xref: /linux/net/ipv6/ip6_output.c (revision 81ee0eb6c0fe34490ed92667538197d9295e899e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	IPv6 output functions
4  *	Linux INET6 implementation
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Based on linux/net/ipv4/ip_output.c
10  *
11  *	Changes:
12  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
13  *				extension headers are implemented.
14  *				route changes now work.
15  *				ip6_forward does not confuse sniffers.
16  *				etc.
17  *
18  *      H. von Brand    :       Added missing #include <linux/string.h>
19  *	Imran Patel	:	frag id should be in NBO
20  *      Kazunori MIYAZAWA @USAGI
21  *			:       add ip6_append_data and related functions
22  *				for datagram xmit
23  */
24 
25 #include <linux/errno.h>
26 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/socket.h>
29 #include <linux/net.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/in6.h>
33 #include <linux/tcp.h>
34 #include <linux/route.h>
35 #include <linux/module.h>
36 #include <linux/slab.h>
37 
38 #include <linux/bpf-cgroup.h>
39 #include <linux/netfilter.h>
40 #include <linux/netfilter_ipv6.h>
41 
42 #include <net/sock.h>
43 #include <net/snmp.h>
44 
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 #include <net/protocol.h>
48 #include <net/ip6_route.h>
49 #include <net/addrconf.h>
50 #include <net/rawv6.h>
51 #include <net/icmp.h>
52 #include <net/xfrm.h>
53 #include <net/checksum.h>
54 #include <linux/mroute6.h>
55 #include <net/l3mdev.h>
56 #include <net/lwtunnel.h>
57 #include <net/ip_tunnels.h>
58 
59 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
60 {
61 	struct dst_entry *dst = skb_dst(skb);
62 	struct net_device *dev = dst->dev;
63 	struct inet6_dev *idev = ip6_dst_idev(dst);
64 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
65 	const struct in6_addr *daddr, *nexthop;
66 	struct ipv6hdr *hdr;
67 	struct neighbour *neigh;
68 	int ret;
69 
70 	/* Be paranoid, rather than too clever. */
71 	if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
72 		skb = skb_expand_head(skb, hh_len);
73 		if (!skb) {
74 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
75 			return -ENOMEM;
76 		}
77 	}
78 
79 	hdr = ipv6_hdr(skb);
80 	daddr = &hdr->daddr;
81 	if (ipv6_addr_is_multicast(daddr)) {
82 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
83 		    ((mroute6_is_socket(net, skb) &&
84 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
85 		     ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
86 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
87 
88 			/* Do not check for IFF_ALLMULTI; multicast routing
89 			   is not supported in any case.
90 			 */
91 			if (newskb)
92 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
93 					net, sk, newskb, NULL, newskb->dev,
94 					dev_loopback_xmit);
95 
96 			if (hdr->hop_limit == 0) {
97 				IP6_INC_STATS(net, idev,
98 					      IPSTATS_MIB_OUTDISCARDS);
99 				kfree_skb(skb);
100 				return 0;
101 			}
102 		}
103 
104 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
105 		if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
106 		    !(dev->flags & IFF_LOOPBACK)) {
107 			kfree_skb(skb);
108 			return 0;
109 		}
110 	}
111 
112 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
113 		int res = lwtunnel_xmit(skb);
114 
115 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
116 			return res;
117 	}
118 
119 	rcu_read_lock_bh();
120 	nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
121 	neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
122 	if (unlikely(!neigh))
123 		neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
124 	if (!IS_ERR(neigh)) {
125 		sock_confirm_neigh(skb, neigh);
126 		ret = neigh_output(neigh, skb, false);
127 		rcu_read_unlock_bh();
128 		return ret;
129 	}
130 	rcu_read_unlock_bh();
131 
132 	IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
133 	kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
134 	return -EINVAL;
135 }
136 
137 static int
138 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
139 				    struct sk_buff *skb, unsigned int mtu)
140 {
141 	struct sk_buff *segs, *nskb;
142 	netdev_features_t features;
143 	int ret = 0;
144 
145 	/* Please see corresponding comment in ip_finish_output_gso
146 	 * describing the cases where GSO segment length exceeds the
147 	 * egress MTU.
148 	 */
149 	features = netif_skb_features(skb);
150 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
151 	if (IS_ERR_OR_NULL(segs)) {
152 		kfree_skb(skb);
153 		return -ENOMEM;
154 	}
155 
156 	consume_skb(skb);
157 
158 	skb_list_walk_safe(segs, segs, nskb) {
159 		int err;
160 
161 		skb_mark_not_on_list(segs);
162 		err = ip6_fragment(net, sk, segs, ip6_finish_output2);
163 		if (err && ret == 0)
164 			ret = err;
165 	}
166 
167 	return ret;
168 }
169 
170 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
171 {
172 	unsigned int mtu;
173 
174 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
175 	/* Policy lookup after SNAT yielded a new policy */
176 	if (skb_dst(skb)->xfrm) {
177 		IP6CB(skb)->flags |= IP6SKB_REROUTED;
178 		return dst_output(net, sk, skb);
179 	}
180 #endif
181 
182 	mtu = ip6_skb_dst_mtu(skb);
183 	if (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))
184 		return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
185 
186 	if ((skb->len > mtu && !skb_is_gso(skb)) ||
187 	    dst_allfrag(skb_dst(skb)) ||
188 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
189 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
190 	else
191 		return ip6_finish_output2(net, sk, skb);
192 }
193 
194 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
195 {
196 	int ret;
197 
198 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
199 	switch (ret) {
200 	case NET_XMIT_SUCCESS:
201 		return __ip6_finish_output(net, sk, skb);
202 	case NET_XMIT_CN:
203 		return __ip6_finish_output(net, sk, skb) ? : ret;
204 	default:
205 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
206 		return ret;
207 	}
208 }
209 
210 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
211 {
212 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
213 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
214 
215 	skb->protocol = htons(ETH_P_IPV6);
216 	skb->dev = dev;
217 
218 	if (unlikely(idev->cnf.disable_ipv6)) {
219 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
220 		kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
221 		return 0;
222 	}
223 
224 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
225 			    net, sk, skb, indev, dev,
226 			    ip6_finish_output,
227 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
228 }
229 EXPORT_SYMBOL(ip6_output);
230 
231 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
232 {
233 	if (!np->autoflowlabel_set)
234 		return ip6_default_np_autolabel(net);
235 	else
236 		return np->autoflowlabel;
237 }
238 
239 /*
240  * xmit an sk_buff (used by TCP, SCTP and DCCP)
241  * Note : socket lock is not held for SYNACK packets, but might be modified
242  * by calls to skb_set_owner_w() and ipv6_local_error(),
243  * which are using proper atomic operations or spinlocks.
