xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 2326f3cd)
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(skb);
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 		IPCB(skb)->flags |= IPSKB_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(skb);
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(skb);
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->tstamp = 0;
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 inet6_dev *idev = __in6_dev_get_safely(skb->dev);
468 	struct dst_entry *dst = skb_dst(skb);
469 	struct ipv6hdr *hdr = ipv6_hdr(skb);
470 	struct inet6_skb_parm *opt = IP6CB(skb);
471 	struct net *net = dev_net(dst->dev);
472 	u32 mtu;
473 
474 	if (net->ipv6.devconf_all->forwarding == 0)
475 		goto error;
476 
477 	if (skb->pkt_type != PACKET_HOST)
478 		goto drop;
479 
480 	if (unlikely(skb->sk))
481 		goto drop;
482 
483 	if (skb_warn_if_lro(skb))
484 		goto drop;
485 
486 	if (!net->ipv6.devconf_all->disable_policy &&
487 	    !idev->cnf.disable_policy &&
488 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
489 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
490 		goto drop;
491 	}
492 
493 	skb_forward_csum(skb);
494 
495 	/*
496 	 *	We DO NOT make any processing on
497 	 *	RA packets, pushing them to user level AS IS
498 	 *	without ane WARRANTY that application will be able
499 	 *	to interpret them. The reason is that we
500 	 *	cannot make anything clever here.
501 	 *
502 	 *	We are not end-node, so that if packet contains
503 	 *	AH/ESP, we cannot make anything.
504 	 *	Defragmentation also would be mistake, RA packets
505 	 *	cannot be fragmented, because there is no warranty
506 	 *	that different fragments will go along one path. --ANK
507 	 */
508 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
509 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
510 			return 0;
511 	}
512 
513 	/*
514 	 *	check and decrement ttl
515 	 */
516 	if (hdr->hop_limit <= 1) {
517 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
518 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
519 
520 		kfree_skb(skb);
521 		return -ETIMEDOUT;
522 	}
523 
524 	/* XXX: idev->cnf.proxy_ndp? */
525 	if (net->ipv6.devconf_all->proxy_ndp &&
526 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
527 		int proxied = ip6_forward_proxy_check(skb);
528 		if (proxied > 0) {
529 			hdr->hop_limit--;
530 			return ip6_input(skb);
531 		} else if (proxied < 0) {
532 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
533 			goto drop;
534 		}
535 	}
536 
537 	if (!xfrm6_route_forward(skb)) {
538 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
539 		goto drop;
540 	}
541 	dst = skb_dst(skb);
542 
543 	/* IPv6 specs say nothing about it, but it is clear that we cannot
544 	   send redirects to source routed frames.
545 	   We don't send redirects to frames decapsulated from IPsec.
546 	 */
547 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
548 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
549 		struct in6_addr *target = NULL;
550 		struct inet_peer *peer;
551 		struct rt6_info *rt;
552 
553 		/*
554 		 *	incoming and outgoing devices are the same
555 		 *	send a redirect.
556 		 */
557 
558 		rt = (struct rt6_info *) dst;
559 		if (rt->rt6i_flags & RTF_GATEWAY)
560 			target = &rt->rt6i_gateway;
561 		else
562 			target = &hdr->daddr;
563 
564 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
565 
566 		/* Limit redirects both by destination (here)
567 		   and by source (inside ndisc_send_redirect)
568 		 */
569 		if (inet_peer_xrlim_allow(peer, 1*HZ))
570 			ndisc_send_redirect(skb, target);
571 		if (peer)
572 			inet_putpeer(peer);
573 	} else {
574 		int addrtype = ipv6_addr_type(&hdr->saddr);
575 
576 		/* This check is security critical. */
577 		if (addrtype == IPV6_ADDR_ANY ||
578 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
579 			goto error;
580 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
581 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
582 				    ICMPV6_NOT_NEIGHBOUR, 0);
583 			goto error;
584 		}
585 	}
586 
587 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
588 	if (mtu < IPV6_MIN_MTU)
589 		mtu = IPV6_MIN_MTU;
590 
591 	if (ip6_pkt_too_big(skb, mtu)) {
592 		/* Again, force OUTPUT device used as source address */
593 		skb->dev = dst->dev;
594 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
595 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
596 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
597 				IPSTATS_MIB_FRAGFAILS);
598 		kfree_skb(skb);
599 		return -EMSGSIZE;
600 	}
601 
602 	if (skb_cow(skb, dst->dev->hard_header_len)) {
603 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
604 				IPSTATS_MIB_OUTDISCARDS);
605 		goto drop;
606 	}
607 
608 	hdr = ipv6_hdr(skb);
609 
610 	/* Mangling hops number delayed to point after skb COW */
611 
612 	hdr->hop_limit--;
613 
614 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
615 		       net, NULL, skb, skb->dev, dst->dev,
616 		       ip6_forward_finish);
617 
618 error:
619 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
620 drop:
621 	kfree_skb(skb);
622 	return -EINVAL;
623 }
624 
625 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
626 {
627 	to->pkt_type = from->pkt_type;
628 	to->priority = from->priority;
629 	to->protocol = from->protocol;
630 	skb_dst_drop(to);
631 	skb_dst_set(to, dst_clone(skb_dst(from)));
632 	to->dev = from->dev;
633 	to->mark = from->mark;
634 
635 	skb_copy_hash(to, from);
636 
637 #ifdef CONFIG_NET_SCHED
638 	to->tc_index = from->tc_index;
639 #endif
640 	nf_copy(to, from);
641 	skb_ext_copy(to, from);
