xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 8730046c)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/net/ipv4/ip_output.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  *
15  *	Changes:
16  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
17  *				extension headers are implemented.
18  *				route changes now work.
19  *				ip6_forward does not confuse sniffers.
20  *				etc.
21  *
22  *      H. von Brand    :       Added missing #include <linux/string.h>
23  *	Imran Patel	:	frag id should be in NBO
24  *      Kazunori MIYAZAWA @USAGI
25  *			:       add ip6_append_data and related functions
26  *				for datagram xmit
27  */
28 
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
41 
42 #include <linux/bpf-cgroup.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
61 
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 {
64 	struct dst_entry *dst = skb_dst(skb);
65 	struct net_device *dev = dst->dev;
66 	struct neighbour *neigh;
67 	struct in6_addr *nexthop;
68 	int ret;
69 
70 	skb->protocol = htons(ETH_P_IPV6);
71 	skb->dev = dev;
72 
73 	if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
74 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
75 
76 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
77 		    ((mroute6_socket(net, skb) &&
78 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
79 		     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
80 					 &ipv6_hdr(skb)->saddr))) {
81 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
82 
83 			/* Do not check for IFF_ALLMULTI; multicast routing
84 			   is not supported in any case.
85 			 */
86 			if (newskb)
87 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
88 					net, sk, newskb, NULL, newskb->dev,
89 					dev_loopback_xmit);
90 
91 			if (ipv6_hdr(skb)->hop_limit == 0) {
92 				IP6_INC_STATS(net, idev,
93 					      IPSTATS_MIB_OUTDISCARDS);
94 				kfree_skb(skb);
95 				return 0;
96 			}
97 		}
98 
99 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
100 
101 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
102 		    IPV6_ADDR_SCOPE_NODELOCAL &&
103 		    !(dev->flags & IFF_LOOPBACK)) {
104 			kfree_skb(skb);
105 			return 0;
106 		}
107 	}
108 
109 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
110 		int res = lwtunnel_xmit(skb);
111 
112 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
113 			return res;
114 	}
115 
116 	rcu_read_lock_bh();
117 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
118 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
119 	if (unlikely(!neigh))
120 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
121 	if (!IS_ERR(neigh)) {
122 		ret = dst_neigh_output(dst, neigh, skb);
123 		rcu_read_unlock_bh();
124 		return ret;
125 	}
126 	rcu_read_unlock_bh();
127 
128 	IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
129 	kfree_skb(skb);
130 	return -EINVAL;
131 }
132 
133 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
134 {
135 	int ret;
136 
137 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
138 	if (ret) {
139 		kfree_skb(skb);
140 		return ret;
141 	}
142 
143 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
144 	    dst_allfrag(skb_dst(skb)) ||
145 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
146 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
147 	else
148 		return ip6_finish_output2(net, sk, skb);
149 }
150 
151 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
152 {
153 	struct net_device *dev = skb_dst(skb)->dev;
154 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
155 
156 	if (unlikely(idev->cnf.disable_ipv6)) {
157 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
158 		kfree_skb(skb);
159 		return 0;
160 	}
161 
162 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
163 			    net, sk, skb, NULL, dev,
164 			    ip6_finish_output,
165 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
166 }
167 
168 /*
169  * xmit an sk_buff (used by TCP, SCTP and DCCP)
170  * Note : socket lock is not held for SYNACK packets, but might be modified
171  * by calls to skb_set_owner_w() and ipv6_local_error(),
172  * which are using proper atomic operations or spinlocks.
173  */
174 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
175 	     struct ipv6_txoptions *opt, int tclass)
176 {
177 	struct net *net = sock_net(sk);
178 	const struct ipv6_pinfo *np = inet6_sk(sk);
179 	struct in6_addr *first_hop = &fl6->daddr;
180 	struct dst_entry *dst = skb_dst(skb);
181 	struct ipv6hdr *hdr;
182 	u8  proto = fl6->flowi6_proto;
183 	int seg_len = skb->len;
184 	int hlimit = -1;
185 	u32 mtu;
186 
187 	if (opt) {
188 		unsigned int head_room;
189 
190 		/* First: exthdrs may take lots of space (~8K for now)
191 		   MAX_HEADER is not enough.
192 		 */
193 		head_room = opt->opt_nflen + opt->opt_flen;
194 		seg_len += head_room;
195 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
196 
197 		if (skb_headroom(skb) < head_room) {
198 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
199 			if (!skb2) {
200 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
201 					      IPSTATS_MIB_OUTDISCARDS);
202 				kfree_skb(skb);
203 				return -ENOBUFS;
204 			}
205 			consume_skb(skb);
206 			skb = skb2;
207 			/* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
208 			 * it is safe to call in our context (socket lock not held)
209 			 */
210 			skb_set_owner_w(skb, (struct sock *)sk);
211 		}
212 		if (opt->opt_flen)
213 			ipv6_push_frag_opts(skb, opt, &proto);
214 		if (opt->opt_nflen)
215 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
216 					     &fl6->saddr);
217 	}
218 
219 	skb_push(skb, sizeof(struct ipv6hdr));
220 	skb_reset_network_header(skb);
221 	hdr = ipv6_hdr(skb);
222 
223 	/*
224 	 *	Fill in the IPv6 header
225 	 */
226 	if (np)
227 		hlimit = np->hop_limit;
228 	if (hlimit < 0)
229 		hlimit = ip6_dst_hoplimit(dst);
230 
231 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
232 						     np->autoflowlabel, fl6));
233 
234 	hdr->payload_len = htons(seg_len);
235 	hdr->nexthdr = proto;
236 	hdr->hop_limit = hlimit;
237 
238 	hdr->saddr = fl6->saddr;
239 	hdr->daddr = *first_hop;
240 
241 	skb->protocol = htons(ETH_P_IPV6);
242 	skb->priority = sk->sk_priority;
243 	skb->mark = sk->sk_mark;
244 
245 	mtu = dst_mtu(dst);
246 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
247 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
248 			      IPSTATS_MIB_OUT, skb->len);
249 
250 		/* if egress device is enslaved to an L3 master device pass the
251 		 * skb to its handler for processing
252 		 */
253 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
254 		if (unlikely(!skb))
255 			return 0;
256 
257 		/* hooks should never assume socket lock is held.