244  */
245 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
246 	     __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
247 {
248 	struct net *net = sock_net(sk);
249 	const struct ipv6_pinfo *np = inet6_sk(sk);
250 	struct in6_addr *first_hop = &fl6->daddr;
251 	struct dst_entry *dst = skb_dst(skb);
252 	struct net_device *dev = dst->dev;
253 	struct inet6_dev *idev = ip6_dst_idev(dst);
254 	unsigned int head_room;
255 	struct ipv6hdr *hdr;
256 	u8  proto = fl6->flowi6_proto;
257 	int seg_len = skb->len;
258 	int hlimit = -1;
259 	u32 mtu;
260 
261 	head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dev);
262 	if (opt)
263 		head_room += opt->opt_nflen + opt->opt_flen;
264 
265 	if (unlikely(head_room > skb_headroom(skb))) {
266 		skb = skb_expand_head(skb, head_room);
267 		if (!skb) {
268 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
269 			return -ENOBUFS;
270 		}
271 	}
272 
273 	if (opt) {
274 		seg_len += opt->opt_nflen + opt->opt_flen;
275 
276 		if (opt->opt_flen)
277 			ipv6_push_frag_opts(skb, opt, &proto);
278 
279 		if (opt->opt_nflen)
280 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
281 					     &fl6->saddr);
282 	}
283 
284 	skb_push(skb, sizeof(struct ipv6hdr));
285 	skb_reset_network_header(skb);
286 	hdr = ipv6_hdr(skb);
287 
288 	/*
289 	 *	Fill in the IPv6 header
290 	 */
291 	if (np)
292 		hlimit = np->hop_limit;
293 	if (hlimit < 0)
294 		hlimit = ip6_dst_hoplimit(dst);
295 
296 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
297 				ip6_autoflowlabel(net, np), fl6));
298 
299 	hdr->payload_len = htons(seg_len);
300 	hdr->nexthdr = proto;
301 	hdr->hop_limit = hlimit;
302 
303 	hdr->saddr = fl6->saddr;
304 	hdr->daddr = *first_hop;
305 
306 	skb->protocol = htons(ETH_P_IPV6);
307 	skb->priority = priority;
308 	skb->mark = mark;
309 
310 	mtu = dst_mtu(dst);
311 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
312 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
313 
314 		/* if egress device is enslaved to an L3 master device pass the
315 		 * skb to its handler for processing
316 		 */
317 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
318 		if (unlikely(!skb))
319 			return 0;
320 
321 		/* hooks should never assume socket lock is held.
322 		 * we promote our socket to non const
323 		 */
324 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
325 			       net, (struct sock *)sk, skb, NULL, dev,
326 			       dst_output);
327 	}
328 
329 	skb->dev = dev;
330 	/* ipv6_local_error() does not require socket lock,
331 	 * we promote our socket to non const
332 	 */
333 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
334 
335 	IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
336 	kfree_skb(skb);
337 	return -EMSGSIZE;
338 }
339 EXPORT_SYMBOL(ip6_xmit);
340 
341 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
342 {
343 	struct ip6_ra_chain *ra;
344 	struct sock *last = NULL;
345 
346 	read_lock(&ip6_ra_lock);
347 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
348 		struct sock *sk = ra->sk;
349 		if (sk && ra->sel == sel &&
350 		    (!sk->sk_bound_dev_if ||
351 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
352 			struct ipv6_pinfo *np = inet6_sk(sk);
353 
354 			if (np && np->rtalert_isolate &&
355 			    !net_eq(sock_net(sk), dev_net(skb->dev))) {
356 				continue;
357 			}
358 			if (last) {
359 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
360 				if (skb2)
361 					rawv6_rcv(last, skb2);
362 			}
363 			last = sk;
364 		}
365 	}
366 
367 	if (last) {
368 		rawv6_rcv(last, skb);
369 		read_unlock(&ip6_ra_lock);
370 		return 1;
371 	}
372 	read_unlock(&ip6_ra_lock);
373 	return 0;
374 }
375 
376 static int ip6_forward_proxy_check(struct sk_buff *skb)
377 {
378 	struct ipv6hdr *hdr = ipv6_hdr(skb);
379 	u8 nexthdr = hdr->nexthdr;
380 	__be16 frag_off;
381 	int offset;
382 
383 	if (ipv6_ext_hdr(nexthdr)) {
384 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
385 		if (offset < 0)
386 			return 0;
387 	} else
388 		offset = sizeof(struct ipv6hdr);
389 
390 	if (nexthdr == IPPROTO_ICMPV6) {
391 		struct icmp6hdr *icmp6;
392 
393 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
394 					 offset + 1 - skb->data)))
395 			return 0;
396 
397 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
398 
399 		switch (icmp6->icmp6_type) {
400 		case NDISC_ROUTER_SOLICITATION:
401 		case NDISC_ROUTER_ADVERTISEMENT:
402 		case NDISC_NEIGHBOUR_SOLICITATION:
403 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
404 		case NDISC_REDIRECT:
405 			/* For reaction involving unicast neighbor discovery
406 			 * message destined to the proxied address, pass it to
407 			 * input function.
408 			 */
409 			return 1;
410 		default:
411 			break;
412 		}
413 	}
414 
415 	/*
416 	 * The proxying router can't forward traffic sent to a link-local
417 	 * address, so signal the sender and discard the packet. This
418 	 * behavior is clarified by the MIPv6 specification.
419 	 */
420 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
421 		dst_link_failure(skb);
422 		return -1;
423 	}
424 
425 	return 0;
426 }
427 
428 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
429 				     struct sk_buff *skb)
430 {
431 	struct dst_entry *dst = skb_dst(skb);
432 
433 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
434 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
435 
436 #ifdef CONFIG_NET_SWITCHDEV
437 	if (skb->offload_l3_fwd_mark) {
438 		consume_skb(skb);
439 		return 0;
440 	}
441 #endif
442 
443 	skb_clear_tstamp(skb);
444 	return dst_output(net, sk, skb);
445 }
446 
447 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
448 {
449 	if (skb->len <= mtu)
450 		return false;
451 
452 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
453 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
454 		return true;
455 
456 	if (skb->ignore_df)
457 		return false;
458 
459 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
460 		return false;
461 
462 	return true;
463 }
464 
465 int ip6_forward(struct sk_buff *skb)
466 {
467 	struct dst_entry *dst = skb_dst(skb);
468 	struct ipv6hdr *hdr = ipv6_hdr(skb);
469 	struct inet6_skb_parm *opt = IP6CB(skb);
470 	struct net *net = dev_net(dst->dev);
471 	struct inet6_dev *idev;
472 	SKB_DR(reason);
473 	u32 mtu;
474 
475 	idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
476 	if (net->ipv6.devconf_all->forwarding == 0)
477 		goto error;
478 
479 	if (skb->pkt_type != PACKET_HOST)
480 		goto drop;
481 
482 	if (unlikely(skb->sk))
483 		goto drop;
484 
485 	if (skb_warn_if_lro(skb))
486 		goto drop;
487 
488 	if (!net->ipv6.devconf_all->disable_policy &&
489 	    (!idev || !idev->cnf.disable_policy) &&
490 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
491 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
492 		goto drop;
493 	}
494 
495 	skb_forward_csum(skb);
496 
497 	/*
498 	 *	We DO NOT make any processing on
499 	 *	RA packets, pushing them to user level AS IS
500 	 *	without ane WARRANTY that application will be able
501 	 *	to interpret them. The reason is that we
502 	 *	cannot make anything clever here.
503 	 *
504 	 *	We are not end-node, so that if packet contains
505 	 *	AH/ESP, we cannot make anything.
506 	 *	Defragmentation also would be mistake, RA packets
507 	 *	cannot be fragmented, because there is no warranty
508 	 *	that different fragments will go along one path. --ANK
509 	 */
510 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
511 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
512 			return 0;
513 	}
514 
515 	/*
516 	 *	check and decrement ttl
517 	 */
518 	if (hdr->hop_limit <= 1) {
519 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
520 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
521 
522 		kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
523 		return -ETIMEDOUT;
524 	}
525 
526 	/* XXX: idev->cnf.proxy_ndp? */
527 	if (net->ipv6.devconf_all->proxy_ndp &&
528 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
529 		int proxied = ip6_forward_proxy_check(skb);
530 		if (proxied > 0) {
531 			hdr->hop_limit--;
532 			return ip6_input(skb);
533 		} else if (proxied < 0) {
534 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
535 			goto drop;
536 		}
537 	}
538 
539 	if (!xfrm6_route_forward(skb)) {
540 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
541 		SKB_DR_SET(reason, XFRM_POLICY);
542 		goto drop;
543 	}
544 	dst = skb_dst(skb);
545 
546 	/* IPv6 specs say nothing about it, but it is clear that we cannot
547 	   send redirects to source routed frames.