642 	skb_copy_secmark(to, from);
643 }
644 
645 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
646 		      u8 nexthdr, __be32 frag_id,
647 		      struct ip6_fraglist_iter *iter)
648 {
649 	unsigned int first_len;
650 	struct frag_hdr *fh;
651 
652 	/* BUILD HEADER */
653 	*prevhdr = NEXTHDR_FRAGMENT;
654 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
655 	if (!iter->tmp_hdr)
656 		return -ENOMEM;
657 
658 	iter->frag = skb_shinfo(skb)->frag_list;
659 	skb_frag_list_init(skb);
660 
661 	iter->offset = 0;
662 	iter->hlen = hlen;
663 	iter->frag_id = frag_id;
664 	iter->nexthdr = nexthdr;
665 
666 	__skb_pull(skb, hlen);
667 	fh = __skb_push(skb, sizeof(struct frag_hdr));
668 	__skb_push(skb, hlen);
669 	skb_reset_network_header(skb);
670 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
671 
672 	fh->nexthdr = nexthdr;
673 	fh->reserved = 0;
674 	fh->frag_off = htons(IP6_MF);
675 	fh->identification = frag_id;
676 
677 	first_len = skb_pagelen(skb);
678 	skb->data_len = first_len - skb_headlen(skb);
679 	skb->len = first_len;
680 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
681 
682 	return 0;
683 }
684 EXPORT_SYMBOL(ip6_fraglist_init);
685 
686 void ip6_fraglist_prepare(struct sk_buff *skb,
687 			  struct ip6_fraglist_iter *iter)
688 {
689 	struct sk_buff *frag = iter->frag;
690 	unsigned int hlen = iter->hlen;
691 	struct frag_hdr *fh;
692 
693 	frag->ip_summed = CHECKSUM_NONE;
694 	skb_reset_transport_header(frag);
695 	fh = __skb_push(frag, sizeof(struct frag_hdr));
696 	__skb_push(frag, hlen);
697 	skb_reset_network_header(frag);
698 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
699 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
700 	fh->nexthdr = iter->nexthdr;
701 	fh->reserved = 0;
702 	fh->frag_off = htons(iter->offset);
703 	if (frag->next)
704 		fh->frag_off |= htons(IP6_MF);
705 	fh->identification = iter->frag_id;
706 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
707 	ip6_copy_metadata(frag, skb);
708 }
709 EXPORT_SYMBOL(ip6_fraglist_prepare);
710 
711 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
712 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
713 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
714 {
715 	state->prevhdr = prevhdr;
716 	state->nexthdr = nexthdr;
717 	state->frag_id = frag_id;
718 
719 	state->hlen = hlen;
720 	state->mtu = mtu;
721 
722 	state->left = skb->len - hlen;	/* Space per frame */
723 	state->ptr = hlen;		/* Where to start from */
724 
725 	state->hroom = hdr_room;
726 	state->troom = needed_tailroom;
727 
728 	state->offset = 0;
729 }
730 EXPORT_SYMBOL(ip6_frag_init);
731 
732 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
733 {
734 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
735 	struct sk_buff *frag;
736 	struct frag_hdr *fh;
737 	unsigned int len;
738 
739 	len = state->left;
740 	/* IF: it doesn't fit, use 'mtu' - the data space left */
741 	if (len > state->mtu)
742 		len = state->mtu;
743 	/* IF: we are not sending up to and including the packet end
744 	   then align the next start on an eight byte boundary */
745 	if (len < state->left)
746 		len &= ~7;
747 
748 	/* Allocate buffer */
749 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
750 			 state->hroom + state->troom, GFP_ATOMIC);
751 	if (!frag)
752 		return ERR_PTR(-ENOMEM);
753 
754 	/*
755 	 *	Set up data on packet
756 	 */
757 
758 	ip6_copy_metadata(frag, skb);
759 	skb_reserve(frag, state->hroom);
760 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
761 	skb_reset_network_header(frag);
762 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
763 	frag->transport_header = (frag->network_header + state->hlen +
764 				  sizeof(struct frag_hdr));
765 
766 	/*
767 	 *	Charge the memory for the fragment to any owner
768 	 *	it might possess
769 	 */
770 	if (skb->sk)
771 		skb_set_owner_w(frag, skb->sk);
772 
773 	/*
774 	 *	Copy the packet header into the new buffer.
775 	 */
776 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
777 
778 	fragnexthdr_offset = skb_network_header(frag);
779 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
780 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
781 
782 	/*
783 	 *	Build fragment header.
784 	 */
785 	fh->nexthdr = state->nexthdr;
786 	fh->reserved = 0;
787 	fh->identification = state->frag_id;
788 
789 	/*
790 	 *	Copy a block of the IP datagram.
791 	 */
792 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
793 			     len));
794 	state->left -= len;
795 
796 	fh->frag_off = htons(state->offset);
797 	if (state->left > 0)
798 		fh->frag_off |= htons(IP6_MF);
799 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
800 
801 	state->ptr += len;
802 	state->offset += len;
803 
804 	return frag;
805 }
806 EXPORT_SYMBOL(ip6_frag_next);
807 
808 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
809 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
810 {
811 	struct sk_buff *frag;
812 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
813 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
814 				inet6_sk(skb->sk) : NULL;
815 	struct ip6_frag_state state;
816 	unsigned int mtu, hlen, nexthdr_offset;
817 	ktime_t tstamp = skb->tstamp;
818 	int hroom, err = 0;
819 	__be32 frag_id;
820 	u8 *prevhdr, nexthdr = 0;
821 
822 	err = ip6_find_1stfragopt(skb, &prevhdr);
823 	if (err < 0)
824 		goto fail;
825 	hlen = err;
826 	nexthdr = *prevhdr;
827 	nexthdr_offset = prevhdr - skb_network_header(skb);
828 
829 	mtu = ip6_skb_dst_mtu(skb);
830 
831 	/* We must not fragment if the socket is set to force MTU discovery
832 	 * or if the skb it not generated by a local socket.