258 		 * we promote our socket to non const
259 		 */
260 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
261 			       net, (struct sock *)sk, skb, NULL, dst->dev,
262 			       dst_output);
263 	}
264 
265 	skb->dev = dst->dev;
266 	/* ipv6_local_error() does not require socket lock,
267 	 * we promote our socket to non const
268 	 */
269 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
270 
271 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
272 	kfree_skb(skb);
273 	return -EMSGSIZE;
274 }
275 EXPORT_SYMBOL(ip6_xmit);
276 
277 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
278 {
279 	struct ip6_ra_chain *ra;
280 	struct sock *last = NULL;
281 
282 	read_lock(&ip6_ra_lock);
283 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
284 		struct sock *sk = ra->sk;
285 		if (sk && ra->sel == sel &&
286 		    (!sk->sk_bound_dev_if ||
287 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
288 			if (last) {
289 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
290 				if (skb2)
291 					rawv6_rcv(last, skb2);
292 			}
293 			last = sk;
294 		}
295 	}
296 
297 	if (last) {
298 		rawv6_rcv(last, skb);
299 		read_unlock(&ip6_ra_lock);
300 		return 1;
301 	}
302 	read_unlock(&ip6_ra_lock);
303 	return 0;
304 }
305 
306 static int ip6_forward_proxy_check(struct sk_buff *skb)
307 {
308 	struct ipv6hdr *hdr = ipv6_hdr(skb);
309 	u8 nexthdr = hdr->nexthdr;
310 	__be16 frag_off;
311 	int offset;
312 
313 	if (ipv6_ext_hdr(nexthdr)) {
314 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
315 		if (offset < 0)
316 			return 0;
317 	} else
318 		offset = sizeof(struct ipv6hdr);
319 
320 	if (nexthdr == IPPROTO_ICMPV6) {
321 		struct icmp6hdr *icmp6;
322 
323 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
324 					 offset + 1 - skb->data)))
325 			return 0;
326 
327 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
328 
329 		switch (icmp6->icmp6_type) {
330 		case NDISC_ROUTER_SOLICITATION:
331 		case NDISC_ROUTER_ADVERTISEMENT:
332 		case NDISC_NEIGHBOUR_SOLICITATION:
333 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
334 		case NDISC_REDIRECT:
335 			/* For reaction involving unicast neighbor discovery
336 			 * message destined to the proxied address, pass it to
337 			 * input function.
338 			 */
339 			return 1;
340 		default:
341 			break;
342 		}
343 	}
344 
345 	/*
346 	 * The proxying router can't forward traffic sent to a link-local
347 	 * address, so signal the sender and discard the packet. This
348 	 * behavior is clarified by the MIPv6 specification.
349 	 */
350 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
351 		dst_link_failure(skb);
352 		return -1;
353 	}
354 
355 	return 0;
356 }
357 
358 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
359 				     struct sk_buff *skb)
360 {
361 	return dst_output(net, sk, skb);
362 }
363 
364 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
365 {
366 	unsigned int mtu;
367 	struct inet6_dev *idev;
368 
369 	if (dst_metric_locked(dst, RTAX_MTU)) {
370 		mtu = dst_metric_raw(dst, RTAX_MTU);
371 		if (mtu)
372 			return mtu;
373 	}
374 
375 	mtu = IPV6_MIN_MTU;
376 	rcu_read_lock();
377 	idev = __in6_dev_get(dst->dev);
378 	if (idev)
379 		mtu = idev->cnf.mtu6;
380 	rcu_read_unlock();
381 
382 	return mtu;
383 }
384 
385 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
386 {
387 	if (skb->len <= mtu)
388 		return false;
389 
390 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
391 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
392 		return true;
393 
394 	if (skb->ignore_df)
395 		return false;
396 
397 	if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
398 		return false;
399 
400 	return true;
401 }
402 
403 int ip6_forward(struct sk_buff *skb)
404 {
405 	struct dst_entry *dst = skb_dst(skb);
406 	struct ipv6hdr *hdr = ipv6_hdr(skb);
407 	struct inet6_skb_parm *opt = IP6CB(skb);
408 	struct net *net = dev_net(dst->dev);
409 	u32 mtu;
410 
411 	if (net->ipv6.devconf_all->forwarding == 0)
412 		goto error;
413 
414 	if (skb->pkt_type != PACKET_HOST)
415 		goto drop;
416 
417 	if (unlikely(skb->sk))
418 		goto drop;
419 
420 	if (skb_warn_if_lro(skb))
421 		goto drop;
422 
423 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
424 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
425 				IPSTATS_MIB_INDISCARDS);
426 		goto drop;
427 	}
428 
429 	skb_forward_csum(skb);
430 
431 	/*
432 	 *	We DO NOT make any processing on
433 	 *	RA packets, pushing them to user level AS IS
434 	 *	without ane WARRANTY that application will be able
435 	 *	to interpret them. The reason is that we
436 	 *	cannot make anything clever here.
437 	 *
438 	 *	We are not end-node, so that if packet contains
439 	 *	AH/ESP, we cannot make anything.
440 	 *	Defragmentation also would be mistake, RA packets
441 	 *	cannot be fragmented, because there is no warranty
442 	 *	that different fragments will go along one path. --ANK
443 	 */
444 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
445 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
446 			return 0;
447 	}
448 
449 	/*
450 	 *	check and decrement ttl
451 	 */
452 	if (hdr->hop_limit <= 1) {
453 		/* Force OUTPUT device used as source address */
454 		skb->dev = dst->dev;
455 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
456 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
457 				IPSTATS_MIB_INHDRERRORS);
458 
459 		kfree_skb(skb);
460 		return -ETIMEDOUT;
461 	}
462 
463 	/* XXX: idev->cnf.proxy_ndp? */
464 	if (net->ipv6.devconf_all->proxy_ndp &&
465 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
466 		int proxied = ip6_forward_proxy_check(skb);
467 		if (proxied > 0)
468 			return ip6_input(skb);
469 		else if (proxied < 0) {
470 			__IP6_INC_STATS(net, ip6_dst_idev(dst),
471 					IPSTATS_MIB_INDISCARDS);
472 			goto drop;
473 		}
474 	}
475 
476 	if (!xfrm6_route_forward(skb)) {
477 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
478 				IPSTATS_MIB_INDISCARDS);
479 		goto drop;
480 	}
481 	dst = skb_dst(skb);
482 
483 	/* IPv6 specs say nothing about it, but it is clear that we cannot
484 	   send redirects to source routed frames.
485 	   We don't send redirects to frames decapsulated from IPsec.
486 	 */
487 	if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
488 		struct in6_addr *target = NULL;
489 		struct inet_peer *peer;
490 		struct rt6_info *rt;
491 
492 		/*
493 		 *	incoming and outgoing devices are the same
494 		 *	send a redirect.
