6)
1142 {
1143 	struct ipv6_pinfo *np = inet6_sk(sk);
1144 	unsigned int mtu;
1145 
1146 	/*
1147 	 * setup for corking
1148 	 */
1149 	if (opt) {
1150 		if (WARN_ON(v6_cork->opt))
1151 			return -EINVAL;
1152 
1153 		v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1154 		if (unlikely(v6_cork->opt == NULL))
1155 			return -ENOBUFS;
1156 
1157 		v6_cork->opt->tot_len = opt->tot_len;
1158 		v6_cork->opt->opt_flen = opt->opt_flen;
1159 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1160 
1161 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1162 						    sk->sk_allocation);
1163 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1164 			return -ENOBUFS;
1165 
1166 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1167 						    sk->sk_allocation);
1168 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1169 			return -ENOBUFS;
1170 
1171 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1172 						   sk->sk_allocation);
1173 		if (opt->hopopt && !v6_cork->opt->hopopt)
1174 			return -ENOBUFS;
1175 
1176 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1177 						    sk->sk_allocation);
1178 		if (opt->srcrt && !v6_cork->opt->srcrt)
1179 			return -ENOBUFS;
1180 
1181 		/* need source address above miyazawa*/
1182 	}
1183 	dst_hold(&rt->dst);
1184 	cork->base.dst = &rt->dst;
1185 	cork->fl.u.ip6 = *fl6;
1186 	v6_cork->hop_limit = hlimit;
1187 	v6_cork->tclass = tclass;
1188 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1189 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1190 		      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1191 	else
1192 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1193 		      rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1194 	if (np->frag_size < mtu) {
1195 		if (np->frag_size)
1196 			mtu = np->frag_size;
1197 	}
1198 	cork->base.fragsize = mtu;
1199 	if (dst_allfrag(rt->dst.path))
1200 		cork->base.flags |= IPCORK_ALLFRAG;
1201 	cork->base.length = 0;
1202 
1203 	return 0;
1204 }
1205 
1206 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1207 	int offset, int len, int odd, struct sk_buff *skb),
1208 	void *from, int length, int transhdrlen,
1209 	int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1210 	struct rt6_info *rt, unsigned int flags, int dontfrag)
1211 {
1212 	struct inet_sock *inet = inet_sk(sk);
1213 	struct ipv6_pinfo *np = inet6_sk(sk);
1214 	struct inet_cork *cork;
1215 	struct sk_buff *skb, *skb_prev = NULL;
1216 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1217 	int exthdrlen;
1218 	int dst_exthdrlen;
1219 	int hh_len;
1220 	int copy;
1221 	int err;
1222 	int offset = 0;
1223 	__u8 tx_flags = 0;
1224 	u32 tskey = 0;
1225 
1226 	if (flags&MSG_PROBE)
1227 		return 0;
1228 	cork = &inet->cork.base;
1229 	if (skb_queue_empty(&sk->sk_write_queue)) {
1230 		/*
1231 		 * setup for corking
1232 		 */
1233 		err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1234 				     tclass, opt, rt, fl6);
1235 		if (err)
1236 			return err;
1237 		exthdrlen = (opt ? opt->opt_flen : 0);
1238 		length += exthdrlen;
1239 		transhdrlen += exthdrlen;
1240 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1241 	} else {
1242 		rt = (struct rt6_info *)cork->dst;
1243 		fl6 = &inet->cork.fl.u.ip6;
1244 		opt = np->cork.opt;
1245 		transhdrlen = 0;
1246 		exthdrlen = 0;
1247 		dst_exthdrlen = 0;
1248 	}
1249 	mtu = cork->fragsize;
1250 	orig_mtu = mtu;
1251 
1252 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1253 
1254 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1255 			(opt ? opt->opt_nflen : 0);
1256 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1257 		     sizeof(struct frag_hdr);
1258 
1259 	if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1260 		unsigned int maxnonfragsize, headersize;
1261 
1262 		headersize = sizeof(struct ipv6hdr) +
1263 			     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1264 			     (dst_allfrag(&rt->dst) ?
