xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 7ae5c03a)
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 
123 	if (unlikely(IS_ERR_OR_NULL(neigh))) {
124 		if (unlikely(!neigh))
125 			neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
126 		if (IS_ERR(neigh)) {
127 			rcu_read_unlock_bh();
128 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
129 			kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
130 			return -EINVAL;
131 		}
132 	}
133 	sock_confirm_neigh(skb, neigh);
134 	ret = neigh_output(neigh, skb, false);
135 	rcu_read_unlock_bh();
136 	return ret;
137 }
138 
139 static int
140 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
141 				    struct sk_buff *skb, unsigned int mtu)
142 {
143 	struct sk_buff *segs, *nskb;
144 	netdev_features_t features;
145 	int ret = 0;
146 
147 	/* Please see corresponding comment in ip_finish_output_gso
148 	 * describing the cases where GSO segment length exceeds the
149 	 * egress MTU.
150 	 */
151 	features = netif_skb_features(skb);
152 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
153 	if (IS_ERR_OR_NULL(segs)) {
154 		kfree_skb(skb);
155 		return -ENOMEM;
156 	}
157 
158 	consume_skb(skb);
159 
160 	skb_list_walk_safe(segs, segs, nskb) {
161 		int err;
162 
163 		skb_mark_not_on_list(segs);
164 		err = ip6_fragment(net, sk, segs, ip6_finish_output2);
165 		if (err && ret == 0)
166 			ret = err;
167 	}
168 
169 	return ret;
170 }
171 
172 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
173 {
174 	unsigned int mtu;
175 
176 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
177 	/* Policy lookup after SNAT yielded a new policy */
178 	if (skb_dst(skb)->xfrm) {
179 		IP6CB(skb)->flags |= IP6SKB_REROUTED;
180 		return dst_output(net, sk, skb);
181 	}
182 #endif
183 
184 	mtu = ip6_skb_dst_mtu(skb);
185 	if (skb_is_gso(skb) &&
186 	    !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
187 	    !skb_gso_validate_network_len(skb, mtu))
188 		return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
189 
190 	if ((skb->len > mtu && !skb_is_gso(skb)) ||
191 	    dst_allfrag(skb_dst(skb)) ||
192 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
193 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
194 	else
195 		return ip6_finish_output2(net, sk, skb);
196 }
197 
198 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
199 {
200 	int ret;
201 
202 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
203 	switch (ret) {
204 	case NET_XMIT_SUCCESS:
205 	case NET_XMIT_CN:
206 		return __ip6_finish_output(net, sk, skb) ? : ret;
207 	default:
208 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
209 		return ret;
210 	}
211 }
212 
213 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
214 {
215 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
216 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
217 
218 	skb->protocol = htons(ETH_P_IPV6);
219 	skb->dev = dev;
220 
221 	if (unlikely(idev->cnf.disable_ipv6)) {
222 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
223 		kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
224 		return 0;
225 	}
226 
227 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
228 			    net, sk, skb, indev, dev,
229 			    ip6_finish_output,
230 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
231 }
232 EXPORT_SYMBOL(ip6_output);
233 
234 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
235 {
236 	if (!np->autoflowlabel_set)
237 		return ip6_default_np_autolabel(net);
238 	else
239 		return np->autoflowlabel;
240 }
241 
242 /*
243  * xmit an sk_buff (used by TCP, SCTP and DCCP)
244  * Note : socket lock is not held for SYNACK packets, but might be modified
245  * by calls to skb_set_owner_w() and ipv6_local_error(),
246  * which are using proper atomic operations or spinlocks.