495 		 */
496 
497 		rt = (struct rt6_info *) dst;
498 		if (rt->rt6i_flags & RTF_GATEWAY)
499 			target = &rt->rt6i_gateway;
500 		else
501 			target = &hdr->daddr;
502 
503 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
504 
505 		/* Limit redirects both by destination (here)
506 		   and by source (inside ndisc_send_redirect)
507 		 */
508 		if (inet_peer_xrlim_allow(peer, 1*HZ))
509 			ndisc_send_redirect(skb, target);
510 		if (peer)
511 			inet_putpeer(peer);
512 	} else {
513 		int addrtype = ipv6_addr_type(&hdr->saddr);
514 
515 		/* This check is security critical. */
516 		if (addrtype == IPV6_ADDR_ANY ||
517 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
518 			goto error;
519 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
520 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
521 				    ICMPV6_NOT_NEIGHBOUR, 0);
522 			goto error;
523 		}
524 	}
525 
526 	mtu = ip6_dst_mtu_forward(dst);
527 	if (mtu < IPV6_MIN_MTU)
528 		mtu = IPV6_MIN_MTU;
529 
530 	if (ip6_pkt_too_big(skb, mtu)) {
531 		/* Again, force OUTPUT device used as source address */
532 		skb->dev = dst->dev;
533 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
534 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
535 				IPSTATS_MIB_INTOOBIGERRORS);
536 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
537 				IPSTATS_MIB_FRAGFAILS);
538 		kfree_skb(skb);
539 		return -EMSGSIZE;
540 	}
541 
542 	if (skb_cow(skb, dst->dev->hard_header_len)) {
543 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
544 				IPSTATS_MIB_OUTDISCARDS);
545 		goto drop;
546 	}
547 
548 	hdr = ipv6_hdr(skb);
549 
550 	/* Mangling hops number delayed to point after skb COW */
551 
552 	hdr->hop_limit--;
553 
554 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
555 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
556 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
557 		       net, NULL, skb, skb->dev, dst->dev,
558 		       ip6_forward_finish);
559 
560 error:
561 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
562 drop:
563 	kfree_skb(skb);
564 	return -EINVAL;
565 }
566 
567 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
568 {
569 	to->pkt_type = from->pkt_type;
570 	to->priority = from->priority;
571 	to->protocol = from->protocol;
572 	skb_dst_drop(to);
573 	skb_dst_set(to, dst_clone(skb_dst(from)));
574 	to->dev = from->dev;
575 	to->mark = from->mark;
576 
577 #ifdef CONFIG_NET_SCHED
578 	to->tc_index = from->tc_index;
579 #endif
580 	nf_copy(to, from);
581 	skb_copy_secmark(to, from);
582 }
583 
584 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
585 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
586 {
587 	struct sk_buff *frag;
588 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
589 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
590 				inet6_sk(skb->sk) : NULL;
591 	struct ipv6hdr *tmp_hdr;
592 	struct frag_hdr *fh;
593 	unsigned int mtu, hlen, left, len;
594 	int hroom, troom;
595 	__be32 frag_id;
596 	int ptr, offset = 0, err = 0;
597 	u8 *prevhdr, nexthdr = 0;
598 
599 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
600 	nexthdr = *prevhdr;
601 
602 	mtu = ip6_skb_dst_mtu(skb);
603 
604 	/* We must not fragment if the socket is set to force MTU discovery
605 	 * or if the skb it not generated by a local socket.
606 	 */
607 	if (unlikely(!skb->ignore_df && skb->len > mtu))
608 		goto fail_toobig;
609 
610 	if (IP6CB(skb)->frag_max_size) {
611 		if (IP6CB(skb)->frag_max_size > mtu)
612 			goto fail_toobig;
613 
614 		/* don't send fragments larger than what we received */
615 		mtu = IP6CB(skb)->frag_max_size;
616 		if (mtu < IPV6_MIN_MTU)
617 			mtu = IPV6_MIN_MTU;
618 	}
619 
620 	if (np && np->frag_size < mtu) {
621 		if (np->frag_size)
622 			mtu = np->frag_size;
623 	}
624 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
625 		goto fail_toobig;
626 	mtu -= hlen + sizeof(struct frag_hdr);
627 
628 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
629 				    &ipv6_hdr(skb)->saddr);
630 
631 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
632 	    (err = skb_checksum_help(skb)))
633 		goto fail;
634 
635 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
636 	if (skb_has_frag_list(skb)) {
637 		unsigned int first_len = skb_pagelen(skb);
638 		struct sk_buff *frag2;
639 
640 		if (first_len - hlen > mtu ||
641 		    ((first_len - hlen) & 7) ||
642 		    skb_cloned(skb) ||
643 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
644 			goto slow_path;
645 
646 		skb_walk_frags(skb, frag) {
647 			/* Correct geometry. */
648 			if (frag->len > mtu ||
649 			    ((frag->len & 7) && frag->next) ||
650 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
651 				goto slow_path_clean;
652 
653 			/* Partially cloned skb? */
654 			if (skb_shared(frag))
655 				goto slow_path_clean;
656 
657 			BUG_ON(frag->sk);
658 			if (skb->sk) {
659 				frag->sk = skb->sk;
660 				frag->destructor = sock_wfree;
661 			}
662 			skb->truesize -= frag->truesize;
663 		}
664 
665 		err = 0;
666 		offset = 0;
667 		/* BUILD HEADER */
668 
669 		*prevhdr = NEXTHDR_FRAGMENT;
670 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
671 		if (!tmp_hdr) {
672 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
673 				      IPSTATS_MIB_FRAGFAILS);
674 			err = -ENOMEM;
675 			goto fail;
676 		}
677 		frag = skb_shinfo(skb)->frag_list;
678 		skb_frag_list_init(skb);
679 
680 		__skb_pull(skb, hlen);
681 		fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
682 		__skb_push(skb, hlen);
683 		skb_reset_network_header(skb);
684 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
685 
686 		fh->nexthdr = nexthdr;
687 		fh->reserved = 0;
688 		fh->frag_off = htons(IP6_MF);
689 		fh->identification = frag_id;
690 
691 		first_len = skb_pagelen(skb);
692 		skb->data_len = first_len - skb_headlen(skb);
693 		skb->len = first_len;
694 		ipv6_hdr(skb)->payload_len = htons(first_len -
695 						   sizeof(struct ipv6hdr));
696 
697 		dst_hold(&rt->dst);
698 
699 		for (;;) {
700 			/* Prepare header of the next frame,
701 			 * before previous one went down. */
702 			if (frag) {
703 				frag->ip_summed = CHECKSUM_NONE;
704 				skb_reset_transport_header(frag);
705 				fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
706 				__skb_push(frag, hlen);
707 				skb_reset_network_header(frag);
708 				memcpy(skb_network_header(frag), tmp_hdr,
709 				       hlen);
710 				offset += skb->len - hlen - sizeof(struct frag_hdr);
711 				fh->nexthdr = nexthdr;
712 				fh->reserved = 0;
713 				fh->frag_off = htons(offset);
714 				if (frag->next)
715 					fh->frag_off |= htons(IP6_MF);
716 				fh->identification = frag_id;
717 				ipv6_hdr(frag)->payload_len =
718 						htons(frag->len -
719 						      sizeof(struct ipv6hdr));
720 				ip6_copy_metadata(frag, skb);
721 			}
722 
723 			err = output(net, sk, skb);
724 			if (!err)
725 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
726 					      IPSTATS_MIB_FRAGCREATES);
727 
728 			if (err || !frag)
729 				break;
730 
731 			skb = frag;
732 			frag = skb->next;
733 			skb->next = NULL;
734 		}
735 
736 		kfree(tmp_hdr);
737 
738 		if (err == 0) {
739 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
740 				      IPSTATS_MIB_FRAGOKS);
741 			ip6_rt_put(rt);
742 			return 0;
743 		}
744 
745 		kfree_skb_list(frag);
746 
747 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
748 			      IPSTATS_MIB_FRAGFAILS);
749 		ip6_rt_put(rt);
750 		return err;
751 
752 slow_path_clean:
753 		skb_walk_frags(skb, frag2) {
754 			if (frag2 == frag)
755 				break;
756 			frag2->sk = NULL;
757 			frag2->destructor = NULL;
758 			skb->truesize += frag2->truesize;
759 		}
760 	}
761 
762 slow_path:
763 	left = skb->len - hlen;		/* Space per frame */
764 	ptr = hlen;			/* Where to start from */
765 
766 	/*
767 	 *	Fragment the datagram.