548 	   We don't send redirects to frames decapsulated from IPsec.
549 	 */
550 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
551 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
552 		struct in6_addr *target = NULL;
553 		struct inet_peer *peer;
554 		struct rt6_info *rt;
555 
556 		/*
557 		 *	incoming and outgoing devices are the same
558 		 *	send a redirect.
559 		 */
560 
561 		rt = (struct rt6_info *) dst;
562 		if (rt->rt6i_flags & RTF_GATEWAY)
563 			target = &rt->rt6i_gateway;
564 		else
565 			target = &hdr->daddr;
566 
567 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
568 
569 		/* Limit redirects both by destination (here)
570 		   and by source (inside ndisc_send_redirect)
571 		 */
572 		if (inet_peer_xrlim_allow(peer, 1*HZ))
573 			ndisc_send_redirect(skb, target);
574 		if (peer)
575 			inet_putpeer(peer);
576 	} else {
577 		int addrtype = ipv6_addr_type(&hdr->saddr);
578 
579 		/* This check is security critical. */
580 		if (addrtype == IPV6_ADDR_ANY ||
581 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
582 			goto error;
583 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
584 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
585 				    ICMPV6_NOT_NEIGHBOUR, 0);
586 			goto error;
587 		}
588 	}
589 
590 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
591 	if (mtu < IPV6_MIN_MTU)
592 		mtu = IPV6_MIN_MTU;
593 
594 	if (ip6_pkt_too_big(skb, mtu)) {
595 		/* Again, force OUTPUT device used as source address */
596 		skb->dev = dst->dev;
597 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
598 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
599 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
600 				IPSTATS_MIB_FRAGFAILS);
601 		kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
602 		return -EMSGSIZE;
603 	}
604 
605 	if (skb_cow(skb, dst->dev->hard_header_len)) {
606 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
607 				IPSTATS_MIB_OUTDISCARDS);
608 		goto drop;
609 	}
610 
611 	hdr = ipv6_hdr(skb);
612 
613 	/* Mangling hops number delayed to point after skb COW */
614 
615 	hdr->hop_limit--;
616 
617 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
618 		       net, NULL, skb, skb->dev, dst->dev,
619 		       ip6_forward_finish);
620 
621 error:
622 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
623 	SKB_DR_SET(reason, IP_INADDRERRORS);
624 drop:
625 	kfree_skb_reason(skb, reason);
626 	return -EINVAL;
627 }
628 
629 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
630 {
631 	to->pkt_type = from->pkt_type;
632 	to->priority = from->priority;
633 	to->protocol = from->protocol;
634 	skb_dst_drop(to);
635 	skb_dst_set(to, dst_clone(skb_dst(from)));
636 	to->dev = from->dev;
637 	to->mark = from->mark;
638 
639 	skb_copy_hash(to, from);
640 
641 #ifdef CONFIG_NET_SCHED
642 	to->tc_index = from->tc_index;
643 #endif
644 	nf_copy(to, from);
645 	skb_ext_copy(to, from);
646 	skb_copy_secmark(to, from);
647 }
648 
649 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
650 		      u8 nexthdr, __be32 frag_id,
651 		      struct ip6_fraglist_iter *iter)
652 {
653 	unsigned int first_len;
654 	struct frag_hdr *fh;
655 
656 	/* BUILD HEADER */
657 	*prevhdr = NEXTHDR_FRAGMENT;
658 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
659 	if (!iter->tmp_hdr)
660 		return -ENOMEM;
661 
662 	iter->frag = skb_shinfo(skb)->frag_list;
663 	skb_frag_list_init(skb);
664 
665 	iter->offset = 0;
666 	iter->hlen = hlen;
667 	iter->frag_id = frag_id;
668 	iter->nexthdr = nexthdr;
669 
670 	__skb_pull(skb, hlen);
671 	fh = __skb_push(skb, sizeof(struct frag_hdr));
672 	__skb_push(skb, hlen);
673 	skb_reset_network_header(skb);
674 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
675 
676 	fh->nexthdr = nexthdr;
677 	fh->reserved = 0;
678 	fh->frag_off = htons(IP6_MF);
679 	fh->identification = frag_id;
680 
681 	first_len = skb_pagelen(skb);
682 	skb->data_len = first_len - skb_headlen(skb);
683 	skb->len = first_len;
684 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
685 
686 	return 0;
687 }
688 EXPORT_SYMBOL(ip6_fraglist_init);
689 
690 void ip6_fraglist_prepare(struct sk_buff *skb,
691 			  struct ip6_fraglist_iter *iter)
692 {
693 	struct sk_buff *frag = iter->frag;
694 	unsigned int hlen = iter->hlen;
695 	struct frag_hdr *fh;
696 
697 	frag->ip_summed = CHECKSUM_NONE;
698 	skb_reset_transport_header(frag);
699 	fh = __skb_push(frag, sizeof(struct frag_hdr));
700 	__skb_push(frag, hlen);
701 	skb_reset_network_header(frag);
702 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
703 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
704 	fh->nexthdr = iter->nexthdr;
705 	fh->reserved = 0;
706 	fh->frag_off = htons(iter->offset);
707 	if (frag->next)
708 		fh->frag_off |= htons(IP6_MF);
709 	fh->identification = iter->frag_id;
710 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
711 	ip6_copy_metadata(frag, skb);
712 }
713 EXPORT_SYMBOL(ip6_fraglist_prepare);
714 
715 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
716 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
717 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
718 {
719 	state->prevhdr = prevhdr;
720 	state->nexthdr = nexthdr;
721 	state->frag_id = frag_id;
722 
723 	state->hlen = hlen;
724 	state->mtu = mtu;
725 
726 	state->left = skb->len - hlen;	/* Space per frame */
727 	state->ptr = hlen;		/* Where to start from */
728 
729 	state->hroom = hdr_room;
730 	state->troom = needed_tailroom;
731 
732 	state->offset = 0;
733 }
734 EXPORT_SYMBOL(ip6_frag_init);
735 
736 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
737 {
738 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
739 	struct sk_buff *frag;
740 	struct frag_hdr *fh;
741 	unsigned int len;
742 
743 	len = state->left;
744 	/* IF: it doesn't fit, use 'mtu' - the data space left */
745 	if (len > state->mtu)
746 		len = state->mtu;
747 	/* IF: we are not sending up to and including the packet end
748 	   then align the next start on an eight byte boundary */
749 	if (len < state->left)
750 		len &= ~7;
751 
752 	/* Allocate buffer */
753 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
754 			 state->hroom + state->troom, GFP_ATOMIC);
755 	if (!frag)
756 		return ERR_PTR(-ENOMEM);
757 
758 	/*
759 	 *	Set up data on packet
760 	 */
761 
762 	ip6_copy_metadata(frag, skb);
763 	skb_reserve(frag, state->hroom);
764 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
765 	skb_reset_network_header(frag);
766 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
767 	frag->transport_header = (frag->network_header + state->hlen +
768 				  sizeof(struct frag_hdr));
769 
770 	/*
771 	 *	Charge the memory for the fragment to any owner
772 	 *	it might possess
773 	 */
774 	if (skb->sk)
775 		skb_set_owner_w(frag, skb->sk);
776 
777 	/*
778 	 *	Copy the packet header into the new buffer.
779 	 */
780 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
781 
782 	fragnexthdr_offset = skb_network_header(frag);
783 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
784 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
785 
786 	/*
787 	 *	Build fragment header.
788 	 */
789 	fh->nexthdr = state->nexthdr;
790 	fh->reserved = 0;
791 	fh->identification = state->frag_id;
792 
793 	/*
794 	 *	Copy a block of the IP datagram.