833 	 */
834 	if (unlikely(!skb->ignore_df && skb->len > mtu))
835 		goto fail_toobig;
836 
837 	if (IP6CB(skb)->frag_max_size) {
838 		if (IP6CB(skb)->frag_max_size > mtu)
839 			goto fail_toobig;
840 
841 		/* don't send fragments larger than what we received */
842 		mtu = IP6CB(skb)->frag_max_size;
843 		if (mtu < IPV6_MIN_MTU)
844 			mtu = IPV6_MIN_MTU;
845 	}
846 
847 	if (np && np->frag_size < mtu) {
848 		if (np->frag_size)
849 			mtu = np->frag_size;
850 	}
851 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
852 		goto fail_toobig;
853 	mtu -= hlen + sizeof(struct frag_hdr);
854 
855 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
856 				    &ipv6_hdr(skb)->saddr);
857 
858 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
859 	    (err = skb_checksum_help(skb)))
860 		goto fail;
861 
862 	prevhdr = skb_network_header(skb) + nexthdr_offset;
863 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
864 	if (skb_has_frag_list(skb)) {
865 		unsigned int first_len = skb_pagelen(skb);
866 		struct ip6_fraglist_iter iter;
867 		struct sk_buff *frag2;
868 
869 		if (first_len - hlen > mtu ||
870 		    ((first_len - hlen) & 7) ||
871 		    skb_cloned(skb) ||
872 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
873 			goto slow_path;
874 
875 		skb_walk_frags(skb, frag) {
876 			/* Correct geometry. */
877 			if (frag->len > mtu ||
878 			    ((frag->len & 7) && frag->next) ||
879 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
880 				goto slow_path_clean;
881 
882 			/* Partially cloned skb? */
883 			if (skb_shared(frag))
884 				goto slow_path_clean;
885 
886 			BUG_ON(frag->sk);
887 			if (skb->sk) {
888 				frag->sk = skb->sk;
889 				frag->destructor = sock_wfree;
890 			}
891 			skb->truesize -= frag->truesize;
892 		}
893 
894 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
895 					&iter);
896 		if (err < 0)
897 			goto fail;
898 
899 		for (;;) {
900 			/* Prepare header of the next frame,
901 			 * before previous one went down. */
902 			if (iter.frag)
903 				ip6_fraglist_prepare(skb, &iter);
904 
905 			skb->tstamp = tstamp;
906 			err = output(net, sk, skb);
907 			if (!err)
908 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
909 					      IPSTATS_MIB_FRAGCREATES);
910 
911 			if (err || !iter.frag)
912 				break;
913 
914 			skb = ip6_fraglist_next(&iter);
915 		}
916 
917 		kfree(iter.tmp_hdr);
918 
919 		if (err == 0) {
920 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
921 				      IPSTATS_MIB_FRAGOKS);
922 			return 0;
923 		}
924 
925 		kfree_skb_list(iter.frag);
926 
927 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
928 			      IPSTATS_MIB_FRAGFAILS);
929 		return err;
930 
931 slow_path_clean:
932 		skb_walk_frags(skb, frag2) {
933 			if (frag2 == frag)
934 				break;
935 			frag2->sk = NULL;
936 			frag2->destructor = NULL;
937 			skb->truesize += frag2->truesize;
938 		}
939 	}
940 
941 slow_path:
942 	/*
943 	 *	Fragment the datagram.
944 	 */
945 
946 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
947 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
948 		      &state);
949 
950 	/*
951 	 *	Keep copying data until we run out.
952 	 */
953 
954 	while (state.left > 0) {
955 		frag = ip6_frag_next(skb, &state);
956 		if (IS_ERR(frag)) {
957 			err = PTR_ERR(frag);
958 			goto fail;
959 		}
960 
961 		/*
962 		 *	Put this fragment into the sending queue.
963 		 */
964 		frag->tstamp = tstamp;
965 		err = output(net, sk, frag);
966 		if (err)
967 			goto fail;
968 
969 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
970 			      IPSTATS_MIB_FRAGCREATES);
971 	}
972 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
973 		      IPSTATS_MIB_FRAGOKS);
974 	consume_skb(skb);
975 	return err;
976 
977 fail_toobig:
978 	if (skb->sk && dst_allfrag(skb_dst(skb)))
979 		sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
980 
981 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
982 	err = -EMSGSIZE;
983 
984 fail:
985 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
986 		      IPSTATS_MIB_FRAGFAILS);
987 	kfree_skb(skb);
988 	return err;
989 }
990 
991 static inline int ip6_rt_check(const struct rt6key *rt_key,
992 			       const struct in6_addr *fl_addr,
993 			       const struct in6_addr *addr_cache)
994 {
995 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
996 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
997 }
998 
999 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1000 					  struct dst_entry *dst,
1001 					  const struct flowi6 *fl6)
1002 {
1003 	struct ipv6_pinfo *np = inet6_sk(sk);
1004 	struct rt6_info *rt;
1005 
1006 	if (!dst)
1007 		goto out;
1008 
1009 	if (dst->ops->family != AF_INET6) {
1010 		dst_release(dst);
1011 		return NULL;
1012 	}
1013 
1014 	rt = (struct rt6_info *)dst;
1015 	/* Yes, checking route validity in not connected
1016 	 * case is not very simple. Take into account,
1017 	 * that we do not support routing by source, TOS,
1018 	 * and MSG_DONTROUTE		--ANK (980726)
1019 	 *
1020 	 * 1. ip6_rt_check(): If route was host route,
1021 	 *    check that cached destination is current.
1022 	 *    If it is network route, we still may
1023 	 *    check its validity using saved pointer
1024 	 *    to the last used address: daddr_cache.
1025 	 *    We do not want to save whole address now,
1026 	 *    (because main consumer of this service
1027 	 *    is tcp, which has not this problem),
1028 	 *    so that the last trick works only on connected
1029 	 *    sockets.
1030 	 * 2. oif also should be the same.
1031 	 */
1032 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1033 #ifdef CONFIG_IPV6_SUBTREES
1034 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1035 #endif
1036 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
1037 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
1038 		dst_release(dst);
1039 		dst = NULL;
1040 	}
1041 
1042 out:
1043 	return dst;
1044 }
1045 
1046 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1047 			       struct dst_entry **dst, struct flowi6 *fl6)
1048 {
1049 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1050 	struct neighbour *n;
1051 	struct rt6_info *rt;
1052 #endif
1053 	int err;
1054 	int flags = 0;
1055 
1056 	/* The correct way to handle this would be to do
1057 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1058 	 * the route-specific preferred source forces the
1059 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1060 	 *
1061 	 * In source specific routing (no src=any default route),
1062 	 * ip6_route_output will fail given src=any saddr, though, so
1063 	 * that's why we try it again later.