768 	 */
769 
770 	*prevhdr = NEXTHDR_FRAGMENT;
771 	troom = rt->dst.dev->needed_tailroom;
772 
773 	/*
774 	 *	Keep copying data until we run out.
775 	 */
776 	while (left > 0)	{
777 		len = left;
778 		/* IF: it doesn't fit, use 'mtu' - the data space left */
779 		if (len > mtu)
780 			len = mtu;
781 		/* IF: we are not sending up to and including the packet end
782 		   then align the next start on an eight byte boundary */
783 		if (len < left)	{
784 			len &= ~7;
785 		}
786 
787 		/* Allocate buffer */
788 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
789 				 hroom + troom, GFP_ATOMIC);
790 		if (!frag) {
791 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
792 				      IPSTATS_MIB_FRAGFAILS);
793 			err = -ENOMEM;
794 			goto fail;
795 		}
796 
797 		/*
798 		 *	Set up data on packet
799 		 */
800 
801 		ip6_copy_metadata(frag, skb);
802 		skb_reserve(frag, hroom);
803 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
804 		skb_reset_network_header(frag);
805 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
806 		frag->transport_header = (frag->network_header + hlen +
807 					  sizeof(struct frag_hdr));
808 
809 		/*
810 		 *	Charge the memory for the fragment to any owner
811 		 *	it might possess
812 		 */
813 		if (skb->sk)
814 			skb_set_owner_w(frag, skb->sk);
815 
816 		/*
817 		 *	Copy the packet header into the new buffer.
818 		 */
819 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
820 
821 		/*
822 		 *	Build fragment header.
823 		 */
824 		fh->nexthdr = nexthdr;
825 		fh->reserved = 0;
826 		fh->identification = frag_id;
827 
828 		/*
829 		 *	Copy a block of the IP datagram.
830 		 */
831 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
832 				     len));
833 		left -= len;
834 
835 		fh->frag_off = htons(offset);
836 		if (left > 0)
837 			fh->frag_off |= htons(IP6_MF);
838 		ipv6_hdr(frag)->payload_len = htons(frag->len -
839 						    sizeof(struct ipv6hdr));
840 
841 		ptr += len;
842 		offset += len;
843 
844 		/*
845 		 *	Put this fragment into the sending queue.
846 		 */
847 		err = output(net, sk, frag);
848 		if (err)
849 			goto fail;
850 
851 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
852 			      IPSTATS_MIB_FRAGCREATES);
853 	}
854 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
855 		      IPSTATS_MIB_FRAGOKS);
856 	consume_skb(skb);
857 	return err;
858 
859 fail_toobig:
860 	if (skb->sk && dst_allfrag(skb_dst(skb)))
861 		sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
862 
863 	skb->dev = skb_dst(skb)->dev;
864 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
865 	err = -EMSGSIZE;
866 
867 fail:
868 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
869 		      IPSTATS_MIB_FRAGFAILS);
870 	kfree_skb(skb);
871 	return err;
872 }
873 
874 static inline int ip6_rt_check(const struct rt6key *rt_key,
875 			       const struct in6_addr *fl_addr,
876 			       const struct in6_addr *addr_cache)
877 {
878 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
879 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
880 }
881 
882 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
883 					  struct dst_entry *dst,
884 					  const struct flowi6 *fl6)
885 {
886 	struct ipv6_pinfo *np = inet6_sk(sk);
887 	struct rt6_info *rt;
888 
889 	if (!dst)
890 		goto out;
891 
892 	if (dst->ops->family != AF_INET6) {
893 		dst_release(dst);
894 		return NULL;
895 	}
896 
897 	rt = (struct rt6_info *)dst;
898 	/* Yes, checking route validity in not connected
899 	 * case is not very simple. Take into account,
900 	 * that we do not support routing by source, TOS,
901 	 * and MSG_DONTROUTE		--ANK (980726)
902 	 *
903 	 * 1. ip6_rt_check(): If route was host route,
904 	 *    check that cached destination is current.
905 	 *    If it is network route, we still may
906 	 *    check its validity using saved pointer
907 	 *    to the last used address: daddr_cache.
908 	 *    We do not want to save whole address now,
909 	 *    (because main consumer of this service
910 	 *    is tcp, which has not this problem),
911 	 *    so that the last trick works only on connected
912 	 *    sockets.
913 	 * 2. oif also should be the same.
914 	 */
915 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
916 #ifdef CONFIG_IPV6_SUBTREES
917 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
918 #endif
919 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
920 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
921 		dst_release(dst);
922 		dst = NULL;
923 	}
924 
925 out:
926 	return dst;
927 }
928 
929 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
930 			       struct dst_entry **dst, struct flowi6 *fl6)
931 {
932 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
933 	struct neighbour *n;
934 	struct rt6_info *rt;
935 #endif
936 	int err;
937 	int flags = 0;
938 
939 	/* The correct way to handle this would be to do
940 	 * ip6_route_get_saddr, and then ip6_route_output; however,
941 	 * the route-specific preferred source forces the
942 	 * ip6_route_output call _before_ ip6_route_get_saddr.
943 	 *
944 	 * In source specific routing (no src=any default route),
945 	 * ip6_route_output will fail given src=any saddr, though, so
946 	 * that's why we try it again later.
947 	 */
948 	if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
949 		struct rt6_info *rt;
950 		bool had_dst = *dst != NULL;
951 
952 		if (!had_dst)
953 			*dst = ip6_route_output(net, sk, fl6);
954 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
955 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
956 					  sk ? inet6_sk(sk)->srcprefs : 0,
957 					  &fl6->saddr);
958 		if (err)
959 			goto out_err_release;
960 
961 		/* If we had an erroneous initial result, pretend it
962 		 * never existed and let the SA-enabled version take
963 		 * over.