795 	 */
796 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
797 			     len));
798 	state->left -= len;
799 
800 	fh->frag_off = htons(state->offset);
801 	if (state->left > 0)
802 		fh->frag_off |= htons(IP6_MF);
803 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
804 
805 	state->ptr += len;
806 	state->offset += len;
807 
808 	return frag;
809 }
810 EXPORT_SYMBOL(ip6_frag_next);
811 
812 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
813 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
814 {
815 	struct sk_buff *frag;
816 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
817 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
818 				inet6_sk(skb->sk) : NULL;
819 	bool mono_delivery_time = skb->mono_delivery_time;
820 	struct ip6_frag_state state;
821 	unsigned int mtu, hlen, nexthdr_offset;
822 	ktime_t tstamp = skb->tstamp;
823 	int hroom, err = 0;
824 	__be32 frag_id;
825 	u8 *prevhdr, nexthdr = 0;
826 
827 	err = ip6_find_1stfragopt(skb, &prevhdr);
828 	if (err < 0)
829 		goto fail;
830 	hlen = err;
831 	nexthdr = *prevhdr;
832 	nexthdr_offset = prevhdr - skb_network_header(skb);
833 
834 	mtu = ip6_skb_dst_mtu(skb);
835 
836 	/* We must not fragment if the socket is set to force MTU discovery
837 	 * or if the skb it not generated by a local socket.
838 	 */
839 	if (unlikely(!skb->ignore_df && skb->len > mtu))
840 		goto fail_toobig;
841 
842 	if (IP6CB(skb)->frag_max_size) {
843 		if (IP6CB(skb)->frag_max_size > mtu)
844 			goto fail_toobig;
845 
846 		/* don't send fragments larger than what we received */
847 		mtu = IP6CB(skb)->frag_max_size;
848 		if (mtu < IPV6_MIN_MTU)
849 			mtu = IPV6_MIN_MTU;
850 	}
851 
852 	if (np && np->frag_size < mtu) {
853 		if (np->frag_size)
854 			mtu = np->frag_size;
855 	}
856 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
857 		goto fail_toobig;
858 	mtu -= hlen + sizeof(struct frag_hdr);
859 
860 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
861 				    &ipv6_hdr(skb)->saddr);
862 
863 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
864 	    (err = skb_checksum_help(skb)))
865 		goto fail;
866 
867 	prevhdr = skb_network_header(skb) + nexthdr_offset;
868 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
869 	if (skb_has_frag_list(skb)) {
870 		unsigned int first_len = skb_pagelen(skb);
871 		struct ip6_fraglist_iter iter;
872 		struct sk_buff *frag2;
873 
874 		if (first_len - hlen > mtu ||
875 		    ((first_len - hlen) & 7) ||
876 		    skb_cloned(skb) ||
877 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
878 			goto slow_path;
879 
880 		skb_walk_frags(skb, frag) {
881 			/* Correct geometry. */
882 			if (frag->len > mtu ||
883 			    ((frag->len & 7) && frag->next) ||
884 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
885 				goto slow_path_clean;
886 
887 			/* Partially cloned skb? */
888 			if (skb_shared(frag))
889 				goto slow_path_clean;
890 
891 			BUG_ON(frag->sk);
892 			if (skb->sk) {
893 				frag->sk = skb->sk;
894 				frag->destructor = sock_wfree;
895 			}
896 			skb->truesize -= frag->truesize;
897 		}
898 
899 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
900 					&iter);
901 		if (err < 0)
902 			goto fail;
903 
904 		for (;;) {
905 			/* Prepare header of the next frame,
906 			 * before previous one went down. */
907 			if (iter.frag)
908 				ip6_fraglist_prepare(skb, &iter);
909 
910 			skb_set_delivery_time(skb, tstamp, mono_delivery_time);
911 			err = output(net, sk, skb);
912 			if (!err)
913 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
914 					      IPSTATS_MIB_FRAGCREATES);
915 
916 			if (err || !iter.frag)
917 				break;
918 
919 			skb = ip6_fraglist_next(&iter);
920 		}
921 
922 		kfree(iter.tmp_hdr);
923 
924 		if (err == 0) {
925 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
926 				      IPSTATS_MIB_FRAGOKS);
927 			return 0;
928 		}
929 
930 		kfree_skb_list(iter.frag);
931 
932 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
933 			      IPSTATS_MIB_FRAGFAILS);
934 		return err;
935 
936 slow_path_clean:
937 		skb_walk_frags(skb, frag2) {
938 			if (frag2 == frag)
939 				break;
940 			frag2->sk = NULL;
941 			frag2->destructor = NULL;
942 			skb->truesize += frag2->truesize;
943 		}
944 	}
945 
946 slow_path:
947 	/*
948 	 *	Fragment the datagram.
949 	 */
950 
951 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
952 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
953 		      &state);
954 
955 	/*
956 	 *	Keep copying data until we run out.
957 	 */
958 
959 	while (state.left > 0) {
960 		frag = ip6_frag_next(skb, &state);
961 		if (IS_ERR(frag)) {
962 			err = PTR_ERR(frag);
963 			goto fail;
964 		}
965 
966 		/*
967 		 *	Put this fragment into the sending queue.
968 		 */
969 		skb_set_delivery_time(frag, tstamp, mono_delivery_time);
970 		err = output(net, sk, frag);
971 		if (err)
972 			goto fail;
973 
974 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
975 			      IPSTATS_MIB_FRAGCREATES);
976 	}
977 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
978 		      IPSTATS_MIB_FRAGOKS);
979 	consume_skb(skb);
980 	return err;
981 
982 fail_toobig:
983 	if (skb->sk && dst_allfrag(skb_dst(skb)))
984 		sk_gso_disable(skb->sk);
985 
986 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
987 	err = -EMSGSIZE;
988 
989 fail:
990 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
991 		      IPSTATS_MIB_FRAGFAILS);
992 	kfree_skb(skb);
993 	return err;
994 }
995 
996 static inline int ip6_rt_check(const struct rt6key *rt_key,
997 			       const struct in6_addr *fl_addr,
998 			       const struct in6_addr *addr_cache)
999 {
1000 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1001 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1002 }
1003 
1004 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1005 					  struct dst_entry *dst,
1006 					  const struct flowi6 *fl6)
1007 {
1008 	struct ipv6_pinfo *np = inet6_sk(sk);
1009 	struct rt6_info *rt;
1010 
1011 	if (!dst)
1012 		goto out;
1013 
1014 	if (dst->ops->family != AF_INET6) {
1015 		dst_release(dst);
1016 		return NULL;
1017 	}
1018 
1019 	rt = (struct rt6_info *)dst;
1020 	/* Yes, checking route validity in not connected
1021 	 * case is not very simple. Take into account,
1022 	 * that we do not support routing by source, TOS,
1023 	 * and MSG_DONTROUTE		--ANK (980726)
1024 	 *
1025 	 * 1. ip6_rt_check(): If route was host route,
1026 	 *    check that cached destination is current.
1027 	 *    If it is network route, we still may
1028 	 *    check its validity using saved pointer
1029 	 *    to the last used address: daddr_cache.
1030 	 *    We do not want to save whole address now,
1031 	 *    (because main consumer of this service
1032 	 *    is tcp, which has not this problem),
1033 	 *    so that the last trick works only on connected
1034 	 *    sockets.
1035 	 * 2. oif also should be the same.
1036 	 */
1037 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1038 #ifdef CONFIG_IPV6_SUBTREES
1039 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1040 #endif
1041 	   (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1042 		dst_release(dst);
1043 		dst = NULL;
1044 	}
1045 
1046 out:
1047 	return dst;
1048 }
1049 
1050 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1051 			       struct dst_entry **dst, struct flowi6 *fl6)
1052 {
1053 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1054 	struct neighbour *n;
1055 	struct rt6_info *rt;
1056 #endif
1057 	int err;
1058 	int flags = 0;
1059 
1060 	/* The correct way to handle this would be to do
1061 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1062 	 * the route-specific preferred source forces the
1063 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1064 	 *
1065 	 * In source specific routing (no src=any default route),
1066 	 * ip6_route_output will fail given src=any saddr, though, so
1067 	 * that's why we try it again later.