247  */
248 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
249 	     __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
250 {
251 	struct net *net = sock_net(sk);
252 	const struct ipv6_pinfo *np = inet6_sk(sk);
253 	struct in6_addr *first_hop = &fl6->daddr;
254 	struct dst_entry *dst = skb_dst(skb);
255 	struct net_device *dev = dst->dev;
256 	struct inet6_dev *idev = ip6_dst_idev(dst);
257 	struct hop_jumbo_hdr *hop_jumbo;
258 	int hoplen = sizeof(*hop_jumbo);
259 	unsigned int head_room;
260 	struct ipv6hdr *hdr;
261 	u8  proto = fl6->flowi6_proto;
262 	int seg_len = skb->len;
263 	int hlimit = -1;
264 	u32 mtu;
265 
266 	head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
267 	if (opt)
268 		head_room += opt->opt_nflen + opt->opt_flen;
269 
270 	if (unlikely(head_room > skb_headroom(skb))) {
271 		skb = skb_expand_head(skb, head_room);
272 		if (!skb) {
273 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
274 			return -ENOBUFS;
275 		}
276 	}
277 
278 	if (opt) {
279 		seg_len += opt->opt_nflen + opt->opt_flen;
280 
281 		if (opt->opt_flen)
282 			ipv6_push_frag_opts(skb, opt, &proto);
283 
284 		if (opt->opt_nflen)
285 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
286 					     &fl6->saddr);
287 	}
288 
289 	if (unlikely(seg_len > IPV6_MAXPLEN)) {
290 		hop_jumbo = skb_push(skb, hoplen);
291 
292 		hop_jumbo->nexthdr = proto;
293 		hop_jumbo->hdrlen = 0;
294 		hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
295 		hop_jumbo->tlv_len = 4;
296 		hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
297 
298 		proto = IPPROTO_HOPOPTS;
299 		seg_len = 0;
300 		IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
301 	}
302 
303 	skb_push(skb, sizeof(struct ipv6hdr));
304 	skb_reset_network_header(skb);
305 	hdr = ipv6_hdr(skb);
306 
307 	/*
308 	 *	Fill in the IPv6 header
309 	 */
310 	if (np)
311 		hlimit = np->hop_limit;
312 	if (hlimit < 0)
313 		hlimit = ip6_dst_hoplimit(dst);
314 
315 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
316 				ip6_autoflowlabel(net, np), fl6));
317 
318 	hdr->payload_len = htons(seg_len);
319 	hdr->nexthdr = proto;
320 	hdr->hop_limit = hlimit;
321 
322 	hdr->saddr = fl6->saddr;
323 	hdr->daddr = *first_hop;
324 
325 	skb->protocol = htons(ETH_P_IPV6);
326 	skb->priority = priority;
327 	skb->mark = mark;
328 
329 	mtu = dst_mtu(dst);
330 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
331 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
332 
333 		/* if egress device is enslaved to an L3 master device pass the
334 		 * skb to its handler for processing
335 		 */
336 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
337 		if (unlikely(!skb))
338 			return 0;
339 
340 		/* hooks should never assume socket lock is held.
341 		 * we promote our socket to non const
342 		 */
343 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
344 			       net, (struct sock *)sk, skb, NULL, dev,
345 			       dst_output);
346 	}
347 
348 	skb->dev = dev;
349 	/* ipv6_local_error() does not require socket lock,
350 	 * we promote our socket to non const
351 	 */
352 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
353 
354 	IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
355 	kfree_skb(skb);
356 	return -EMSGSIZE;
357 }
358 EXPORT_SYMBOL(ip6_xmit);
359 
360 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
361 {
362 	struct ip6_ra_chain *ra;
363 	struct sock *last = NULL;
364 
365 	read_lock(&ip6_ra_lock);
366 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
367 		struct sock *sk = ra->sk;
368 		if (sk && ra->sel == sel &&
369 		    (!sk->sk_bound_dev_if ||
370 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
371 			struct ipv6_pinfo *np = inet6_sk(sk);
372 
373 			if (np && np->rtalert_isolate &&
374 			    !net_eq(sock_net(sk), dev_net(skb->dev))) {
375 				continue;
376 			}
377 			if (last) {
378 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
379 				if (skb2)
380 					rawv6_rcv(last, skb2);
381 			}
382 			last = sk;
383 		}
384 	}
385 
386 	if (last) {
387 		rawv6_rcv(last, skb);
388 		read_unlock(&ip6_ra_lock);
389 		return 1;
390 	}
391 	read_unlock(&ip6_ra_lock);
392 	return 0;
393 }
394 
395 static int ip6_forward_proxy_check(struct sk_buff *skb)
396 {
397 	struct ipv6hdr *hdr = ipv6_hdr(skb);
398 	u8 nexthdr = hdr->nexthdr;
399 	__be16 frag_off;
400 	int offset;
401 
402 	if (ipv6_ext_hdr(nexthdr)) {
403 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
404 		if (offset < 0)
405 			return 0;
406 	} else
407 		offset = sizeof(struct ipv6hdr);
408 
409 	if (nexthdr == IPPROTO_ICMPV6) {
410 		struct icmp6hdr *icmp6;
411 
412 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
413 					 offset + 1 - skb->data)))
414 			return 0;
415 
416 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
417 
418 		switch (icmp6->icmp6_type) {
419 		case NDISC_ROUTER_SOLICITATION:
420 		case NDISC_ROUTER_ADVERTISEMENT:
421 		case NDISC_NEIGHBOUR_SOLICITATION:
422 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
423 		case NDISC_REDIRECT:
424 			/* For reaction involving unicast neighbor discovery
425 			 * message destined to the proxied address, pass it to
426 			 * input function.
427 			 */
428 			return 1;
429 		default:
430 			break;
431 		}
432 	}
433 
434 	/*
435 	 * The proxying router can't forward traffic sent to a link-local
436 	 * address, so signal the sender and discard the packet. This
437 	 * behavior is clarified by the MIPv6 specification.