964 		 */
965 		if (!had_dst && (*dst)->error) {
966 			dst_release(*dst);
967 			*dst = NULL;
968 		}
969 
970 		if (fl6->flowi6_oif)
971 			flags |= RT6_LOOKUP_F_IFACE;
972 	}
973 
974 	if (!*dst)
975 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
976 
977 	err = (*dst)->error;
978 	if (err)
979 		goto out_err_release;
980 
981 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
982 	/*
983 	 * Here if the dst entry we've looked up
984 	 * has a neighbour entry that is in the INCOMPLETE
985 	 * state and the src address from the flow is
986 	 * marked as OPTIMISTIC, we release the found
987 	 * dst entry and replace it instead with the
988 	 * dst entry of the nexthop router
989 	 */
990 	rt = (struct rt6_info *) *dst;
991 	rcu_read_lock_bh();
992 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
993 				      rt6_nexthop(rt, &fl6->daddr));
994 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
995 	rcu_read_unlock_bh();
996 
997 	if (err) {
998 		struct inet6_ifaddr *ifp;
999 		struct flowi6 fl_gw6;
1000 		int redirect;
1001 
1002 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1003 				      (*dst)->dev, 1);
1004 
1005 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1006 		if (ifp)
1007 			in6_ifa_put(ifp);
1008 
1009 		if (redirect) {
1010 			/*
1011 			 * We need to get the dst entry for the
1012 			 * default router instead
1013 			 */
1014 			dst_release(*dst);
1015 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1016 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1017 			*dst = ip6_route_output(net, sk, &fl_gw6);
1018 			err = (*dst)->error;
1019 			if (err)
1020 				goto out_err_release;
1021 		}
1022 	}
1023 #endif
1024 
1025 	return 0;
1026 
1027 out_err_release:
1028 	dst_release(*dst);
1029 	*dst = NULL;
1030 
1031 	if (err == -ENETUNREACH)
1032 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1033 	return err;
1034 }
1035 
1036 /**
1037  *	ip6_dst_lookup - perform route lookup on flow
1038  *	@sk: socket which provides route info
1039  *	@dst: pointer to dst_entry * for result
1040  *	@fl6: flow to lookup
1041  *
1042  *	This function performs a route lookup on the given flow.
1043  *
1044  *	It returns zero on success, or a standard errno code on error.
1045  */
1046 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1047 		   struct flowi6 *fl6)
1048 {
1049 	*dst = NULL;
1050 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1051 }
1052 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1053 
1054 /**
1055  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1056  *	@sk: socket which provides route info
1057  *	@fl6: flow to lookup
1058  *	@final_dst: final destination address for ipsec lookup
1059  *
1060  *	This function performs a route lookup on the given flow.
1061  *
1062  *	It returns a valid dst pointer on success, or a pointer encoded
1063  *	error code.
1064  */
1065 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1066 				      const struct in6_addr *final_dst)
1067 {
1068 	struct dst_entry *dst = NULL;
1069 	int err;
1070 
1071 	err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1072 	if (err)
1073 		return ERR_PTR(err);
1074 	if (final_dst)
1075 		fl6->daddr = *final_dst;
1076 
1077 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1078 }
1079 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1080 
1081 /**
1082  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1083  *	@sk: socket which provides the dst cache and route info
1084  *	@fl6: flow to lookup
1085  *	@final_dst: final destination address for ipsec lookup
1086  *
1087  *	This function performs a route lookup on the given flow with the
1088  *	possibility of using the cached route in the socket if it is valid.
1089  *	It will take the socket dst lock when operating on the dst cache.
1090  *	As a result, this function can only be used in process context.
1091  *
1092  *	It returns a valid dst pointer on success, or a pointer encoded
1093  *	error code.
1094  */
1095 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1096 					 const struct in6_addr *final_dst)
1097 {
1098 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1099 
1100 	dst = ip6_sk_dst_check(sk, dst, fl6);
1101 	if (!dst)
1102 		dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1103 
1104 	return dst;
1105 }
1106 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1107 
1108 static inline int ip6_ufo_append_data(struct sock *sk,
1109 			struct sk_buff_head *queue,
1110 			int getfrag(void *from, char *to, int offset, int len,
1111 			int odd, struct sk_buff *skb),
1112 			void *from, int length, int hh_len, int fragheaderlen,
1113 			int exthdrlen, int transhdrlen, int mtu,
1114 			unsigned int flags, const struct flowi6 *fl6)
1115 
1116 {
1117 	struct sk_buff *skb;
1118 	int err;
1119 
1120 	/* There is support for UDP large send offload by network
1121 	 * device, so create one single skb packet containing complete
1122 	 * udp datagram
1123 	 */
1124 	skb = skb_peek_tail(queue);
1125 	if (!skb) {
1126 		skb = sock_alloc_send_skb(sk,
1127 			hh_len + fragheaderlen + transhdrlen + 20,
1128 			(flags & MSG_DONTWAIT), &err);
1129 		if (!skb)
1130 			return err;
1131 
1132 		/* reserve space for Hardware header */
1133 		skb_reserve(skb, hh_len);
1134 
1135 		/* create space for UDP/IP header */
1136 		skb_put(skb, fragheaderlen + transhdrlen);
1137 
1138 		/* initialize network header pointer */
1139 		skb_set_network_header(skb, exthdrlen);
1140 
1141 		/* initialize protocol header pointer */
1142 		skb->transport_header = skb->network_header + fragheaderlen;
1143 
1144 		skb->protocol = htons(ETH_P_IPV6);
1145 		skb->csum = 0;
1146 
1147 		__skb_queue_tail(queue, skb);
1148 	} else if (skb_is_gso(skb)) {
1149 		goto append;
1150 	}
1151 
1152 	skb->ip_summed = CHECKSUM_PARTIAL;
1153 	/* Specify the length of each IPv6 datagram fragment.
1154 	 * It has to be a multiple of 8.
1155 	 */
1156 	skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1157 				     sizeof(struct frag_hdr)) & ~7;
1158 	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1159 	skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1160 							 &fl6->daddr,
1161 							 &fl6->saddr);
1162 
1163 append:
1164 	return skb_append_datato_frags(sk, skb, getfrag, from,
1165 				       (length - transhdrlen));
1166 }
1167 
1168 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1169 					       gfp_t gfp)
1170 {
1171 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1172 }
1173 
1174 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1175 						gfp_t gfp)
1176 {
1177 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1178 }
1179 
1180 static void ip6_append_data_mtu(unsigned int *mtu,
1181 				int *maxfraglen,
1182 				unsigned int fragheaderlen,
1183 				struct sk_buff *skb,
1184 				struct rt6_info *rt,
1185 				unsigned int orig_mtu)
1186 {
1187 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1188 		if (!skb) {
1189 			/* first fragment, reserve header_len */
1190 			*mtu = orig_mtu - rt->dst.header_len;
1191 
1192 		} else {
1193 			/*
1194 			 * this fragment is not first, the headers
1195 			 * space is regarded as data space.