1068 	 */
1069 	if (ipv6_addr_any(&fl6->saddr)) {
1070 		struct fib6_info *from;
1071 		struct rt6_info *rt;
1072 
1073 		*dst = ip6_route_output(net, sk, fl6);
1074 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1075 
1076 		rcu_read_lock();
1077 		from = rt ? rcu_dereference(rt->from) : NULL;
1078 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1079 					  sk ? inet6_sk(sk)->srcprefs : 0,
1080 					  &fl6->saddr);
1081 		rcu_read_unlock();
1082 
1083 		if (err)
1084 			goto out_err_release;
1085 
1086 		/* If we had an erroneous initial result, pretend it
1087 		 * never existed and let the SA-enabled version take
1088 		 * over.
1089 		 */
1090 		if ((*dst)->error) {
1091 			dst_release(*dst);
1092 			*dst = NULL;
1093 		}
1094 
1095 		if (fl6->flowi6_oif)
1096 			flags |= RT6_LOOKUP_F_IFACE;
1097 	}
1098 
1099 	if (!*dst)
1100 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1101 
1102 	err = (*dst)->error;
1103 	if (err)
1104 		goto out_err_release;
1105 
1106 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1107 	/*
1108 	 * Here if the dst entry we've looked up
1109 	 * has a neighbour entry that is in the INCOMPLETE
1110 	 * state and the src address from the flow is
1111 	 * marked as OPTIMISTIC, we release the found
1112 	 * dst entry and replace it instead with the
1113 	 * dst entry of the nexthop router
1114 	 */
1115 	rt = (struct rt6_info *) *dst;
1116 	rcu_read_lock_bh();
1117 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1118 				      rt6_nexthop(rt, &fl6->daddr));
1119 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1120 	rcu_read_unlock_bh();
1121 
1122 	if (err) {
1123 		struct inet6_ifaddr *ifp;
1124 		struct flowi6 fl_gw6;
1125 		int redirect;
1126 
1127 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1128 				      (*dst)->dev, 1);
1129 
1130 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1131 		if (ifp)
1132 			in6_ifa_put(ifp);
1133 
1134 		if (redirect) {
1135 			/*
1136 			 * We need to get the dst entry for the
1137 			 * default router instead
1138 			 */
1139 			dst_release(*dst);
1140 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1141 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1142 			*dst = ip6_route_output(net, sk, &fl_gw6);
1143 			err = (*dst)->error;
1144 			if (err)
1145 				goto out_err_release;
1146 		}
1147 	}
1148 #endif
1149 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1150 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1151 		err = -EAFNOSUPPORT;
1152 		goto out_err_release;
1153 	}
1154 
1155 	return 0;
1156 
1157 out_err_release:
1158 	dst_release(*dst);
1159 	*dst = NULL;
1160 
1161 	if (err == -ENETUNREACH)
1162 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1163 	return err;
1164 }
1165 
1166 /**
1167  *	ip6_dst_lookup - perform route lookup on flow
1168  *	@net: Network namespace to perform lookup in
1169  *	@sk: socket which provides route info
1170  *	@dst: pointer to dst_entry * for result
1171  *	@fl6: flow to lookup
1172  *
1173  *	This function performs a route lookup on the given flow.
1174  *
1175  *	It returns zero on success, or a standard errno code on error.
1176  */
1177 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1178 		   struct flowi6 *fl6)
1179 {
1180 	*dst = NULL;
1181 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1182 }
1183 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1184 
1185 /**
1186  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1187  *	@net: Network namespace to perform lookup in
1188  *	@sk: socket which provides route info
1189  *	@fl6: flow to lookup
1190  *	@final_dst: final destination address for ipsec lookup
1191  *
1192  *	This function performs a route lookup on the given flow.
1193  *
1194  *	It returns a valid dst pointer on success, or a pointer encoded
1195  *	error code.
1196  */
1197 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1198 				      const struct in6_addr *final_dst)
1199 {
1200 	struct dst_entry *dst = NULL;
1201 	int err;
1202 
1203 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1204 	if (err)
1205 		return ERR_PTR(err);
1206 	if (final_dst)
1207 		fl6->daddr = *final_dst;
1208 
1209 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1210 }
1211 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1212 
1213 /**
1214  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1215  *	@sk: socket which provides the dst cache and route info
1216  *	@fl6: flow to lookup
1217  *	@final_dst: final destination address for ipsec lookup
1218  *	@connected: whether @sk is connected or not
1219  *
1220  *	This function performs a route lookup on the given flow with the
1221  *	possibility of using the cached route in the socket if it is valid.
1222  *	It will take the socket dst lock when operating on the dst cache.
1223  *	As a result, this function can only be used in process context.
1224  *
1225  *	In addition, for a connected socket, cache the dst in the socket
1226  *	if the current cache is not valid.
1227  *
1228  *	It returns a valid dst pointer on success, or a pointer encoded
1229  *	error code.
1230  */
1231 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1232 					 const struct in6_addr *final_dst,
1233 					 bool connected)
1234 {
1235 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1236 
1237 	dst = ip6_sk_dst_check(sk, dst, fl6);
1238 	if (dst)
1239 		return dst;
1240 
1241 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1242 	if (connected && !IS_ERR(dst))
1243 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1244 
1245 	return dst;
1246 }
1247 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1248 
1249 /**
1250  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1251  *      @skb: Packet for which lookup is done
1252  *      @dev: Tunnel device
1253  *      @net: Network namespace of tunnel device
1254  *      @sock: Socket which provides route info
1255  *      @saddr: Memory to store the src ip address
1256  *      @info: Tunnel information
1257  *      @protocol: IP protocol
1258  *      @use_cache: Flag to enable cache usage
1259  *      This function performs a route lookup on a tunnel
1260  *
1261  *      It returns a valid dst pointer and stores src address to be used in
1262  *      tunnel in param saddr on success, else a pointer encoded error code.
1263  */
1264 
1265 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1266 					struct net_device *dev,
1267 					struct net *net,
1268 					struct socket *sock,
1269 					struct in6_addr *saddr,
1270 					const struct ip_tunnel_info *info,
1271 					u8 protocol,
1272 					bool use_cache)
1273 {
1274 	struct dst_entry *dst = NULL;
1275 #ifdef CONFIG_DST_CACHE
1276 	struct dst_cache *dst_cache;
1277 #endif
1278 	struct flowi6 fl6;
1279 	__u8 prio;
1280 
1281 #ifdef CONFIG_DST_CACHE
1282 	dst_cache = (struct dst_cache *)&info->dst_cache;
1283 	if (use_cache) {
1284 		dst = dst_cache_get_ip6(dst_cache, saddr);
1285 		if (dst)
1286 			return dst;
1287 	}
1288 #endif
1289 	memset(&fl6, 0, sizeof(fl6));
1290 	fl6.flowi6_mark = skb->mark;
1291 	fl6.flowi6_proto = protocol;
1292 	fl6.daddr = info->key.u.ipv6.dst;
1293 	fl6.saddr = info->key.u.ipv6.src;
1294 	prio = info->key.tos;
1295 	fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1296 					  info->key.label);
1297 
1298 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1299 					      NULL);
1300 	if (IS_ERR(dst)) {
1301 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1302 		return ERR_PTR(-ENETUNREACH);
1303 	}
1304 	if (dst->dev == dev) { /* is this necessary? */
1305 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1306 		dst_release(dst);
1307 		return ERR_PTR(-ELOOP);
1308 	}
1309 #ifdef CONFIG_DST_CACHE
1310 	if (use_cache)
1311 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1312 #endif
1313 	*saddr = fl6.saddr;
1314 	return dst;
1315 }
1316 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1317 
1318 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1319 					       gfp_t gfp)
1320 {
1321 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1322 }
1323 
1324 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1325 						gfp_t gfp)
1326 {
1327 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1328 }
1329 
1330 static void ip6_append_data_mtu(unsigned int *mtu,
1331 				int *maxfraglen,
1332 				unsigned int fragheaderlen,
1333 				struct sk_buff *skb,
1334 				struct rt6_info *rt,
1335 				unsigned int orig_mtu)
1336 {
1337 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1338 		if (!skb) {
1339 			/* first fragment, reserve header_len */
1340 			*mtu = orig_mtu - rt->dst.header_len;
1341 
1342 		} else {
1343 			/*
1344 			 * this fragment is not first, the headers
1345 			 * space is regarded as data space.