438 	 */
439 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
440 		dst_link_failure(skb);
441 		return -1;
442 	}
443 
444 	return 0;
445 }
446 
447 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
448 				     struct sk_buff *skb)
449 {
450 	struct dst_entry *dst = skb_dst(skb);
451 
452 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
453 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
454 
455 #ifdef CONFIG_NET_SWITCHDEV
456 	if (skb->offload_l3_fwd_mark) {
457 		consume_skb(skb);
458 		return 0;
459 	}
460 #endif
461 
462 	skb_clear_tstamp(skb);
463 	return dst_output(net, sk, skb);
464 }
465 
466 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
467 {
468 	if (skb->len <= mtu)
469 		return false;
470 
471 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
472 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
473 		return true;
474 
475 	if (skb->ignore_df)
476 		return false;
477 
478 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
479 		return false;
480 
481 	return true;
482 }
483 
484 int ip6_forward(struct sk_buff *skb)
485 {
486 	struct dst_entry *dst = skb_dst(skb);
487 	struct ipv6hdr *hdr = ipv6_hdr(skb);
488 	struct inet6_skb_parm *opt = IP6CB(skb);
489 	struct net *net = dev_net(dst->dev);
490 	struct inet6_dev *idev;
491 	SKB_DR(reason);
492 	u32 mtu;
493 
494 	idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
495 	if (net->ipv6.devconf_all->forwarding == 0)
496 		goto error;
497 
498 	if (skb->pkt_type != PACKET_HOST)
499 		goto drop;
500 
501 	if (unlikely(skb->sk))
502 		goto drop;
503 
504 	if (skb_warn_if_lro(skb))
505 		goto drop;
506 
507 	if (!net->ipv6.devconf_all->disable_policy &&
508 	    (!idev || !idev->cnf.disable_policy) &&
509 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
510 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
511 		goto drop;
512 	}
513 
514 	skb_forward_csum(skb);
515 
516 	/*
517 	 *	We DO NOT make any processing on
518 	 *	RA packets, pushing them to user level AS IS
519 	 *	without ane WARRANTY that application will be able
520 	 *	to interpret them. The reason is that we
521 	 *	cannot make anything clever here.
522 	 *
523 	 *	We are not end-node, so that if packet contains
524 	 *	AH/ESP, we cannot make anything.
525 	 *	Defragmentation also would be mistake, RA packets
526 	 *	cannot be fragmented, because there is no warranty
527 	 *	that different fragments will go along one path. --ANK
528 	 */
529 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
530 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
531 			return 0;
532 	}
533 
534 	/*
535 	 *	check and decrement ttl
536 	 */
537 	if (hdr->hop_limit <= 1) {
538 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
539 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
540 
541 		kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
542 		return -ETIMEDOUT;
543 	}
544 
545 	/* XXX: idev->cnf.proxy_ndp? */
546 	if (net->ipv6.devconf_all->proxy_ndp &&
547 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
548 		int proxied = ip6_forward_proxy_check(skb);
549 		if (proxied > 0) {
550 			hdr->hop_limit--;
551 			return ip6_input(skb);
552 		} else if (proxied < 0) {
553 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
554 			goto drop;
555 		}
556 	}
557 
558 	if (!xfrm6_route_forward(skb)) {
559 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
560 		SKB_DR_SET(reason, XFRM_POLICY);
561 		goto drop;
562 	}
563 	dst = skb_dst(skb);
564 
565 	/* IPv6 specs say nothing about it, but it is clear that we cannot
566 	   send redirects to source routed frames.
567 	   We don't send redirects to frames decapsulated from IPsec.
568 	 */
569 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
570 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
571 		struct in6_addr *target = NULL;
572 		struct inet_peer *peer;
573 		struct rt6_info *rt;
574 
575 		/*
576 		 *	incoming and outgoing devices are the same
577 		 *	send a redirect.