1064 	 */
1065 	if (ipv6_addr_any(&fl6->saddr)) {
1066 		struct fib6_info *from;
1067 		struct rt6_info *rt;
1068 
1069 		*dst = ip6_route_output(net, sk, fl6);
1070 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1071 
1072 		rcu_read_lock();
1073 		from = rt ? rcu_dereference(rt->from) : NULL;
1074 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1075 					  sk ? inet6_sk(sk)->srcprefs : 0,
1076 					  &fl6->saddr);
1077 		rcu_read_unlock();
1078 
1079 		if (err)
1080 			goto out_err_release;
1081 
1082 		/* If we had an erroneous initial result, pretend it
1083 		 * never existed and let the SA-enabled version take
1084 		 * over.
1085 		 */
1086 		if ((*dst)->error) {
1087 			dst_release(*dst);
1088 			*dst = NULL;
1089 		}
1090 
1091 		if (fl6->flowi6_oif)
1092 			flags |= RT6_LOOKUP_F_IFACE;
1093 	}
1094 
1095 	if (!*dst)
1096 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1097 
1098 	err = (*dst)->error;
1099 	if (err)
1100 		goto out_err_release;
1101 
1102 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1103 	/*
1104 	 * Here if the dst entry we've looked up
1105 	 * has a neighbour entry that is in the INCOMPLETE
1106 	 * state and the src address from the flow is
1107 	 * marked as OPTIMISTIC, we release the found
1108 	 * dst entry and replace it instead with the
1109 	 * dst entry of the nexthop router
1110 	 */
1111 	rt = (struct rt6_info *) *dst;
1112 	rcu_read_lock_bh();
1113 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1114 				      rt6_nexthop(rt, &fl6->daddr));
1115 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1116 	rcu_read_unlock_bh();
1117 
1118 	if (err) {
1119 		struct inet6_ifaddr *ifp;
1120 		struct flowi6 fl_gw6;
1121 		int redirect;
1122 
1123 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1124 				      (*dst)->dev, 1);
1125 
1126 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1127 		if (ifp)
1128 			in6_ifa_put(ifp);
1129 
1130 		if (redirect) {
1131 			/*
1132 			 * We need to get the dst entry for the
1133 			 * default router instead
1134 			 */
1135 			dst_release(*dst);
1136 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1137 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1138 			*dst = ip6_route_output(net, sk, &fl_gw6);
1139 			err = (*dst)->error;
1140 			if (err)
1141 				goto out_err_release;
1142 		}
1143 	}
1144 #endif
1145 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1146 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1147 		err = -EAFNOSUPPORT;
1148 		goto out_err_release;
1149 	}
1150 
1151 	return 0;
1152 
1153 out_err_release:
1154 	dst_release(*dst);
1155 	*dst = NULL;
1156 
1157 	if (err == -ENETUNREACH)
1158 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1159 	return err;
1160 }
1161 
1162 /**
1163  *	ip6_dst_lookup - perform route lookup on flow
1164  *	@net: Network namespace to perform lookup in
1165  *	@sk: socket which provides route info
1166  *	@dst: pointer to dst_entry * for result
1167  *	@fl6: flow to lookup
1168  *
1169  *	This function performs a route lookup on the given flow.
1170  *
1171  *	It returns zero on success, or a standard errno code on error.
1172  */
1173 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1174 		   struct flowi6 *fl6)
1175 {
1176 	*dst = NULL;
1177 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1178 }
1179 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1180 
1181 /**
1182  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1183  *	@net: Network namespace to perform lookup in
1184  *	@sk: socket which provides route info
1185  *	@fl6: flow to lookup
1186  *	@final_dst: final destination address for ipsec lookup
1187  *
1188  *	This function performs a route lookup on the given flow.
1189  *
1190  *	It returns a valid dst pointer on success, or a pointer encoded
1191  *	error code.
1192  */
1193 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1194 				      const struct in6_addr *final_dst)
1195 {
1196 	struct dst_entry *dst = NULL;
1197 	int err;
1198 
1199 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1200 	if (err)
1201 		return ERR_PTR(err);
1202 	if (final_dst)
1203 		fl6->daddr = *final_dst;
1204 
1205 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1206 }
1207 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1208 
1209 /**
1210  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1211  *	@sk: socket which provides the dst cache and route info
1212  *	@fl6: flow to lookup
1213  *	@final_dst: final destination address for ipsec lookup
1214  *	@connected: whether @sk is connected or not
1215  *
1216  *	This function performs a route lookup on the given flow with the
1217  *	possibility of using the cached route in the socket if it is valid.
1218  *	It will take the socket dst lock when operating on the dst cache.
1219  *	As a result, this function can only be used in process context.
1220  *
1221  *	In addition, for a connected socket, cache the dst in the socket
1222  *	if the current cache is not valid.
1223  *
1224  *	It returns a valid dst pointer on success, or a pointer encoded
1225  *	error code.
1226  */
1227 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1228 					 const struct in6_addr *final_dst,
1229 					 bool connected)
1230 {
1231 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1232 
1233 	dst = ip6_sk_dst_check(sk, dst, fl6);
1234 	if (dst)
1235 		return dst;
1236 
1237 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1238 	if (connected && !IS_ERR(dst))
1239 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1240 
1241 	return dst;
1242 }
1243 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1244 
1245 /**
1246  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1247  *      @skb: Packet for which lookup is done
1248  *      @dev: Tunnel device
1249  *      @net: Network namespace of tunnel device
1250  *      @sock: Socket which provides route info
1251  *      @saddr: Memory to store the src ip address
1252  *      @info: Tunnel information
1253  *      @protocol: IP protocol
1254  *      @use_cache: Flag to enable cache usage
1255  *      This function performs a route lookup on a tunnel
1256  *
1257  *      It returns a valid dst pointer and stores src address to be used in
1258  *      tunnel in param saddr on success, else a pointer encoded error code.