1265 			      sizeof(struct frag_hdr) : 0) +
1266 			     rt->rt6i_nfheader_len;
1267 
1268 		if (ip6_sk_ignore_df(sk))
1269 			maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1270 		else
1271 			maxnonfragsize = mtu;
1272 
1273 		/* dontfrag active */
1274 		if ((cork->length + length > mtu - headersize) && dontfrag &&
1275 		    (sk->sk_protocol == IPPROTO_UDP ||
1276 		     sk->sk_protocol == IPPROTO_RAW)) {
1277 			ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1278 						   sizeof(struct ipv6hdr));
1279 			goto emsgsize;
1280 		}
1281 
1282 		if (cork->length + length > maxnonfragsize - headersize) {
1283 emsgsize:
1284 			ipv6_local_error(sk, EMSGSIZE, fl6,
1285 					 mtu - headersize +
1286 					 sizeof(struct ipv6hdr));
1287 			return -EMSGSIZE;
1288 		}
1289 	}
1290 
1291 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1292 		sock_tx_timestamp(sk, &tx_flags);
1293 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1294 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1295 			tskey = sk->sk_tskey++;
1296 	}
1297 
1298 	/*
1299 	 * Let's try using as much space as possible.
1300 	 * Use MTU if total length of the message fits into the MTU.
1301 	 * Otherwise, we need to reserve fragment header and
1302 	 * fragment alignment (= 8-15 octects, in total).
1303 	 *
1304 	 * Note that we may need to "move" the data from the tail of
1305 	 * of the buffer to the new fragment when we split
1306 	 * the message.
1307 	 *
1308 	 * FIXME: It may be fragmented into multiple chunks
1309 	 *        at once if non-fragmentable extension headers
1310 	 *        are too large.
1311 	 * --yoshfuji
1312 	 */
1313 
1314 	skb = skb_peek_tail(&sk->sk_write_queue);
1315 	cork->length += length;
1316 	if (((length > mtu) ||
1317 	     (skb && skb_is_gso(skb))) &&
1318 	    (sk->sk_protocol == IPPROTO_UDP) &&
1319 	    (rt->dst.dev->features & NETIF_F_UFO)) {
1320 		err = ip6_ufo_append_data(sk, getfrag, from, length,
1321 					  hh_len, fragheaderlen,
1322 					  transhdrlen, mtu, flags, rt);
1323 		if (err)
1324 			goto error;
1325 		return 0;
1326 	}
1327 
1328 	if (!skb)
1329 		goto alloc_new_skb;
1330 
1331 	while (length > 0) {
1332 		/* Check if the remaining data fits into current packet. */
1333 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1334 		if (copy < length)
1335 			copy = maxfraglen - skb->len;
1336 
1337 		if (copy <= 0) {
1338 			char *data;
1339 			unsigned int datalen;
1340 			unsigned int fraglen;
1341 			unsigned int fraggap;
1342 			unsigned int alloclen;
1343 alloc_new_skb:
1344 			/* There's no room in the current skb */
1345 			if (skb)
1346 				fraggap = skb->len - maxfraglen;
1347 			else
1348 				fraggap = 0;
1349 			/* update mtu and maxfraglen if necessary */
1350 			if (skb == NULL || skb_prev == NULL)
1351 				ip6_append_data_mtu(&mtu, &maxfraglen,
1352 						    fragheaderlen, skb, rt,
1353 						    orig_mtu);
1354 
1355 			skb_prev = skb;
1356 
1357 			/*
1358 			 * If remaining data exceeds the mtu,
1359 			 * we know we need more fragment(s).
1360 			 */
1361 			datalen = length + fraggap;
1362 
1363 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1364 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1365 			if ((flags & MSG_MORE) &&
1366 			    !(rt->dst.dev->features&NETIF_F_SG))
1367 				alloclen = mtu;
1368 			else
1369 				alloclen = datalen + fragheaderlen;
1370 
1371 			alloclen += dst_exthdrlen;
1372 
1373 			if (datalen != length + fraggap) {
1374 				/*
1375 				 * this is not the last fragment, the trailer
1376 				 * space is regarded as data space.
1377 				 */
1378 				datalen += rt->dst.trailer_len;
1379 			}
1380 
1381 			alloclen += rt->dst.trailer_len;
1382 			fraglen = datalen + fragheaderlen;
1383 
1384 			/*
1385 			 * We just reserve space for fragment header.
1386 			 * Note: this may be overallocation if the message
1387 			 * (without MSG_MORE) fits into the MTU.