578 		 */
579 
580 		rt = (struct rt6_info *) dst;
581 		if (rt->rt6i_flags & RTF_GATEWAY)
582 			target = &rt->rt6i_gateway;
583 		else
584 			target = &hdr->daddr;
585 
586 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
587 
588 		/* Limit redirects both by destination (here)
589 		   and by source (inside ndisc_send_redirect)
590 		 */
591 		if (inet_peer_xrlim_allow(peer, 1*HZ))
592 			ndisc_send_redirect(skb, target);
593 		if (peer)
594 			inet_putpeer(peer);
595 	} else {
596 		int addrtype = ipv6_addr_type(&hdr->saddr);
597 
598 		/* This check is security critical. */
599 		if (addrtype == IPV6_ADDR_ANY ||
600 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
601 			goto error;
602 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
603 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
604 				    ICMPV6_NOT_NEIGHBOUR, 0);
605 			goto error;
606 		}
607 	}
608 
609 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
610 	if (mtu < IPV6_MIN_MTU)
611 		mtu = IPV6_MIN_MTU;
612 
613 	if (ip6_pkt_too_big(skb, mtu)) {
614 		/* Again, force OUTPUT device used as source address */
615 		skb->dev = dst->dev;
616 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
617 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
618 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
619 				IPSTATS_MIB_FRAGFAILS);
620 		kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
621 		return -EMSGSIZE;
622 	}
623 
624 	if (skb_cow(skb, dst->dev->hard_header_len)) {
625 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
626 				IPSTATS_MIB_OUTDISCARDS);
627 		goto drop;
628 	}
629 
630 	hdr = ipv6_hdr(skb);
631 
632 	/* Mangling hops number delayed to point after skb COW */
633 
634 	hdr->hop_limit--;
635 
636 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
637 		       net, NULL, skb, skb->dev, dst->dev,
638 		       ip6_forward_finish);
639 
640 error:
641 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
642 	SKB_DR_SET(reason, IP_INADDRERRORS);
643 drop:
644 	kfree_skb_reason(skb, reason);
645 	return -EINVAL;
646 }
647 
648 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
649 {
650 	to->pkt_type = from->pkt_type;
651 	to->priority = from->priority;
652 	to->protocol = from->protocol;
653 	skb_dst_drop(to);
654 	skb_dst_set(to, dst_clone(skb_dst(from)));
655 	to->dev = from->dev;
656 	to->mark = from->mark;
657 
658 	skb_copy_hash(to, from);
659 
660 #ifdef CONFIG_NET_SCHED
661 	to->tc_index = from->tc_index;
662 #endif
663 	nf_copy(to, from);
664 	skb_ext_copy(to, from);
665 	skb_copy_secmark(to, from);
666 }
667 
668 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
669 		      u8 nexthdr, __be32 frag_id,
670 		      struct ip6_fraglist_iter *iter)
671 {
672 	unsigned int first_len;
673 	struct frag_hdr *fh;
674 
675 	/* BUILD HEADER */
676 	*prevhdr = NEXTHDR_FRAGMENT;
677 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
678 	if (!iter->tmp_hdr)
679 		return -ENOMEM;
680 
681 	iter->frag = skb_shinfo(skb)->frag_list;
682 	skb_frag_list_init(skb);
683 
684 	iter->offset = 0;
685 	iter->hlen = hlen;
686 	iter->frag_id = frag_id;
687 	iter->nexthdr = nexthdr;
688 
689 	__skb_pull(skb, hlen);
690 	fh = __skb_push(skb, sizeof(struct frag_hdr));
691 	__skb_push(skb, hlen);
692 	skb_reset_network_header(skb);
693 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
694 
695 	fh->nexthdr = nexthdr;
696 	fh->reserved = 0;
697 	fh->frag_off = htons(IP6_MF);
698 	fh->identification = frag_id;
699 
700 	first_len = skb_pagelen(skb);
701 	skb->data_len = first_len - skb_headlen(skb);
702 	skb->len = first_len;
703 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
704 
705 	return 0;
706 }
707 EXPORT_SYMBOL(ip6_fraglist_init);
708 
709 void ip6_fraglist_prepare(struct sk_buff *skb,
710 			  struct ip6_fraglist_iter *iter)
711 {
712 	struct sk_buff *frag = iter->frag;
713 	unsigned int hlen = iter->hlen;
714 	struct frag_hdr *fh;
715 
716 	frag->ip_summed = CHECKSUM_NONE;
717 	skb_reset_transport_header(frag);
718 	fh = __skb_push(frag, sizeof(struct frag_hdr));
719 	__skb_push(frag, hlen);
720 	skb_reset_network_header(frag);
721 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
722 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
723 	fh->nexthdr = iter->nexthdr;
724 	fh->reserved = 0;
725 	fh->frag_off = htons(iter->offset);
726 	if (frag->next)
727 		fh->frag_off |= htons(IP6_MF);
728 	fh->identification = iter->frag_id;
729 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
730 	ip6_copy_metadata(frag, skb);
731 }
732 EXPORT_SYMBOL(ip6_fraglist_prepare);
733 