1346 			 */
1347 			*mtu = orig_mtu;
1348 		}
1349 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1350 			      + fragheaderlen - sizeof(struct frag_hdr);
1351 	}
1352 }
1353 
1354 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1355 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1356 			  struct rt6_info *rt)
1357 {
1358 	struct ipv6_pinfo *np = inet6_sk(sk);
1359 	unsigned int mtu;
1360 	struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1361 
1362 	/* callers pass dst together with a reference, set it first so
1363 	 * ip6_cork_release() can put it down even in case of an error.
1364 	 */
1365 	cork->base.dst = &rt->dst;
1366 
1367 	/*
1368 	 * setup for corking
1369 	 */
1370 	if (opt) {
1371 		if (WARN_ON(v6_cork->opt))
1372 			return -EINVAL;
1373 
1374 		nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1375 		if (unlikely(!nopt))
1376 			return -ENOBUFS;
1377 
1378 		nopt->tot_len = sizeof(*opt);
1379 		nopt->opt_flen = opt->opt_flen;
1380 		nopt->opt_nflen = opt->opt_nflen;
1381 
1382 		nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1383 		if (opt->dst0opt && !nopt->dst0opt)
1384 			return -ENOBUFS;
1385 
1386 		nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1387 		if (opt->dst1opt && !nopt->dst1opt)
1388 			return -ENOBUFS;
1389 
1390 		nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1391 		if (opt->hopopt && !nopt->hopopt)
1392 			return -ENOBUFS;
1393 
1394 		nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1395 		if (opt->srcrt && !nopt->srcrt)
1396 			return -ENOBUFS;
1397 
1398 		/* need source address above miyazawa*/
1399 	}
1400 	v6_cork->hop_limit = ipc6->hlimit;
1401 	v6_cork->tclass = ipc6->tclass;
1402 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1403 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1404 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1405 	else
1406 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1407 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1408 	if (np->frag_size < mtu) {
1409 		if (np->frag_size)
1410 			mtu = np->frag_size;
1411 	}
1412 	cork->base.fragsize = mtu;
1413 	cork->base.gso_size = ipc6->gso_size;
1414 	cork->base.tx_flags = 0;
1415 	cork->base.mark = ipc6->sockc.mark;
1416 	sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1417 
1418 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1419 		cork->base.flags |= IPCORK_ALLFRAG;
1420 	cork->base.length = 0;
1421 
1422 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1423 
1424 	return 0;
1425 }
1426 
1427 static int __ip6_append_data(struct sock *sk,
1428 			     struct sk_buff_head *queue,
1429 			     struct inet_cork_full *cork_full,
1430 			     struct inet6_cork *v6_cork,
1431 			     struct page_frag *pfrag,
1432 			     int getfrag(void *from, char *to, int offset,
1433 					 int len, int odd, struct sk_buff *skb),
1434 			     void *from, int length, int transhdrlen,
1435 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1436 {
1437 	struct sk_buff *skb, *skb_prev = NULL;
1438 	struct inet_cork *cork = &cork_full->base;
1439 	struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1440 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1441 	struct ubuf_info *uarg = NULL;
1442 	int exthdrlen = 0;
1443 	int dst_exthdrlen = 0;
1444 	int hh_len;
1445 	int copy;
1446 	int err;
1447 	int offset = 0;
1448 	u32 tskey = 0;
1449 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1450 	struct ipv6_txoptions *opt = v6_cork->opt;
1451 	int csummode = CHECKSUM_NONE;
1452 	unsigned int maxnonfragsize, headersize;
1453 	unsigned int wmem_alloc_delta = 0;
1454 	bool paged, extra_uref = false;
1455 
1456 	skb = skb_peek_tail(queue);
1457 	if (!skb) {
1458 		exthdrlen = opt ? opt->opt_flen : 0;
1459 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1460 	}
1461 
1462 	paged = !!cork->gso_size;
1463 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1464 	orig_mtu = mtu;
1465 
1466 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1467 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1468 		tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1469 
1470 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1471 
1472 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1473 			(opt ? opt->opt_nflen : 0);
1474 
1475 	headersize = sizeof(struct ipv6hdr) +
1476 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1477 		     (dst_allfrag(&rt->dst) ?
1478 		      sizeof(struct frag_hdr) : 0) +
1479 		     rt->rt6i_nfheader_len;
1480 
1481 	if (mtu <= fragheaderlen ||
1482 	    ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1483 		goto emsgsize;
1484 
1485 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1486 		     sizeof(struct frag_hdr);
1487 
1488 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1489 	 * the first fragment
1490 	 */
1491 	if (headersize + transhdrlen > mtu)
1492 		goto emsgsize;
1493 
1494 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1495 	    (sk->sk_protocol == IPPROTO_UDP ||
1496 	     sk->sk_protocol == IPPROTO_ICMPV6 ||
1497 	     sk->sk_protocol == IPPROTO_RAW)) {
1498 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1499 				sizeof(struct ipv6hdr));
1500 		goto emsgsize;
1501 	}
1502 
1503 	if (ip6_sk_ignore_df(sk))
1504 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1505 	else
1506 		maxnonfragsize = mtu;
1507 
1508 	if (cork->length + length > maxnonfragsize - headersize) {
1509 emsgsize:
1510 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1511 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1512 		return -EMSGSIZE;
1513 	}
1514 
1515 	/* CHECKSUM_PARTIAL only with no extension headers and when
1516 	 * we are not going to fragment
1517 	 */
1518 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1519 	    headersize == sizeof(struct ipv6hdr) &&
1520 	    length <= mtu - headersize &&
1521 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1522 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1523 		csummode = CHECKSUM_PARTIAL;
1524 
1525 	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1526 		uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1527 		if (!uarg)
1528 			return -ENOBUFS;
1529 		extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1530 		if (rt->dst.dev->features & NETIF_F_SG &&
1531 		    csummode == CHECKSUM_PARTIAL) {
1532 			paged = true;
1533 		} else {
1534 			uarg->zerocopy = 0;
1535 			skb_zcopy_set(skb, uarg, &extra_uref);
1536 		}
1537 	}
1538 
1539 	/*
1540 	 * Let's try using as much space as possible.
1541 	 * Use MTU if total length of the message fits into the MTU.
1542 	 * Otherwise, we need to reserve fragment header and
1543 	 * fragment alignment (= 8-15 octects, in total).
1544 	 *
1545 	 * Note that we may need to "move" the data from the tail
1546 	 * of the buffer to the new fragment when we split
1547 	 * the message.
1548 	 *
1549 	 * FIXME: It may be fragmented into multiple chunks
1550 	 *        at once if non-fragmentable extension headers
1551 	 *        are too large.