1388 			 */
1389 			alloclen += sizeof(struct frag_hdr);
1390 
1391 			if (transhdrlen) {
1392 				skb = sock_alloc_send_skb(sk,
1393 						alloclen + hh_len,
1394 						(flags & MSG_DONTWAIT), &err);
1395 			} else {
1396 				skb = NULL;
1397 				if (atomic_read(&sk->sk_wmem_alloc) <=
1398 				    2 * sk->sk_sndbuf)
1399 					skb = sock_wmalloc(sk,
1400 							   alloclen + hh_len, 1,
1401 							   sk->sk_allocation);
1402 				if (unlikely(skb == NULL))
1403 					err = -ENOBUFS;
1404 			}
1405 			if (skb == NULL)
1406 				goto error;
1407 			/*
1408 			 *	Fill in the control structures
1409 			 */
1410 			skb->protocol = htons(ETH_P_IPV6);
1411 			skb->ip_summed = CHECKSUM_NONE;
1412 			skb->csum = 0;
1413 			/* reserve for fragmentation and ipsec header */
1414 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1415 				    dst_exthdrlen);
1416 
1417 			/* Only the initial fragment is time stamped */
1418 			skb_shinfo(skb)->tx_flags = tx_flags;
1419 			tx_flags = 0;
1420 			skb_shinfo(skb)->tskey = tskey;
1421 			tskey = 0;
1422 
1423 			/*
1424 			 *	Find where to start putting bytes
1425 			 */
1426 			data = skb_put(skb, fraglen);
1427 			skb_set_network_header(skb, exthdrlen);
1428 			data += fragheaderlen;
1429 			skb->transport_header = (skb->network_header +
1430 						 fragheaderlen);
1431 			if (fraggap) {
1432 				skb->csum = skb_copy_and_csum_bits(
1433 					skb_prev, maxfraglen,
1434 					data + transhdrlen, fraggap, 0);
1435 				skb_prev->csum = csum_sub(skb_prev->csum,
1436 							  skb->csum);
1437 				data += fraggap;
1438 				pskb_trim_unique(skb_prev, maxfraglen);
1439 			}
1440 			copy = datalen - transhdrlen - fraggap;
1441 
1442 			if (copy < 0) {
1443 				err = -EINVAL;
1444 				kfree_skb(skb);
1445 				goto error;
1446 			} else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1447 				err = -EFAULT;
1448 				kfree_skb(skb);
1449 				goto error;
1450 			}
1451 
1452 			offset += copy;
1453 			length -= datalen - fraggap;
1454 			transhdrlen = 0;
1455 			exthdrlen = 0;
1456 			dst_exthdrlen = 0;
1457 
1458 			/*
1459 			 * Put the packet on the pending queue
1460 			 */
1461 			__skb_queue_tail(&sk->sk_write_queue, skb);
1462 			continue;
1463 		}
1464 
1465 		if (copy > length)
1466 			copy = length;
1467 
1468 		if (!(rt->dst.dev->features&NETIF_F_SG)) {
1469 			unsigned int off;
1470 
1471 			off = skb->len;
1472 			if (getfrag(from, skb_put(skb, copy),
1473 						offset, copy, off, skb) < 0) {
1474 				__skb_trim(skb, off);
1475 				err = -EFAULT;
1476 				goto error;
1477 			}
1478 		} else {
1479 			int i = skb_shinfo(skb)->nr_frags;
1480 			struct page_frag *pfrag = sk_page_frag(sk);
1481 
1482 			err = -ENOMEM;
1483 			if (!sk_page_frag_refill(sk, pfrag))
1484 				goto error;
1485 
1486 			if (!skb_can_coalesce(skb, i, pfrag->page,
1487 					      pfrag->offset)) {
1488 				err = -EMSGSIZE;
1489 				if (i == MAX_SKB_FRAGS)
1490 					goto error;
1491 
1492 				__skb_fill_page_desc(skb, i, pfrag->page,
1493 						     pfrag->offset, 0);
1494 				skb_shinfo(skb)->nr_frags = ++i;
1495 				get_page(pfrag->page);
1496 			}
1497 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1498 			if (getfrag(from,
1499 				    page_address(pfrag->page) + pfrag->offset,
1500 				    offset, copy, skb->len, skb) < 0)
1501 				goto error_efault;
1502 
1503 			pfrag->offset += copy;
1504 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1505 			skb->len += copy;
1506 			skb->data_len += copy;
1507 			skb->truesize += copy;
1508 			atomic_add(copy, &sk->sk_wmem_alloc);
1509 		}
1510 		offset += copy;
1511 		length -= copy;
1512 	}
1513 
1514 	return 0;
1515 
1516 error_efault:
1517 	err = -EFAULT;
1518 error:
1519 	cork->length -= length;
1520 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1521 	return err;
1522 }
1523 EXPORT_SYMBOL_GPL(ip6_append_data);