1259  */
1260 
1261 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1262 					struct net_device *dev,
1263 					struct net *net,
1264 					struct socket *sock,
1265 					struct in6_addr *saddr,
1266 					const struct ip_tunnel_info *info,
1267 					u8 protocol,
1268 					bool use_cache)
1269 {
1270 	struct dst_entry *dst = NULL;
1271 #ifdef CONFIG_DST_CACHE
1272 	struct dst_cache *dst_cache;
1273 #endif
1274 	struct flowi6 fl6;
1275 	__u8 prio;
1276 
1277 #ifdef CONFIG_DST_CACHE
1278 	dst_cache = (struct dst_cache *)&info->dst_cache;
1279 	if (use_cache) {
1280 		dst = dst_cache_get_ip6(dst_cache, saddr);
1281 		if (dst)
1282 			return dst;
1283 	}
1284 #endif
1285 	memset(&fl6, 0, sizeof(fl6));
1286 	fl6.flowi6_mark = skb->mark;
1287 	fl6.flowi6_proto = protocol;
1288 	fl6.daddr = info->key.u.ipv6.dst;
1289 	fl6.saddr = info->key.u.ipv6.src;
1290 	prio = info->key.tos;
1291 	fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1292 					  info->key.label);
1293 
1294 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1295 					      NULL);
1296 	if (IS_ERR(dst)) {
1297 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1298 		return ERR_PTR(-ENETUNREACH);
1299 	}
1300 	if (dst->dev == dev) { /* is this necessary? */
1301 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1302 		dst_release(dst);
1303 		return ERR_PTR(-ELOOP);
1304 	}
1305 #ifdef CONFIG_DST_CACHE
1306 	if (use_cache)
1307 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1308 #endif
1309 	*saddr = fl6.saddr;
1310 	return dst;
1311 }
1312 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1313 
1314 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1315 					       gfp_t gfp)
1316 {
1317 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1318 }
1319 
1320 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1321 						gfp_t gfp)
1322 {
1323 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1324 }
1325 
1326 static void ip6_append_data_mtu(unsigned int *mtu,
1327 				int *maxfraglen,
1328 				unsigned int fragheaderlen,
1329 				struct sk_buff *skb,
1330 				struct rt6_info *rt,
1331 				unsigned int orig_mtu)
1332 {
1333 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1334 		if (!skb) {
1335 			/* first fragment, reserve header_len */
1336 			*mtu = orig_mtu - rt->dst.header_len;
1337 
1338 		} else {
1339 			/*
1340 			 * this fragment is not first, the headers
1341 			 * space is regarded as data space.
1342 			 */
1343 			*mtu = orig_mtu;
1344 		}
1345 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1346 			      + fragheaderlen - sizeof(struct frag_hdr);
1347 	}
1348 }
1349 
1350 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1351 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1352 			  struct rt6_info *rt, struct flowi6 *fl6)
1353 {
1354 	struct ipv6_pinfo *np = inet6_sk(sk);
1355 	unsigned int mtu;
1356 	struct ipv6_txoptions *opt = ipc6->opt;
1357 
1358 	/*
1359 	 * setup for corking
1360 	 */
1361 	if (opt) {
1362 		if (WARN_ON(v6_cork->opt))
1363 			return -EINVAL;
1364 
1365 		v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1366 		if (unlikely(!v6_cork->opt))
1367 			return -ENOBUFS;
1368 
1369 		v6_cork->opt->tot_len = sizeof(*opt);
1370 		v6_cork->opt->opt_flen = opt->opt_flen;
1371 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1372 
1373 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1374 						    sk->sk_allocation);
1375 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1376 			return -ENOBUFS;
1377 
1378 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1379 						    sk->sk_allocation);
1380 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1381 			return -ENOBUFS;
1382 
1383 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1384 						   sk->sk_allocation);
1385 		if (opt->hopopt && !v6_cork->opt->hopopt)
1386 			return -ENOBUFS;
1387 
1388 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1389 						    sk->sk_allocation);
1390 		if (opt->srcrt && !v6_cork->opt->srcrt)
1391 			return -ENOBUFS;
1392 
1393 		/* need source address above miyazawa*/
1394 	}
1395 	dst_hold(&rt->dst);
1396 	cork->base.dst = &rt->dst;
1397 	cork->fl.u.ip6 = *fl6;
1398 	v6_cork->hop_limit = ipc6->hlimit;
1399 	v6_cork->tclass = ipc6->tclass;
1400 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1401 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1402 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1403 	else
1404 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1405 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1406 	if (np->frag_size < mtu) {
1407 		if (np->frag_size)
1408 			mtu = np->frag_size;
1409 	}
1410 	if (mtu < IPV6_MIN_MTU)
1411 		return -EINVAL;
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 flowi6 *fl6,
1429 			     struct sk_buff_head *queue,
1430 			     struct inet_cork *cork,
1431 			     struct inet6_cork *v6_cork,
1432 			     struct page_frag *pfrag,
1433 			     int getfrag(void *from, char *to, int offset,
1434 					 int len, int odd, struct sk_buff *skb),
1435 			     void *from, int length, int transhdrlen,
1436 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1437 {
1438 	struct sk_buff *skb, *skb_prev = NULL;
1439 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1440 	struct ubuf_info *uarg = NULL;
1441 	int exthdrlen = 0;
1442 	int dst_exthdrlen = 0;
1443 	int hh_len;
1444 	int copy;
1445 	int err;
1446 	int offset = 0;
1447 	u32 tskey = 0;
1448 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1449 	struct ipv6_txoptions *opt = v6_cork->opt;
1450 	int csummode = CHECKSUM_NONE;
1451 	unsigned int maxnonfragsize, headersize;
1452 	unsigned int wmem_alloc_delta = 0;
1453 	bool paged, extra_uref = false;
1454 
1455 	skb = skb_peek_tail(queue);
1456 	if (!skb) {
1457 		exthdrlen = opt ? opt->opt_flen : 0;
1458 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1459 	}
1460 
1461 	paged = !!cork->gso_size;
1462 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1463 	orig_mtu = mtu;
1464 
1465 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1466 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1467 		tskey = sk->sk_tskey++;
1468 
1469 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1470 
1471 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1472 			(opt ? opt->opt_nflen : 0);
1473 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1474 		     sizeof(struct frag_hdr);
1475 
1476 	headersize = sizeof(struct ipv6hdr) +
1477 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1478 		     (dst_allfrag(&rt->dst) ?