734 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
735 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
736 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
737 {
738 	state->prevhdr = prevhdr;
739 	state->nexthdr = nexthdr;
740 	state->frag_id = frag_id;
741 
742 	state->hlen = hlen;
743 	state->mtu = mtu;
744 
745 	state->left = skb->len - hlen;	/* Space per frame */
746 	state->ptr = hlen;		/* Where to start from */
747 
748 	state->hroom = hdr_room;
749 	state->troom = needed_tailroom;
750 
751 	state->offset = 0;
752 }
753 EXPORT_SYMBOL(ip6_frag_init);
754 
755 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
756 {
757 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
758 	struct sk_buff *frag;
759 	struct frag_hdr *fh;
760 	unsigned int len;
761 
762 	len = state->left;
763 	/* IF: it doesn't fit, use 'mtu' - the data space left */
764 	if (len > state->mtu)
765 		len = state->mtu;
766 	/* IF: we are not sending up to and including the packet end
767 	   then align the next start on an eight byte boundary */
768 	if (len < state->left)
769 		len &= ~7;
770 
771 	/* Allocate buffer */
772 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
773 			 state->hroom + state->troom, GFP_ATOMIC);
774 	if (!frag)
775 		return ERR_PTR(-ENOMEM);
776 
777 	/*
778 	 *	Set up data on packet
779 	 */
780 
781 	ip6_copy_metadata(frag, skb);
782 	skb_reserve(frag, state->hroom);
783 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
784 	skb_reset_network_header(frag);
785 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
786 	frag->transport_header = (frag->network_header + state->hlen +
787 				  sizeof(struct frag_hdr));
788 
789 	/*
790 	 *	Charge the memory for the fragment to any owner
791 	 *	it might possess
792 	 */
793 	if (skb->sk)
794 		skb_set_owner_w(frag, skb->sk);
795 
796 	/*
797 	 *	Copy the packet header into the new buffer.
798 	 */
799 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
800 
801 	fragnexthdr_offset = skb_network_header(frag);
802 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
803 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
804 
805 	/*
806 	 *	Build fragment header.
807 	 */
808 	fh->nexthdr = state->nexthdr;
809 	fh->reserved = 0;
810 	fh->identification = state->frag_id;
811 
812 	/*
813 	 *	Copy a block of the IP datagram.
814 	 */
815 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
816 			     len));
817 	state->left -= len;
818 
819 	fh->frag_off = htons(state->offset);
820 	if (state->left > 0)
821 		fh->frag_off |= htons(IP6_MF);
822 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
823 
824 	state->ptr += len;
825 	state->offset += len;
826 
827 	return frag;
828 }
829 EXPORT_SYMBOL(ip6_frag_next);
830 
831 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
832 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
833 {
834 	struct sk_buff *frag;
835 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
836 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
837 				inet6_sk(skb->sk) : NULL;
838 	bool mono_delivery_time = skb->mono_delivery_time;
839 	struct ip6_frag_state state;
840 	unsigned int mtu, hlen, nexthdr_offset;
841 	ktime_t tstamp = skb->tstamp;
842 	int hroom, err = 0;
843 	__be32 frag_id;
844 	u8 *prevhdr, nexthdr = 0;
845 
846 	err = ip6_find_1stfragopt(skb, &prevhdr);
847 	if (err < 0)
848 		goto fail;
849 	hlen = err;
850 	nexthdr = *prevhdr;
851 	nexthdr_offset = prevhdr - skb_network_header(skb);
852 
853 	mtu = ip6_skb_dst_mtu(skb);
854 
855 	/* We must not fragment if the socket is set to force MTU discovery
856 	 * or if the skb it not generated by a local socket.
857 	 */
858 	if (unlikely(!skb->ignore_df && skb->len > mtu))
859 		goto fail_toobig;
860 
861 	if (IP6CB(skb)->frag_max_size) {
862 		if (IP6CB(skb)->frag_max_size > mtu)
863 			goto fail_toobig;
864 
865 		/* don't send fragments larger than what we received */
866 		mtu = IP6CB(skb)->frag_max_size;
867 		if (mtu < IPV6_MIN_MTU)
868 			mtu = IPV6_MIN_MTU;
869 	}
870 
871 	if (np && np->frag_size < mtu) {
872 		if (np->frag_size)
873 			mtu = np->frag_size;
874 	}
875 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
876 		goto fail_toobig;
877 	mtu -= hlen + sizeof(struct frag_hdr);
878 
879 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
880 				    &ipv6_hdr(skb)->saddr);
881 
882 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
883 	    (err = skb_checksum_help(skb)))
884 		goto fail;
885 
886 	prevhdr = skb_network_header(skb) + nexthdr_offset;
887 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
888 	if (skb_has_frag_list(skb)) {