1196 			 */
1197 			*mtu = orig_mtu;
1198 		}
1199 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1200 			      + fragheaderlen - sizeof(struct frag_hdr);
1201 	}
1202 }
1203 
1204 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1205 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1206 			  struct rt6_info *rt, struct flowi6 *fl6)
1207 {
1208 	struct ipv6_pinfo *np = inet6_sk(sk);
1209 	unsigned int mtu;
1210 	struct ipv6_txoptions *opt = ipc6->opt;
1211 
1212 	/*
1213 	 * setup for corking
1214 	 */
1215 	if (opt) {
1216 		if (WARN_ON(v6_cork->opt))
1217 			return -EINVAL;
1218 
1219 		v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1220 		if (unlikely(!v6_cork->opt))
1221 			return -ENOBUFS;
1222 
1223 		v6_cork->opt->tot_len = opt->tot_len;
1224 		v6_cork->opt->opt_flen = opt->opt_flen;
1225 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1226 
1227 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1228 						    sk->sk_allocation);
1229 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1230 			return -ENOBUFS;
1231 
1232 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1233 						    sk->sk_allocation);
1234 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1235 			return -ENOBUFS;
1236 
1237 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1238 						   sk->sk_allocation);
1239 		if (opt->hopopt && !v6_cork->opt->hopopt)
1240 			return -ENOBUFS;
1241 
1242 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1243 						    sk->sk_allocation);
1244 		if (opt->srcrt && !v6_cork->opt->srcrt)
1245 			return -ENOBUFS;
1246 
1247 		/* need source address above miyazawa*/
1248 	}
1249 	dst_hold(&rt->dst);
1250 	cork->base.dst = &rt->dst;
1251 	cork->fl.u.ip6 = *fl6;
1252 	v6_cork->hop_limit = ipc6->hlimit;
1253 	v6_cork->tclass = ipc6->tclass;
1254 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1255 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1256 		      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1257 	else
1258 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1259 		      rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1260 	if (np->frag_size < mtu) {
1261 		if (np->frag_size)
1262 			mtu = np->frag_size;
1263 	}
1264 	cork->base.fragsize = mtu;
1265 	if (dst_allfrag(rt->dst.path))
1266 		cork->base.flags |= IPCORK_ALLFRAG;
1267 	cork->base.length = 0;
1268 
1269 	return 0;
1270 }
1271 
1272 static int __ip6_append_data(struct sock *sk,
1273 			     struct flowi6 *fl6,
1274 			     struct sk_buff_head *queue,
1275 			     struct inet_cork *cork,
1276 			     struct inet6_cork *v6_cork,
1277 			     struct page_frag *pfrag,
1278 			     int getfrag(void *from, char *to, int offset,
1279 					 int len, int odd, struct sk_buff *skb),
1280 			     void *from, int length, int transhdrlen,
1281 			     unsigned int flags, struct ipcm6_cookie *ipc6,
1282 			     const struct sockcm_cookie *sockc)
1283 {
1284 	struct sk_buff *skb, *skb_prev = NULL;
1285 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1286 	int exthdrlen = 0;
1287 	int dst_exthdrlen = 0;
1288 	int hh_len;
1289 	int copy;
1290 	int err;
1291 	int offset = 0;
1292 	__u8 tx_flags = 0;
1293 	u32 tskey = 0;
1294 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1295 	struct ipv6_txoptions *opt = v6_cork->opt;
1296 	int csummode = CHECKSUM_NONE;
1297 	unsigned int maxnonfragsize, headersize;
1298 
1299 	skb = skb_peek_tail(queue);
1300 	if (!skb) {
1301 		exthdrlen = opt ? opt->opt_flen : 0;
1302 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1303 	}
1304 
1305 	mtu = cork->fragsize;
1306 	orig_mtu = mtu;
1307 
1308 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1309 
1310 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1311 			(opt ? opt->opt_nflen : 0);
1312 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1313 		     sizeof(struct frag_hdr);
1314 
1315 	headersize = sizeof(struct ipv6hdr) +
1316 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1317 		     (dst_allfrag(&rt->dst) ?
1318 		      sizeof(struct frag_hdr) : 0) +
1319 		     rt->rt6i_nfheader_len;
1320 
1321 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1322 	    (sk->sk_protocol == IPPROTO_UDP ||
1323 	     sk->sk_protocol == IPPROTO_RAW)) {
1324 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1325 				sizeof(struct ipv6hdr));
1326 		goto emsgsize;
1327 	}
1328 
1329 	if (ip6_sk_ignore_df(sk))
1330 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1331 	else
1332 		maxnonfragsize = mtu;
1333 
1334 	if (cork->length + length > maxnonfragsize - headersize) {
1335 emsgsize:
1336 		ipv6_local_error(sk, EMSGSIZE, fl6,
1337 				 mtu - headersize +
1338 				 sizeof(struct ipv6hdr));
1339 		return -EMSGSIZE;
1340 	}
1341 
1342 	/* CHECKSUM_PARTIAL only with no extension headers and when
1343 	 * we are not going to fragment
1344 	 */
1345 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1346 	    headersize == sizeof(struct ipv6hdr) &&
1347 	    length < mtu - headersize &&
1348 	    !(flags & MSG_MORE) &&
1349 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1350 		csummode = CHECKSUM_PARTIAL;
1351 
1352 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1353 		sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1354 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1355 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1356 			tskey = sk->sk_tskey++;
1357 	}
1358 
1359 	/*
1360 	 * Let's try using as much space as possible.
1361 	 * Use MTU if total length of the message fits into the MTU.
1362 	 * Otherwise, we need to reserve fragment header and
1363 	 * fragment alignment (= 8-15 octects, in total).
1364 	 *
1365 	 * Note that we may need to "move" the data from the tail of
1366 	 * of the buffer to the new fragment when we split
1367 	 * the message.
1368 	 *
1369 	 * FIXME: It may be fragmented into multiple chunks
1370 	 *        at once if non-fragmentable extension headers
1371 	 *        are too large.
1372 	 * --yoshfuji
1373 	 */
1374 
1375 	cork->length += length;
1376 	if ((((length + fragheaderlen) > mtu) ||
1377 	     (skb && skb_is_gso(skb))) &&
1378 	    (sk->sk_protocol == IPPROTO_UDP) &&
1379 	    (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
1380 	    (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) {
1381 		err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1382 					  hh_len, fragheaderlen, exthdrlen,
1383 					  transhdrlen, mtu, flags, fl6);
1384 		if (err)
1385 			goto error;
1386 		return 0;
1387 	}
1388 
1389 	if (!skb)
1390 		goto alloc_new_skb;
1391 
1392 	while (length > 0) {
1393 		/* Check if the remaining data fits into current packet. */
1394 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1395 		if (copy < length)
1396 			copy = maxfraglen - skb->len;
1397 
1398 		if (copy <= 0) {
1399 			char *data;
1400 			unsigned int datalen;
1401 			unsigned int fraglen;
1402 			unsigned int fraggap;
1403 			unsigned int alloclen;
1404 alloc_new_skb:
1405 			/* There's no room in the current skb */
1406 			if (skb)
1407 				fraggap = skb->len - maxfraglen;
1408 			else
1409 				fraggap = 0;
1410 			/* update mtu and maxfraglen if necessary */
1411 			if (!skb || !skb_prev)
1412 				ip6_append_data_mtu(&mtu, &maxfraglen,
1413 						    fragheaderlen, skb, rt,
1414 						    orig_mtu);
1415 
1416 			skb_prev = skb;
1417 
1418 			/*
1419 			 * If remaining data exceeds the mtu,
1420 			 * we know we need more fragment(s).