1479 		      sizeof(struct frag_hdr) : 0) +
1480 		     rt->rt6i_nfheader_len;
1481 
1482 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1483 	 * the first fragment
1484 	 */
1485 	if (headersize + transhdrlen > mtu)
1486 		goto emsgsize;
1487 
1488 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1489 	    (sk->sk_protocol == IPPROTO_UDP ||
1490 	     sk->sk_protocol == IPPROTO_RAW)) {
1491 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1492 				sizeof(struct ipv6hdr));
1493 		goto emsgsize;
1494 	}
1495 
1496 	if (ip6_sk_ignore_df(sk))
1497 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1498 	else
1499 		maxnonfragsize = mtu;
1500 
1501 	if (cork->length + length > maxnonfragsize - headersize) {
1502 emsgsize:
1503 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1504 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1505 		return -EMSGSIZE;
1506 	}
1507 
1508 	/* CHECKSUM_PARTIAL only with no extension headers and when
1509 	 * we are not going to fragment
1510 	 */
1511 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1512 	    headersize == sizeof(struct ipv6hdr) &&
1513 	    length <= mtu - headersize &&
1514 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1515 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1516 		csummode = CHECKSUM_PARTIAL;
1517 
1518 	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1519 		uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1520 		if (!uarg)
1521 			return -ENOBUFS;
1522 		extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1523 		if (rt->dst.dev->features & NETIF_F_SG &&
1524 		    csummode == CHECKSUM_PARTIAL) {
1525 			paged = true;
1526 		} else {
1527 			uarg->zerocopy = 0;
1528 			skb_zcopy_set(skb, uarg, &extra_uref);
1529 		}
1530 	}
1531 
1532 	/*
1533 	 * Let's try using as much space as possible.
1534 	 * Use MTU if total length of the message fits into the MTU.
1535 	 * Otherwise, we need to reserve fragment header and
1536 	 * fragment alignment (= 8-15 octects, in total).
1537 	 *
1538 	 * Note that we may need to "move" the data from the tail
1539 	 * of the buffer to the new fragment when we split
1540 	 * the message.
1541 	 *
1542 	 * FIXME: It may be fragmented into multiple chunks
1543 	 *        at once if non-fragmentable extension headers
1544 	 *        are too large.
1545 	 * --yoshfuji
1546 	 */
1547 
1548 	cork->length += length;
1549 	if (!skb)
1550 		goto alloc_new_skb;
1551 
1552 	while (length > 0) {
1553 		/* Check if the remaining data fits into current packet. */
1554 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1555 		if (copy < length)
1556 			copy = maxfraglen - skb->len;
1557 
1558 		if (copy <= 0) {
1559 			char *data;
1560 			unsigned int datalen;
1561 			unsigned int fraglen;
1562 			unsigned int fraggap;
1563 			unsigned int alloclen, alloc_extra;
1564 			unsigned int pagedlen;
1565 alloc_new_skb:
1566 			/* There's no room in the current skb */
1567 			if (skb)
1568 				fraggap = skb->len - maxfraglen;
1569 			else
1570 				fraggap = 0;
1571 			/* update mtu and maxfraglen if necessary */
1572 			if (!skb || !skb_prev)
1573 				ip6_append_data_mtu(&mtu, &maxfraglen,
1574 						    fragheaderlen, skb, rt,
1575 						    orig_mtu);
1576 
1577 			skb_prev = skb;
1578 
1579 			/*
1580 			 * If remaining data exceeds the mtu,
1581 			 * we know we need more fragment(s).
1582 			 */
1583 			datalen = length + fraggap;
1584 
1585 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1586 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1587 			fraglen = datalen + fragheaderlen;
1588 			pagedlen = 0;
1589 
1590 			alloc_extra = hh_len;
1591 			alloc_extra += dst_exthdrlen;
1592 			alloc_extra += rt->dst.trailer_len;
1593 
1594 			/* We just reserve space for fragment header.
1595 			 * Note: this may be overallocation if the message
1596 			 * (without MSG_MORE) fits into the MTU.
1597 			 */
1598 			alloc_extra += sizeof(struct frag_hdr);
1599 
1600 			if ((flags & MSG_MORE) &&
1601 			    !(rt->dst.dev->features&NETIF_F_SG))
1602 				alloclen = mtu;
1603 			else if (!paged &&
1604 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1605 				  !(rt->dst.dev->features & NETIF_F_SG)))
1606 				alloclen = fraglen;
1607 			else {
1608 				alloclen = min_t(int, fraglen, MAX_HEADER);
1609 				pagedlen = fraglen - alloclen;
1610 			}
1611 			alloclen += alloc_extra;
1612 
1613 			if (datalen != length + fraggap) {
1614 				/*
1615 				 * this is not the last fragment, the trailer
1616 				 * space is regarded as data space.