1552 	 * --yoshfuji
1553 	 */
1554 
1555 	cork->length += length;
1556 	if (!skb)
1557 		goto alloc_new_skb;
1558 
1559 	while (length > 0) {
1560 		/* Check if the remaining data fits into current packet. */
1561 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1562 		if (copy < length)
1563 			copy = maxfraglen - skb->len;
1564 
1565 		if (copy <= 0) {
1566 			char *data;
1567 			unsigned int datalen;
1568 			unsigned int fraglen;
1569 			unsigned int fraggap;
1570 			unsigned int alloclen, alloc_extra;
1571 			unsigned int pagedlen;
1572 alloc_new_skb:
1573 			/* There's no room in the current skb */
1574 			if (skb)
1575 				fraggap = skb->len - maxfraglen;
1576 			else
1577 				fraggap = 0;
1578 			/* update mtu and maxfraglen if necessary */
1579 			if (!skb || !skb_prev)
1580 				ip6_append_data_mtu(&mtu, &maxfraglen,
1581 						    fragheaderlen, skb, rt,
1582 						    orig_mtu);
1583 
1584 			skb_prev = skb;
1585 
1586 			/*
1587 			 * If remaining data exceeds the mtu,
1588 			 * we know we need more fragment(s).
1589 			 */
1590 			datalen = length + fraggap;
1591 
1592 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1593 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1594 			fraglen = datalen + fragheaderlen;
1595 			pagedlen = 0;
1596 
1597 			alloc_extra = hh_len;
1598 			alloc_extra += dst_exthdrlen;
1599 			alloc_extra += rt->dst.trailer_len;
1600 
1601 			/* We just reserve space for fragment header.
1602 			 * Note: this may be overallocation if the message
1603 			 * (without MSG_MORE) fits into the MTU.
1604 			 */
1605 			alloc_extra += sizeof(struct frag_hdr);
1606 
1607 			if ((flags & MSG_MORE) &&
1608 			    !(rt->dst.dev->features&NETIF_F_SG))
1609 				alloclen = mtu;
1610 			else if (!paged &&
1611 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1612 				  !(rt->dst.dev->features & NETIF_F_SG)))
1613 				alloclen = fraglen;
1614 			else {
1615 				alloclen = min_t(int, fraglen, MAX_HEADER);
1616 				pagedlen = fraglen - alloclen;
1617 			}
1618 			alloclen += alloc_extra;
1619 
1620 			if (datalen != length + fraggap) {
1621 				/*
1622 				 * this is not the last fragment, the trailer
1623 				 * space is regarded as data space.
1624 				 */
1625 				datalen += rt->dst.trailer_len;
1626 			}
1627 
1628 			fraglen = datalen + fragheaderlen;
1629 
1630 			copy = datalen - transhdrlen - fraggap - pagedlen;
1631 			if (copy < 0) {
1632 				err = -EINVAL;
1633 				goto error;
1634 			}
1635 			if (transhdrlen) {
1636 				skb = sock_alloc_send_skb(sk, alloclen,
1637 						(flags & MSG_DONTWAIT), &err);
1638 			} else {
1639 				skb = NULL;
1640 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1641 				    2 * sk->sk_sndbuf)
1642 					skb = alloc_skb(alloclen,
1643 							sk->sk_allocation);
1644 				if (unlikely(!skb))
1645 					err = -ENOBUFS;
1646 			}
1647 			if (!skb)
1648 				goto error;
1649 			/*
1650 			 *	Fill in the control structures
1651 			 */
1652 			skb->protocol = htons(ETH_P_IPV6);
1653 			skb->ip_summed = csummode;
1654 			skb->csum = 0;
1655 			/* reserve for fragmentation and ipsec header */
1656 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1657 				    dst_exthdrlen);
1658 
1659 			/*
1660 			 *	Find where to start putting bytes
1661 			 */
1662 			data = skb_put(skb, fraglen - pagedlen);
1663 			skb_set_network_header(skb, exthdrlen);
1664 			data += fragheaderlen;
1665 			skb->transport_header = (skb->network_header +
1666 						 fragheaderlen);
1667 			if (fraggap) {
1668 				skb->csum = skb_copy_and_csum_bits(
1669 					skb_prev, maxfraglen,
1670 					data + transhdrlen, fraggap);
1671 				skb_prev->csum = csum_sub(skb_prev->csum,
1672 							  skb->csum);
1673 				data += fraggap;
1674 				pskb_trim_unique(skb_prev, maxfraglen);
1675 			}
1676 			if (copy > 0 &&
1677 			    getfrag(from, data + transhdrlen, offset,
1678 				    copy, fraggap, skb) < 0) {
1679 				err = -EFAULT;
1680 				kfree_skb(skb);
1681 				goto error;
1682 			}
1683 
1684 			offset += copy;
1685 			length -= copy + transhdrlen;
1686 			transhdrlen = 0;
1687 			exthdrlen = 0;
1688 			dst_exthdrlen = 0;
1689 
1690 			/* Only the initial fragment is time stamped */
1691 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1692 			cork->tx_flags = 0;
1693 			skb_shinfo(skb)->tskey = tskey;
1694 			tskey = 0;
1695 			skb_zcopy_set(skb, uarg, &extra_uref);
1696 
1697 			if ((flags & MSG_CONFIRM) && !skb_prev)
1698 				skb_set_dst_pending_confirm(skb, 1);
1699 
1700 			/*
1701 			 * Put the packet on the pending queue
1702 			 */
1703 			if (!skb->destructor) {
1704 				skb->destructor = sock_wfree;
1705 				skb->sk = sk;
1706 				wmem_alloc_delta += skb->truesize;
1707 			}
1708 			__skb_queue_tail(queue, skb);
1709 			continue;
1710 		}
1711 
1712 		if (copy > length)
1713 			copy = length;
1714 
1715 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1716 		    skb_tailroom(skb) >= copy) {
1717 			unsigned int off;
1718 
1719 			off = skb->len;
1720 			if (getfrag(from, skb_put(skb, copy),
1721 						offset, copy, off, skb) < 0) {
1722 				__skb_trim(skb, off);
1723 				err = -EFAULT;
1724 				goto error;
1725 			}
1726 		} else if (!uarg || !uarg->zerocopy) {
1727 			int i = skb_shinfo(skb)->nr_frags;
1728 
1729 			err = -ENOMEM;
1730 			if (!sk_page_frag_refill(sk, pfrag))
1731 				goto error;
1732 
1733 			if (!skb_can_coalesce(skb, i, pfrag->page,
1734 					      pfrag->offset)) {
1735 				err = -EMSGSIZE;
1736 				if (i == MAX_SKB_FRAGS)
1737 					goto error;
1738 
1739 				__skb_fill_page_desc(skb, i, pfrag->page,
1740 						     pfrag->offset, 0);
1741 				skb_shinfo(skb)->nr_frags = ++i;
1742 				get_page(pfrag->page);
1743 			}
1744 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1745 			if (getfrag(from,
1746 				    page_address(pfrag->page) + pfrag->offset,
1747 				    offset, copy, skb->len, skb) < 0)
1748 				goto error_efault;
1749 
1750 			pfrag->offset += copy;
1751 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1752 			skb->len += copy;
1753 			skb->data_len += copy;
1754 			skb->truesize += copy;
1755 			wmem_alloc_delta += copy;
1756 		} else {
1757 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1758 			if (err < 0)
1759 				goto error;
1760 		}
1761 		offset += copy;
1762 		length -= copy;
1763 	}
1764 
1765 	if (wmem_alloc_delta)
1766 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1767 	return 0;
1768 
1769 error_efault:
1770 	err = -EFAULT;
1771 error:
1772 	net_zcopy_put_abort(uarg, extra_uref);
1773 	cork->length -= length;
1774 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1775 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1776 	return err;
1777 }
1778 
1779 int ip6_append_data(struct sock *sk,
1780 		    int getfrag(void *from, char *to, int offset, int len,
1781 				int odd, struct sk_buff *skb),
1782 		    void *from, int length, int transhdrlen,
1783 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1784 		    struct rt6_info *rt, unsigned int flags)