889 		unsigned int first_len = skb_pagelen(skb);
890 		struct ip6_fraglist_iter iter;
891 		struct sk_buff *frag2;
892 
893 		if (first_len - hlen > mtu ||
894 		    ((first_len - hlen) & 7) ||
895 		    skb_cloned(skb) ||
896 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
897 			goto slow_path;
898 
899 		skb_walk_frags(skb, frag) {
900 			/* Correct geometry. */
901 			if (frag->len > mtu ||
902 			    ((frag->len & 7) && frag->next) ||
903 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
904 				goto slow_path_clean;
905 
906 			/* Partially cloned skb? */
907 			if (skb_shared(frag))
908 				goto slow_path_clean;
909 
910 			BUG_ON(frag->sk);
911 			if (skb->sk) {
912 				frag->sk = skb->sk;
913 				frag->destructor = sock_wfree;
914 			}
915 			skb->truesize -= frag->truesize;
916 		}
917 
918 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
919 					&iter);
920 		if (err < 0)
921 			goto fail;
922 
923 		for (;;) {
924 			/* Prepare header of the next frame,
925 			 * before previous one went down. */
926 			if (iter.frag)
927 				ip6_fraglist_prepare(skb, &iter);
928 
929 			skb_set_delivery_time(skb, tstamp, mono_delivery_time);
930 			err = output(net, sk, skb);
931 			if (!err)
932 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
933 					      IPSTATS_MIB_FRAGCREATES);
934 
935 			if (err || !iter.frag)
936 				break;
937 
938 			skb = ip6_fraglist_next(&iter);
939 		}
940 
941 		kfree(iter.tmp_hdr);
942 
943 		if (err == 0) {
944 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
945 				      IPSTATS_MIB_FRAGOKS);
946 			return 0;
947 		}
948 
949 		kfree_skb_list(iter.frag);
950 
951 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
952 			      IPSTATS_MIB_FRAGFAILS);
953 		return err;
954 
955 slow_path_clean:
956 		skb_walk_frags(skb, frag2) {
957 			if (frag2 == frag)
958 				break;
959 			frag2->sk = NULL;
960 			frag2->destructor = NULL;
961 			skb->truesize += frag2->truesize;
962 		}
963 	}
964 
965 slow_path:
966 	/*
967 	 *	Fragment the datagram.
968 	 */
969 
970 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
971 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
972 		      &state);
973 
974 	/*
975 	 *	Keep copying data until we run out.
976 	 */
977 
978 	while (state.left > 0) {
979 		frag = ip6_frag_next(skb, &state);
980 		if (IS_ERR(frag)) {
981 			err = PTR_ERR(frag);
982 			goto fail;
983 		}
984 
985 		/*
986 		 *	Put this fragment into the sending queue.
987 		 */
988 		skb_set_delivery_time(frag, tstamp, mono_delivery_time);
989 		err = output(net, sk, frag);
990 		if (err)
991 			goto fail;
992 
993 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
994 			      IPSTATS_MIB_FRAGCREATES);
995 	}
996 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
997 		      IPSTATS_MIB_FRAGOKS);
998 	consume_skb(skb);
999 	return err;
1000 
1001 fail_toobig:
1002 	if (skb->sk && dst_allfrag(skb_dst(skb)))
1003 		sk_gso_disable(skb->sk);
1004 
1005 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1006 	err = -EMSGSIZE;
1007 
1008 fail:
1009 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1010 		      IPSTATS_MIB_FRAGFAILS);
1011 	kfree_skb(skb);
1012 	return err;
1013 }
1014 
1015 static inline int ip6_rt_check(const struct rt6key *rt_key,
1016 			       const struct in6_addr *fl_addr,
1017 			       const struct in6_addr *addr_cache)
1018 {
1019 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1020 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1021 }
1022 
1023 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1024 					  struct dst_entry *dst,
1025 					  const struct flowi6 *fl6)
1026 {
1027 	struct ipv6_pinfo *np = inet6_sk(sk);
1028 	struct rt6_info *rt;
1029 
1030 	if (!dst)
1031 		goto out;
1032 
1033 	if (dst->ops->family != AF_INET6) {
1034 		dst_release(dst);
1035 		return NULL;
1036 	}
1037 
1038 	rt = (struct rt6_info *)dst;
1039 	/* Yes, checking route validity in not connected
1040 	 * case is not very simple. Take into account,
1041 	 * that we do not support routing by source, TOS,
1042 	 * and MSG_DONTROUTE		--ANK (980726)
1043 	 *
1044 	 * 1. ip6_rt_check(): If route was host route,
1045 	 *    check that cached destination is current.
1046 	 *    If it is network route, we still may
1047 	 *    check its validity using saved pointer
1048 	 *    to the last used address: daddr_cache.
1049 	 *    We do not want to save whole address now,
1050 	 *    (because main consumer of this service
1051 	 *    is tcp, which has not this problem),
1052 	 *    so that the last trick works only on connected
1053 	 *    sockets.
1054 	 * 2. oif also should be the same.