1617 				 */
1618 				datalen += rt->dst.trailer_len;
1619 			}
1620 
1621 			fraglen = datalen + fragheaderlen;
1622 
1623 			copy = datalen - transhdrlen - fraggap - pagedlen;
1624 			if (copy < 0) {
1625 				err = -EINVAL;
1626 				goto error;
1627 			}
1628 			if (transhdrlen) {
1629 				skb = sock_alloc_send_skb(sk, alloclen,
1630 						(flags & MSG_DONTWAIT), &err);
1631 			} else {
1632 				skb = NULL;
1633 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1634 				    2 * sk->sk_sndbuf)
1635 					skb = alloc_skb(alloclen,
1636 							sk->sk_allocation);
1637 				if (unlikely(!skb))
1638 					err = -ENOBUFS;
1639 			}
1640 			if (!skb)
1641 				goto error;
1642 			/*
1643 			 *	Fill in the control structures
1644 			 */
1645 			skb->protocol = htons(ETH_P_IPV6);
1646 			skb->ip_summed = csummode;
1647 			skb->csum = 0;
1648 			/* reserve for fragmentation and ipsec header */
1649 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1650 				    dst_exthdrlen);
1651 
1652 			/*
1653 			 *	Find where to start putting bytes
1654 			 */
1655 			data = skb_put(skb, fraglen - pagedlen);
1656 			skb_set_network_header(skb, exthdrlen);
1657 			data += fragheaderlen;
1658 			skb->transport_header = (skb->network_header +
1659 						 fragheaderlen);
1660 			if (fraggap) {
1661 				skb->csum = skb_copy_and_csum_bits(
1662 					skb_prev, maxfraglen,
1663 					data + transhdrlen, fraggap);
1664 				skb_prev->csum = csum_sub(skb_prev->csum,
1665 							  skb->csum);
1666 				data += fraggap;
1667 				pskb_trim_unique(skb_prev, maxfraglen);
1668 			}
1669 			if (copy > 0 &&
1670 			    getfrag(from, data + transhdrlen, offset,
1671 				    copy, fraggap, skb) < 0) {
1672 				err = -EFAULT;
1673 				kfree_skb(skb);
1674 				goto error;
1675 			}
1676 
1677 			offset += copy;
1678 			length -= copy + transhdrlen;
1679 			transhdrlen = 0;
1680 			exthdrlen = 0;
1681 			dst_exthdrlen = 0;
1682 
1683 			/* Only the initial fragment is time stamped */
1684 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1685 			cork->tx_flags = 0;
1686 			skb_shinfo(skb)->tskey = tskey;
1687 			tskey = 0;
1688 			skb_zcopy_set(skb, uarg, &extra_uref);
1689 
1690 			if ((flags & MSG_CONFIRM) && !skb_prev)
1691 				skb_set_dst_pending_confirm(skb, 1);
1692 
1693 			/*
1694 			 * Put the packet on the pending queue
1695 			 */
1696 			if (!skb->destructor) {
1697 				skb->destructor = sock_wfree;
1698 				skb->sk = sk;
1699 				wmem_alloc_delta += skb->truesize;
1700 			}
1701 			__skb_queue_tail(queue, skb);
1702 			continue;
1703 		}
1704 
1705 		if (copy > length)
1706 			copy = length;
1707 
1708 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1709 		    skb_tailroom(skb) >= copy) {
1710 			unsigned int off;
1711 
1712 			off = skb->len;
1713 			if (getfrag(from, skb_put(skb, copy),
1714 						offset, copy, off, skb) < 0) {
1715 				__skb_trim(skb, off);
1716 				err = -EFAULT;
1717 				goto error;
1718 			}
1719 		} else if (!uarg || !uarg->zerocopy) {
1720 			int i = skb_shinfo(skb)->nr_frags;
1721 
1722 			err = -ENOMEM;
1723 			if (!sk_page_frag_refill(sk, pfrag))
1724 				goto error;
1725 
1726 			if (!skb_can_coalesce(skb, i, pfrag->page,
1727 					      pfrag->offset)) {
1728 				err = -EMSGSIZE;
1729 				if (i == MAX_SKB_FRAGS)
1730 					goto error;
1731 
1732 				__skb_fill_page_desc(skb, i, pfrag->page,
1733 						     pfrag->offset, 0);
1734 				skb_shinfo(skb)->nr_frags = ++i;
1735 				get_page(pfrag->page);
1736 			}
1737 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1738 			if (getfrag(from,
1739 				    page_address(pfrag->page) + pfrag->offset,
1740 				    offset, copy, skb->len, skb) < 0)
1741 				goto error_efault;
1742 
1743 			pfrag->offset += copy;
1744 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1745 			skb->len += copy;
1746 			skb->data_len += copy;
1747 			skb->truesize += copy;
1748 			wmem_alloc_delta += copy;
1749 		} else {
1750 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1751 			if (err < 0)
1752 				goto error;
1753 		}
1754 		offset += copy;
1755 		length -= copy;
1756 	}
1757 
1758 	if (wmem_alloc_delta)
1759 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1760 	return 0;
1761 
1762 error_efault:
1763 	err = -EFAULT;
1764 error:
1765 	net_zcopy_put_abort(uarg, extra_uref);
1766 	cork->length -= length;
1767 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1768 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1769 	return err;
1770 }
1771 
1772 int ip6_append_data(struct sock *sk,
1773 		    int getfrag(void *from, char *to, int offset, int len,
1774 				int odd, struct sk_buff *skb),
1775 		    void *from, int length, int transhdrlen,
1776 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1777 		    struct rt6_info *rt, unsigned int flags)
1778 {
1779 	struct inet_sock *inet = inet_sk(sk);
1780 	struct ipv6_pinfo *np = inet6_sk(sk);
1781 	int exthdrlen;
1782 	int err;
1783 
1784 	if (flags&MSG_PROBE)
1785 		return 0;
1786 	if (skb_queue_empty(&sk->sk_write_queue)) {
1787 		/*
1788 		 * setup for corking
1789 		 */
1790 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1791 				     ipc6, rt, fl6);
1792 		if (err)
1793 			return err;
1794 
1795 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1796 		length += exthdrlen;
1797 		transhdrlen += exthdrlen;
1798 	} else {
1799 		fl6 = &inet->cork.fl.u.ip6;
1800 		transhdrlen = 0;
1801 	}
1802 
1803 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1804 				 &np->cork, sk_page_frag(sk), getfrag,
1805 				 from, length, transhdrlen, flags, ipc6);
1806 }
1807 EXPORT_SYMBOL_GPL(ip6_append_data);
1808 
1809 static void ip6_cork_release(struct inet_cork_full *cork,
1810 			     struct inet6_cork *v6_cork)
1811 {
1812 	if (v6_cork->opt) {
1813 		kfree(v6_cork->opt->dst0opt);