1421 			 */
1422 			datalen = length + fraggap;
1423 
1424 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1425 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1426 			if ((flags & MSG_MORE) &&
1427 			    !(rt->dst.dev->features&NETIF_F_SG))
1428 				alloclen = mtu;
1429 			else
1430 				alloclen = datalen + fragheaderlen;
1431 
1432 			alloclen += dst_exthdrlen;
1433 
1434 			if (datalen != length + fraggap) {
1435 				/*
1436 				 * this is not the last fragment, the trailer
1437 				 * space is regarded as data space.
1438 				 */
1439 				datalen += rt->dst.trailer_len;
1440 			}
1441 
1442 			alloclen += rt->dst.trailer_len;
1443 			fraglen = datalen + fragheaderlen;
1444 
1445 			/*
1446 			 * We just reserve space for fragment header.
1447 			 * Note: this may be overallocation if the message
1448 			 * (without MSG_MORE) fits into the MTU.
1449 			 */
1450 			alloclen += sizeof(struct frag_hdr);
1451 
1452 			if (transhdrlen) {
1453 				skb = sock_alloc_send_skb(sk,
1454 						alloclen + hh_len,
1455 						(flags & MSG_DONTWAIT), &err);
1456 			} else {
1457 				skb = NULL;
1458 				if (atomic_read(&sk->sk_wmem_alloc) <=
1459 				    2 * sk->sk_sndbuf)
1460 					skb = sock_wmalloc(sk,
1461 							   alloclen + hh_len, 1,
1462 							   sk->sk_allocation);
1463 				if (unlikely(!skb))
1464 					err = -ENOBUFS;
1465 			}
1466 			if (!skb)
1467 				goto error;
1468 			/*
1469 			 *	Fill in the control structures
1470 			 */
1471 			skb->protocol = htons(ETH_P_IPV6);
1472 			skb->ip_summed = csummode;
1473 			skb->csum = 0;
1474 			/* reserve for fragmentation and ipsec header */
1475 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1476 				    dst_exthdrlen);
1477 
1478 			/* Only the initial fragment is time stamped */
1479 			skb_shinfo(skb)->tx_flags = tx_flags;
1480 			tx_flags = 0;
1481 			skb_shinfo(skb)->tskey = tskey;
1482 			tskey = 0;
1483 
1484 			/*
1485 			 *	Find where to start putting bytes
1486 			 */
1487 			data = skb_put(skb, fraglen);
1488 			skb_set_network_header(skb, exthdrlen);
1489 			data += fragheaderlen;
1490 			skb->transport_header = (skb->network_header +
1491 						 fragheaderlen);
1492 			if (fraggap) {
1493 				skb->csum = skb_copy_and_csum_bits(
1494 					skb_prev, maxfraglen,
1495 					data + transhdrlen, fraggap, 0);
1496 				skb_prev->csum = csum_sub(skb_prev->csum,
1497 							  skb->csum);
1498 				data += fraggap;
1499 				pskb_trim_unique(skb_prev, maxfraglen);
1500 			}
1501 			copy = datalen - transhdrlen - fraggap;
1502 
1503 			if (copy < 0) {
1504 				err = -EINVAL;
1505 				kfree_skb(skb);
1506 				goto error;
1507 			} else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1508 				err = -EFAULT;
1509 				kfree_skb(skb);
1510 				goto error;
1511 			}
1512 
1513 			offset += copy;
1514 			length -= datalen - fraggap;
1515 			transhdrlen = 0;
1516 			exthdrlen = 0;
1517 			dst_exthdrlen = 0;
1518 
1519 			/*
1520 			 * Put the packet on the pending queue
1521 			 */
1522 			__skb_queue_tail(queue, skb);
1523 			continue;
1524 		}
1525 
1526 		if (copy > length)
1527 			copy = length;
1528 
1529 		if (!(rt->dst.dev->features&NETIF_F_SG)) {
1530 			unsigned int off;
1531 
1532 			off = skb->len;
1533 			if (getfrag(from, skb_put(skb, copy),
1534 						offset, copy, off, skb) < 0) {
1535 				__skb_trim(skb, off);
1536 				err = -EFAULT;
1537 				goto error;
1538 			}
1539 		} else {
1540 			int i = skb_shinfo(skb)->nr_frags;
1541 
1542 			err = -ENOMEM;
1543 			if (!sk_page_frag_refill(sk, pfrag))
1544 				goto error;
1545 
1546 			if (!skb_can_coalesce(skb, i, pfrag->page,
1547 					      pfrag->offset)) {
1548 				err = -EMSGSIZE;
1549 				if (i == MAX_SKB_FRAGS)
1550 					goto error;
1551 
1552 				__skb_fill_page_desc(skb, i, pfrag->page,
1553 						     pfrag->offset, 0);
1554 				skb_shinfo(skb)->nr_frags = ++i;
1555 				get_page(pfrag->page);
1556 			}
1557 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1558 			if (getfrag(from,
1559 				    page_address(pfrag->page) + pfrag->offset,
1560 				    offset, copy, skb->len, skb) < 0)
1561 				goto error_efault;
1562 
1563 			pfrag->offset += copy;
1564 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1565 			skb->len += copy;
1566 			skb->data_len += copy;
1567 			skb->truesize += copy;
1568 			atomic_add(copy, &sk->sk_wmem_alloc);
1569 		}
1570 		offset += copy;
1571 		length -= copy;
1572 	}
1573 
1574 	return 0;
1575 
1576 error_efault:
1577 	err = -EFAULT;
1578 error:
1579 	cork->length -= length;
1580 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1581 	return err;
1582 }
1583 
1584 int ip6_append_data(struct sock *sk,
1585 		    int getfrag(void *from, char *to, int offset, int len,
1586 				int odd, struct sk_buff *skb),
1587 		    void *from, int length, int transhdrlen,
1588 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1589 		    struct rt6_info *rt, unsigned int flags,
1590 		    const struct sockcm_cookie *sockc)
1591 {
1592 	struct inet_sock *inet = inet_sk(sk);
1593 	struct ipv6_pinfo *np = inet6_sk(sk);
1594 	int exthdrlen;
1595 	int err;
1596 
1597 	if (flags&MSG_PROBE)
1598 		return 0;
1599 	if (skb_queue_empty(&sk->sk_write_queue)) {
1600 		/*
1601 		 * setup for corking
1602 		 */
1603 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1604 				     ipc6, rt, fl6);
1605 		if (err)
1606 			return err;
1607 
1608 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1609 		length += exthdrlen;
1610 		transhdrlen += exthdrlen;
1611 	} else {
1612 		fl6 = &inet->cork.fl.u.ip6;
1613 		transhdrlen = 0;
1614 	}
1615 
1616 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1617 				 &np->cork, sk_page_frag(sk), getfrag,
1618 				 from, length, transhdrlen, flags, ipc6, sockc);
1619 }
1620 EXPORT_SYMBOL_GPL(ip6_append_data);
1621 
1622 static void ip6_cork_release(struct inet_cork_full *cork,