1785 {
1786 	struct inet_sock *inet = inet_sk(sk);
1787 	struct ipv6_pinfo *np = inet6_sk(sk);
1788 	int exthdrlen;
1789 	int err;
1790 
1791 	if (flags&MSG_PROBE)
1792 		return 0;
1793 	if (skb_queue_empty(&sk->sk_write_queue)) {
1794 		/*
1795 		 * setup for corking
1796 		 */
1797 		dst_hold(&rt->dst);
1798 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1799 				     ipc6, rt);
1800 		if (err)
1801 			return err;
1802 
1803 		inet->cork.fl.u.ip6 = *fl6;
1804 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1805 		length += exthdrlen;
1806 		transhdrlen += exthdrlen;
1807 	} else {
1808 		transhdrlen = 0;
1809 	}
1810 
1811 	return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1812 				 &np->cork, sk_page_frag(sk), getfrag,
1813 				 from, length, transhdrlen, flags, ipc6);
1814 }
1815 EXPORT_SYMBOL_GPL(ip6_append_data);
1816 
1817 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1818 {
1819 	struct dst_entry *dst = cork->base.dst;
1820 
1821 	cork->base.dst = NULL;
1822 	cork->base.flags &= ~IPCORK_ALLFRAG;
1823 	skb_dst_set(skb, dst);
1824 }
1825 
1826 static void ip6_cork_release(struct inet_cork_full *cork,
1827 			     struct inet6_cork *v6_cork)
1828 {
1829 	if (v6_cork->opt) {
1830 		struct ipv6_txoptions *opt = v6_cork->opt;
1831 
1832 		kfree(opt->dst0opt);
1833 		kfree(opt->dst1opt);
1834 		kfree(opt->hopopt);
1835 		kfree(opt->srcrt);
1836 		kfree(opt);
1837 		v6_cork->opt = NULL;
1838 	}
1839 
1840 	if (cork->base.dst) {
1841 		dst_release(cork->base.dst);
1842 		cork->base.dst = NULL;
1843 		cork->base.flags &= ~IPCORK_ALLFRAG;
1844 	}
1845 }
1846 
1847 struct sk_buff *__ip6_make_skb(struct sock *sk,
1848 			       struct sk_buff_head *queue,
1849 			       struct inet_cork_full *cork,
1850 			       struct inet6_cork *v6_cork)
1851 {
1852 	struct sk_buff *skb, *tmp_skb;
1853 	struct sk_buff **tail_skb;
1854 	struct in6_addr *final_dst;
1855 	struct ipv6_pinfo *np = inet6_sk(sk);
1856 	struct net *net = sock_net(sk);
1857 	struct ipv6hdr *hdr;
1858 	struct ipv6_txoptions *opt = v6_cork->opt;
1859 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1860 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1861 	unsigned char proto = fl6->flowi6_proto;
1862 
1863 	skb = __skb_dequeue(queue);
1864 	if (!skb)
1865 		goto out;
1866 	tail_skb = &(skb_shinfo(skb)->frag_list);
1867 
1868 	/* move skb->data to ip header from ext header */
1869 	if (skb->data < skb_network_header(skb))
1870 		__skb_pull(skb, skb_network_offset(skb));
1871 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1872 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1873 		*tail_skb = tmp_skb;
1874 		tail_skb = &(tmp_skb->next);
1875 		skb->len += tmp_skb->len;
1876 		skb->data_len += tmp_skb->len;
1877 		skb->truesize += tmp_skb->truesize;
1878 		tmp_skb->destructor = NULL;
1879 		tmp_skb->sk = NULL;
1880 	}
1881 
1882 	/* Allow local fragmentation. */
1883 	skb->ignore_df = ip6_sk_ignore_df(sk);
1884 	__skb_pull(skb, skb_network_header_len(skb));
1885 
1886 	final_dst = &fl6->daddr;
1887 	if (opt && opt->opt_flen)
1888 		ipv6_push_frag_opts(skb, opt, &proto);
1889 	if (opt && opt->opt_nflen)
1890 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1891 
1892 	skb_push(skb, sizeof(struct ipv6hdr));
1893 	skb_reset_network_header(skb);
1894 	hdr = ipv6_hdr(skb);
1895 
1896 	ip6_flow_hdr(hdr, v6_cork->tclass,
1897 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1898 					ip6_autoflowlabel(net, np), fl6));
1899 	hdr->hop_limit = v6_cork->hop_limit;
1900 	hdr->nexthdr = proto;
1901 	hdr->saddr = fl6->saddr;
1902 	hdr->daddr = *final_dst;
1903 
1904 	skb->priority = sk->sk_priority;
1905 	skb->mark = cork->base.mark;
1906 	skb->tstamp = cork->base.transmit_time;
1907 
1908 	ip6_cork_steal_dst(skb, cork);
1909 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1910 	if (proto == IPPROTO_ICMPV6) {
1911 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1912 
1913 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1914 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1915 	}
1916 
1917 	ip6_cork_release(cork, v6_cork);
1918 out:
1919 	return skb;
1920 }
1921 
1922 int ip6_send_skb(struct sk_buff *skb)
1923 {
1924 	struct net *net = sock_net(skb->sk);
1925 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1926 	int err;
1927 
1928 	err = ip6_local_out(net, skb->sk, skb);
1929 	if (err) {
1930 		if (err > 0)
1931 			err = net_xmit_errno(err);
1932 		if (err)
1933 			IP6_INC_STATS(net, rt->rt6i_idev,
1934 				      IPSTATS_MIB_OUTDISCARDS);
1935 	}
1936 
1937 	return err;
1938 }
1939 
1940 int ip6_push_pending_frames(struct sock *sk)
1941 {
1942 	struct sk_buff *skb;
1943 
1944 	skb = ip6_finish_skb(sk);
1945 	if (!skb)
1946 		return 0;
1947 
1948 	return ip6_send_skb(skb);
1949 }
1950 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1951 
1952 static void __ip6_flush_pending_frames(struct sock *sk,
1953 				       struct sk_buff_head *queue,
1954 				       struct inet_cork_full *cork,
1955 				       struct inet6_cork *v6_cork)
1956 {
1957 	struct sk_buff *skb;
1958 
1959 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1960 		if (skb_dst(skb))
1961 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1962 				      IPSTATS_MIB_OUTDISCARDS);
1963 		kfree_skb(skb);
1964 	}
1965 
1966 	ip6_cork_release(cork, v6_cork);
1967 }
1968 
1969 void ip6_flush_pending_frames(struct sock *sk)
1970 {
1971 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1972 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1973 }
1974 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1975 
1976 struct sk_buff *ip6_make_skb(struct sock *sk,
1977 			     int getfrag(void *from, char *to, int offset,
1978 					 int len, int odd, struct sk_buff *skb),
1979 			     void *from, int length, int transhdrlen,
1980 			     struct ipcm6_cookie *ipc6, struct rt6_info *rt,
1981 			     unsigned int flags, struct inet_cork_full *cork)
1982 {
1983 	struct inet6_cork v6_cork;
1984 	struct sk_buff_head queue;
1985 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1986 	int err;
1987 
1988 	if (flags & MSG_PROBE) {
1989 		dst_release(&rt->dst);
1990 		return NULL;
1991 	}
1992 
1993 	__skb_queue_head_init(&queue);
1994 
1995 	cork->base.flags = 0;
1996 	cork->base.addr = 0;
1997 	cork->base.opt = NULL;
1998 	v6_cork.opt = NULL;
1999 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2000 	if (err) {
2001 		ip6_cork_release(cork, &v6_cork);
2002 		return ERR_PTR(err);
2003 	}
2004 	if (ipc6->dontfrag < 0)
2005 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2006 
2007 	err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2008 				&current->task_frag, getfrag, from,
2009 				length + exthdrlen, transhdrlen + exthdrlen,
2010 				flags, ipc6);
2011 	if (err) {
2012 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2013 		return ERR_PTR(err);
2014 	}
2015 
2016 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2017 }
2018