1055 	 */
1056 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1057 #ifdef CONFIG_IPV6_SUBTREES
1058 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1059 #endif
1060 	   (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1061 		dst_release(dst);
1062 		dst = NULL;
1063 	}
1064 
1065 out:
1066 	return dst;
1067 }
1068 
1069 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1070 			       struct dst_entry **dst, struct flowi6 *fl6)
1071 {
1072 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1073 	struct neighbour *n;
1074 	struct rt6_info *rt;
1075 #endif
1076 	int err;
1077 	int flags = 0;
1078 
1079 	/* The correct way to handle this would be to do
1080 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1081 	 * the route-specific preferred source forces the
1082 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1083 	 *
1084 	 * In source specific routing (no src=any default route),
1085 	 * ip6_route_output will fail given src=any saddr, though, so
1086 	 * that's why we try it again later.
1087 	 */
1088 	if (ipv6_addr_any(&fl6->saddr)) {
1089 		struct fib6_info *from;
1090 		struct rt6_info *rt;
1091 
1092 		*dst = ip6_route_output(net, sk, fl6);
1093 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1094 
1095 		rcu_read_lock();
1096 		from = rt ? rcu_dereference(rt->from) : NULL;
1097 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1098 					  sk ? inet6_sk(sk)->srcprefs : 0,
1099 					  &fl6->saddr);
1100 		rcu_read_unlock();
1101 
1102 		if (err)
1103 			goto out_err_release;
1104 
1105 		/* If we had an erroneous initial result, pretend it
1106 		 * never existed and let the SA-enabled version take
1107 		 * over.
1108 		 */
1109 		if ((*dst)->error) {
1110 			dst_release(*dst);
1111 			*dst = NULL;
1112 		}
1113 
1114 		if (fl6->flowi6_oif)
1115 			flags |= RT6_LOOKUP_F_IFACE;
1116 	}
1117 
1118 	if (!*dst)
1119 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1120 
1121 	err = (*dst)->error;
1122 	if (err)
1123 		goto out_err_release;
1124 
1125 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1126 	/*
1127 	 * Here if the dst entry we've looked up
1128 	 * has a neighbour entry that is in the INCOMPLETE
1129 	 * state and the src address from the flow is
1130 	 * marked as OPTIMISTIC, we release the found
1131 	 * dst entry and replace it instead with the
1132 	 * dst entry of the nexthop router
1133 	 */
1134 	rt = (struct rt6_info *) *dst;
1135 	rcu_read_lock_bh();
1136 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1137 				      rt6_nexthop(rt, &fl6->daddr));
1138 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1139 	rcu_read_unlock_bh();
1140 
1141 	if (err) {
1142 		struct inet6_ifaddr *ifp;
1143 		struct flowi6 fl_gw6;
1144 		int redirect;
1145 
1146 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1147 				      (*dst)->dev, 1);
1148 
1149 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1150 		if (ifp)
1151 			in6_ifa_put(ifp);
1152 
1153 		if (redirect) {
1154 			/*
1155 			 * We need to get the dst entry for the
1156 			 * default router instead
1157 			 */
1158 			dst_release(*dst);
1159 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1160 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1161 			*dst = ip6_route_output(net, sk, &fl_gw6);
1162 			err = (*dst)->error;
1163 			if (err)
1164 				goto out_err_release;
1165 		}
1166 	}
1167 #endif
1168 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1169 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1170 		err = -EAFNOSUPPORT;
1171 		goto out_err_release;
1172 	}
1173 
1174 	return 0;
1175 
1176 out_err_release:
1177 	dst_release(*dst);
1178 	*dst = NULL;
1179 
1180 	if (err == -ENETUNREACH)
1181 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1182 	return err;
1183 }
1184 
1185 /**
1186  *	ip6_dst_lookup - perform route lookup on flow
1187  *	@net: Network namespace to perform lookup in
1188  *	@sk: socket which provides route info
1189  *	@dst: pointer to dst_entry * for result
1190  *	@fl6: flow to lookup
1191  *
1192  *	This function performs a route lookup on the given flow.
1193  *
1194  *	It returns zero on success, or a standard errno code on error.
1195  */
1196 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1197 		   struct flowi6 *fl6)
1198 {
1199 	*dst = NULL;
1200 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1201 }
1202 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1203 
1204 /**
1205  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1206  *	@net: Network namespace to perform lookup in
1207  *	@sk: socket which provides route info
1208  *	@fl6: flow to lookup
1209  *	@final_dst: final destination address for ipsec lookup
1210  *
1211  *	This function performs a route lookup on the given flow.
1212  *
1213  *	It returns a valid dst pointer on success, or a pointer encoded
1214  *	error code.