1524 
1525 static void ip6_cork_release(struct inet_cork_full *cork,
1526 			     struct inet6_cork *v6_cork)
1527 {
1528 	if (v6_cork->opt) {
1529 		kfree(v6_cork->opt->dst0opt);
1530 		kfree(v6_cork->opt->dst1opt);
1531 		kfree(v6_cork->opt->hopopt);
1532 		kfree(v6_cork->opt->srcrt);
1533 		kfree(v6_cork->opt);
1534 		v6_cork->opt = NULL;
1535 	}
1536 
1537 	if (cork->base.dst) {
1538 		dst_release(cork->base.dst);
1539 		cork->base.dst = NULL;
1540 		cork->base.flags &= ~IPCORK_ALLFRAG;
1541 	}
1542 	memset(&cork->fl, 0, sizeof(cork->fl));
1543 }
1544 
1545 int ip6_push_pending_frames(struct sock *sk)
1546 {
1547 	struct sk_buff *skb, *tmp_skb;
1548 	struct sk_buff **tail_skb;
1549 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1550 	struct inet_sock *inet = inet_sk(sk);
1551 	struct ipv6_pinfo *np = inet6_sk(sk);
1552 	struct net *net = sock_net(sk);
1553 	struct ipv6hdr *hdr;
1554 	struct ipv6_txoptions *opt = np->cork.opt;
1555 	struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1556 	struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1557 	unsigned char proto = fl6->flowi6_proto;
1558 	int err = 0;
1559 
1560 	skb = __skb_dequeue(&sk->sk_write_queue);
1561 	if (skb == NULL)
1562 		goto out;
1563 	tail_skb = &(skb_shinfo(skb)->frag_list);
1564 
1565 	/* move skb->data to ip header from ext header */
1566 	if (skb->data < skb_network_header(skb))
1567 		__skb_pull(skb, skb_network_offset(skb));
1568 	while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1569 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1570 		*tail_skb = tmp_skb;
1571 		tail_skb = &(tmp_skb->next);
1572 		skb->len += tmp_skb->len;
1573 		skb->data_len += tmp_skb->len;
1574 		skb->truesize += tmp_skb->truesize;
1575 		tmp_skb->destructor = NULL;
1576 		tmp_skb->sk = NULL;
1577 	}
1578 
1579 	/* Allow local fragmentation. */
1580 	skb->ignore_df = ip6_sk_ignore_df(sk);
1581 
1582 	*final_dst = fl6->daddr;
1583 	__skb_pull(skb, skb_network_header_len(skb));
1584 	if (opt && opt->opt_flen)
1585 		ipv6_push_frag_opts(skb, opt, &proto);
1586 	if (opt && opt->opt_nflen)
1587 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1588 
1589 	skb_push(skb, sizeof(struct ipv6hdr));
1590 	skb_reset_network_header(skb);
1591 	hdr = ipv6_hdr(skb);
1592 
1593 	ip6_flow_hdr(hdr, np->cork.tclass,
1594 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1595 					np->autoflowlabel));
1596 	hdr->hop_limit = np->cork.hop_limit;
1597 	hdr->nexthdr = proto;
1598 	hdr->saddr = fl6->saddr;
1599 	hdr->daddr = *final_dst;
1600 
1601 	skb->priority = sk->sk_priority;
1602 	skb->mark = sk->sk_mark;
1603 
1604 	skb_dst_set(skb, dst_clone(&rt->dst));
1605 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1606 	if (proto == IPPROTO_ICMPV6) {
1607 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1608 
1609 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1610 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1611 	}
1612 
1613 	err = ip6_local_out(skb);
1614 	if (err) {
1615 		if (err > 0)
1616 			err = net_xmit_errno(err);
1617 		if (err)
1618 			goto error;
1619 	}
1620 
1621 out:
1622 	ip6_cork_release(&inet->cork, &np->cork);
1623 	return err;
1624 error:
1625 	IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1626 	goto out;
1627 }
1628 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1629 
1630 void ip6_flush_pending_frames(struct sock *sk)
1631 {
1632 	struct sk_buff *skb;
1633 
1634 	while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1635 		if (skb_dst(skb))
1636 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1637 				      IPSTATS_MIB_OUTDISCARDS);
1638 		kfree_skb(skb);
1639 	}
1640 
1641 	ip6_cork_release(&inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1642 }
1643 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1644