1814 		kfree(v6_cork->opt->dst1opt);
1815 		kfree(v6_cork->opt->hopopt);
1816 		kfree(v6_cork->opt->srcrt);
1817 		kfree(v6_cork->opt);
1818 		v6_cork->opt = NULL;
1819 	}
1820 
1821 	if (cork->base.dst) {
1822 		dst_release(cork->base.dst);
1823 		cork->base.dst = NULL;
1824 		cork->base.flags &= ~IPCORK_ALLFRAG;
1825 	}
1826 	memset(&cork->fl, 0, sizeof(cork->fl));
1827 }
1828 
1829 struct sk_buff *__ip6_make_skb(struct sock *sk,
1830 			       struct sk_buff_head *queue,
1831 			       struct inet_cork_full *cork,
1832 			       struct inet6_cork *v6_cork)
1833 {
1834 	struct sk_buff *skb, *tmp_skb;
1835 	struct sk_buff **tail_skb;
1836 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1837 	struct ipv6_pinfo *np = inet6_sk(sk);
1838 	struct net *net = sock_net(sk);
1839 	struct ipv6hdr *hdr;
1840 	struct ipv6_txoptions *opt = v6_cork->opt;
1841 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1842 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1843 	unsigned char proto = fl6->flowi6_proto;
1844 
1845 	skb = __skb_dequeue(queue);
1846 	if (!skb)
1847 		goto out;
1848 	tail_skb = &(skb_shinfo(skb)->frag_list);
1849 
1850 	/* move skb->data to ip header from ext header */
1851 	if (skb->data < skb_network_header(skb))
1852 		__skb_pull(skb, skb_network_offset(skb));
1853 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1854 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1855 		*tail_skb = tmp_skb;
1856 		tail_skb = &(tmp_skb->next);
1857 		skb->len += tmp_skb->len;
1858 		skb->data_len += tmp_skb->len;
1859 		skb->truesize += tmp_skb->truesize;
1860 		tmp_skb->destructor = NULL;
1861 		tmp_skb->sk = NULL;
1862 	}
1863 
1864 	/* Allow local fragmentation. */
1865 	skb->ignore_df = ip6_sk_ignore_df(sk);
1866 
1867 	*final_dst = fl6->daddr;
1868 	__skb_pull(skb, skb_network_header_len(skb));
1869 	if (opt && opt->opt_flen)
1870 		ipv6_push_frag_opts(skb, opt, &proto);
1871 	if (opt && opt->opt_nflen)
1872 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1873 
1874 	skb_push(skb, sizeof(struct ipv6hdr));
1875 	skb_reset_network_header(skb);
1876 	hdr = ipv6_hdr(skb);
1877 
1878 	ip6_flow_hdr(hdr, v6_cork->tclass,
1879 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1880 					ip6_autoflowlabel(net, np), fl6));
1881 	hdr->hop_limit = v6_cork->hop_limit;
1882 	hdr->nexthdr = proto;
1883 	hdr->saddr = fl6->saddr;
1884 	hdr->daddr = *final_dst;
1885 
1886 	skb->priority = sk->sk_priority;
1887 	skb->mark = cork->base.mark;
1888 
1889 	skb->tstamp = cork->base.transmit_time;
1890 
1891 	skb_dst_set(skb, dst_clone(&rt->dst));
1892 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1893 	if (proto == IPPROTO_ICMPV6) {
1894 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1895 
1896 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1897 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1898 	}
1899 
1900 	ip6_cork_release(cork, v6_cork);
1901 out:
1902 	return skb;
1903 }
1904 
1905 int ip6_send_skb(struct sk_buff *skb)
1906 {
1907 	struct net *net = sock_net(skb->sk);
1908 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1909 	int err;
1910 
1911 	err = ip6_local_out(net, skb->sk, skb);
1912 	if (err) {
1913 		if (err > 0)
1914 			err = net_xmit_errno(err);
1915 		if (err)
1916 			IP6_INC_STATS(net, rt->rt6i_idev,
1917 				      IPSTATS_MIB_OUTDISCARDS);
1918 	}
1919 
1920 	return err;
1921 }
1922 
1923 int ip6_push_pending_frames(struct sock *sk)
1924 {
1925 	struct sk_buff *skb;
1926 
1927 	skb = ip6_finish_skb(sk);
1928 	if (!skb)
1929 		return 0;
1930 
1931 	return ip6_send_skb(skb);
1932 }
1933 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1934 
1935 static void __ip6_flush_pending_frames(struct sock *sk,
1936 				       struct sk_buff_head *queue,
1937 				       struct inet_cork_full *cork,
1938 				       struct inet6_cork *v6_cork)
1939 {
1940 	struct sk_buff *skb;
1941 
1942 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1943 		if (skb_dst(skb))
1944 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1945 				      IPSTATS_MIB_OUTDISCARDS);
1946 		kfree_skb(skb);
1947 	}
1948 
1949 	ip6_cork_release(cork, v6_cork);
1950 }
1951 
1952 void ip6_flush_pending_frames(struct sock *sk)
1953 {
1954 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1955 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1956 }
1957 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1958 
1959 struct sk_buff *ip6_make_skb(struct sock *sk,
1960 			     int getfrag(void *from, char *to, int offset,
1961 					 int len, int odd, struct sk_buff *skb),
1962 			     void *from, int length, int transhdrlen,
1963 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1964 			     struct rt6_info *rt, unsigned int flags,
1965 			     struct inet_cork_full *cork)
1966 {
1967 	struct inet6_cork v6_cork;
1968 	struct sk_buff_head queue;
1969 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1970 	int err;
1971 
1972 	if (flags & MSG_PROBE)
1973 		return NULL;
1974 
1975 	__skb_queue_head_init(&queue);
1976 
1977 	cork->base.flags = 0;
1978 	cork->base.addr = 0;
1979 	cork->base.opt = NULL;
1980 	cork->base.dst = NULL;
1981 	v6_cork.opt = NULL;
1982 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1983 	if (err) {
1984 		ip6_cork_release(cork, &v6_cork);
1985 		return ERR_PTR(err);
1986 	}
1987 	if (ipc6->dontfrag < 0)
1988 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1989 
1990 	err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1991 				&current->task_frag, getfrag, from,
1992 				length + exthdrlen, transhdrlen + exthdrlen,
1993 				flags, ipc6);
1994 	if (err) {
1995 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1996 		return ERR_PTR(err);
1997 	}
1998 
1999 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2000 }
2001