1623 			     struct inet6_cork *v6_cork)
1624 {
1625 	if (v6_cork->opt) {
1626 		kfree(v6_cork->opt->dst0opt);
1627 		kfree(v6_cork->opt->dst1opt);
1628 		kfree(v6_cork->opt->hopopt);
1629 		kfree(v6_cork->opt->srcrt);
1630 		kfree(v6_cork->opt);
1631 		v6_cork->opt = NULL;
1632 	}
1633 
1634 	if (cork->base.dst) {
1635 		dst_release(cork->base.dst);
1636 		cork->base.dst = NULL;
1637 		cork->base.flags &= ~IPCORK_ALLFRAG;
1638 	}
1639 	memset(&cork->fl, 0, sizeof(cork->fl));
1640 }
1641 
1642 struct sk_buff *__ip6_make_skb(struct sock *sk,
1643 			       struct sk_buff_head *queue,
1644 			       struct inet_cork_full *cork,
1645 			       struct inet6_cork *v6_cork)
1646 {
1647 	struct sk_buff *skb, *tmp_skb;
1648 	struct sk_buff **tail_skb;
1649 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1650 	struct ipv6_pinfo *np = inet6_sk(sk);
1651 	struct net *net = sock_net(sk);
1652 	struct ipv6hdr *hdr;
1653 	struct ipv6_txoptions *opt = v6_cork->opt;
1654 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1655 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1656 	unsigned char proto = fl6->flowi6_proto;
1657 
1658 	skb = __skb_dequeue(queue);
1659 	if (!skb)
1660 		goto out;
1661 	tail_skb = &(skb_shinfo(skb)->frag_list);
1662 
1663 	/* move skb->data to ip header from ext header */
1664 	if (skb->data < skb_network_header(skb))
1665 		__skb_pull(skb, skb_network_offset(skb));
1666 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1667 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1668 		*tail_skb = tmp_skb;
1669 		tail_skb = &(tmp_skb->next);
1670 		skb->len += tmp_skb->len;
1671 		skb->data_len += tmp_skb->len;
1672 		skb->truesize += tmp_skb->truesize;
1673 		tmp_skb->destructor = NULL;
1674 		tmp_skb->sk = NULL;
1675 	}
1676 
1677 	/* Allow local fragmentation. */
1678 	skb->ignore_df = ip6_sk_ignore_df(sk);
1679 
1680 	*final_dst = fl6->daddr;
1681 	__skb_pull(skb, skb_network_header_len(skb));
1682 	if (opt && opt->opt_flen)
1683 		ipv6_push_frag_opts(skb, opt, &proto);
1684 	if (opt && opt->opt_nflen)
1685 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1686 
1687 	skb_push(skb, sizeof(struct ipv6hdr));
1688 	skb_reset_network_header(skb);
1689 	hdr = ipv6_hdr(skb);
1690 
1691 	ip6_flow_hdr(hdr, v6_cork->tclass,
1692 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1693 					np->autoflowlabel, fl6));
1694 	hdr->hop_limit = v6_cork->hop_limit;
1695 	hdr->nexthdr = proto;
1696 	hdr->saddr = fl6->saddr;
1697 	hdr->daddr = *final_dst;
1698 
1699 	skb->priority = sk->sk_priority;
1700 	skb->mark = sk->sk_mark;
1701 
1702 	skb_dst_set(skb, dst_clone(&rt->dst));
1703 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1704 	if (proto == IPPROTO_ICMPV6) {
1705 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1706 
1707 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1708 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1709 	}
1710 
1711 	ip6_cork_release(cork, v6_cork);
1712 out:
1713 	return skb;
1714 }
1715 
1716 int ip6_send_skb(struct sk_buff *skb)
1717 {
1718 	struct net *net = sock_net(skb->sk);
1719 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1720 	int err;
1721 
1722 	err = ip6_local_out(net, skb->sk, skb);
1723 	if (err) {
1724 		if (err > 0)
1725 			err = net_xmit_errno(err);
1726 		if (err)
1727 			IP6_INC_STATS(net, rt->rt6i_idev,
1728 				      IPSTATS_MIB_OUTDISCARDS);
1729 	}
1730 
1731 	return err;
1732 }
1733 
1734 int ip6_push_pending_frames(struct sock *sk)
1735 {
1736 	struct sk_buff *skb;
1737 
1738 	skb = ip6_finish_skb(sk);
1739 	if (!skb)
1740 		return 0;
1741 
1742 	return ip6_send_skb(skb);
1743 }
1744 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1745 
1746 static void __ip6_flush_pending_frames(struct sock *sk,
1747 				       struct sk_buff_head *queue,
1748 				       struct inet_cork_full *cork,
1749 				       struct inet6_cork *v6_cork)
1750 {
1751 	struct sk_buff *skb;
1752 
1753 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1754 		if (skb_dst(skb))
1755 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1756 				      IPSTATS_MIB_OUTDISCARDS);
1757 		kfree_skb(skb);
1758 	}
1759 
1760 	ip6_cork_release(cork, v6_cork);
1761 }
1762 
1763 void ip6_flush_pending_frames(struct sock *sk)
1764 {
1765 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1766 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1767 }
1768 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1769 
1770 struct sk_buff *ip6_make_skb(struct sock *sk,
1771 			     int getfrag(void *from, char *to, int offset,
1772 					 int len, int odd, struct sk_buff *skb),
1773 			     void *from, int length, int transhdrlen,
1774 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1775 			     struct rt6_info *rt, unsigned int flags,
1776 			     const struct sockcm_cookie *sockc)
1777 {
1778 	struct inet_cork_full cork;
1779 	struct inet6_cork v6_cork;
1780 	struct sk_buff_head queue;
1781 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1782 	int err;
1783 
1784 	if (flags & MSG_PROBE)
1785 		return NULL;
1786 
1787 	__skb_queue_head_init(&queue);
1788 
1789 	cork.base.flags = 0;
1790 	cork.base.addr = 0;
1791 	cork.base.opt = NULL;
1792 	v6_cork.opt = NULL;
1793 	err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
1794 	if (err)
1795 		return ERR_PTR(err);
1796 
1797 	if (ipc6->dontfrag < 0)
1798 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1799 
1800 	err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1801 				&current->task_frag, getfrag, from,
1802 				length + exthdrlen, transhdrlen + exthdrlen,
1803 				flags, ipc6, sockc);
1804 	if (err) {
1805 		__ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1806 		return ERR_PTR(err);
1807 	}
1808 
1809 	return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1810 }
1811