1215  */
1216 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1217 				      const struct in6_addr *final_dst)
1218 {
1219 	struct dst_entry *dst = NULL;
1220 	int err;
1221 
1222 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1223 	if (err)
1224 		return ERR_PTR(err);
1225 	if (final_dst)
1226 		fl6->daddr = *final_dst;
1227 
1228 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1229 }
1230 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1231 
1232 /**
1233  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1234  *	@sk: socket which provides the dst cache and route info
1235  *	@fl6: flow to lookup
1236  *	@final_dst: final destination address for ipsec lookup
1237  *	@connected: whether @sk is connected or not
1238  *
1239  *	This function performs a route lookup on the given flow with the
1240  *	possibility of using the cached route in the socket if it is valid.
1241  *	It will take the socket dst lock when operating on the dst cache.
1242  *	As a result, this function can only be used in process context.
1243  *
1244  *	In addition, for a connected socket, cache the dst in the socket
1245  *	if the current cache is not valid.
1246  *
1247  *	It returns a valid dst pointer on success, or a pointer encoded
1248  *	error code.
1249  */
1250 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1251 					 const struct in6_addr *final_dst,
1252 					 bool connected)
1253 {
1254 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1255 
1256 	dst = ip6_sk_dst_check(sk, dst, fl6);
1257 	if (dst)
1258 		return dst;
1259 
1260 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1261 	if (connected && !IS_ERR(dst))
1262 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1263 
1264 	return dst;
1265 }
1266 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1267 
1268 /**
1269  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1270  *      @skb: Packet for which lookup is done
1271  *      @dev: Tunnel device
1272  *      @net: Network namespace of tunnel device
1273  *      @sock: Socket which provides route info
1274  *      @saddr: Memory to store the src ip address
1275  *      @info: Tunnel information
1276  *      @protocol: IP protocol
1277  *      @use_cache: Flag to enable cache usage
1278  *      This function performs a route lookup on a tunnel
1279  *
1280  *      It returns a valid dst pointer and stores src address to be used in
1281  *      tunnel in param saddr on success, else a pointer encoded error code.
1282  */
1283 
1284 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1285 					struct net_device *dev,
1286 					struct net *net,
1287 					struct socket *sock,
1288 					struct in6_addr *saddr,
1289 					const struct ip_tunnel_info *info,
1290 					u8 protocol,
1291 					bool use_cache)
1292 {
1293 	struct dst_entry *dst = NULL;
1294 #ifdef CONFIG_DST_CACHE
1295 	struct dst_cache *dst_cache;
1296 #endif
1297 	struct flowi6 fl6;
1298 	__u8 prio;
1299 
1300 #ifdef CONFIG_DST_CACHE
1301 	dst_cache = (struct dst_cache *)&info->dst_cache;
1302 	if (use_cache) {
1303 		dst = dst_cache_get_ip6(dst_cache, saddr);
1304 		if (dst)
1305 			return dst;
1306 	}
1307 #endif
1308 	memset(&fl6, 0, sizeof(fl6));
1309 	fl6.flowi6_mark = skb->mark;
1310 	fl6.flowi6_proto = protocol;
1311 	fl6.daddr = info->key.u.ipv6.dst;
1312 	fl6.saddr = info->key.u.ipv6.src;
1313 	prio = info->key.tos;
1314 	fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1315 					  info->key.label);
1316 
1317 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1318 					      NULL);
1319 	if (IS_ERR(dst)) {
1320 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1321 		return ERR_PTR(-ENETUNREACH);
1322 	}
1323 	if (dst->dev == dev) { /* is this necessary? */
1324 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1325 		dst_release(dst);
1326 		return ERR_PTR(-ELOOP);
1327 	}
1328 #ifdef CONFIG_DST_CACHE
1329 	if (use_cache)
1330 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1331 #endif
1332 	*saddr = fl6.saddr;
1333 	return dst;
1334 }
1335 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1336 
1337 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1338 					       gfp_t gfp)
1339 {
1340 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1341 }
1342 
1343 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1344 						gfp_t gfp)
1345 {
1346 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1347 }
1348 
1349 static void ip6_append_data_mtu(unsigned int *mtu,
1350 				int *maxfraglen,
1351 				unsigned int fragheaderlen,
1352 				struct sk_buff *skb,
1353 				struct rt6_info *rt,
1354 				unsigned int orig_mtu)
1355 {
1356 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1357 		if (!skb) {
1358 			/* first fragment, reserve header_len */
1359 			*mtu = orig_mtu - rt->dst.header_len;
1360 
1361 		} else {
1362 			/*
1363 			 * this fragment is not first, the headers
1364 			 * space is regarded as data space.
1365 			 */
1366 			*mtu = orig_mtu;
1367 		}
1368 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1369 			      + fragheaderlen - sizeof(struct frag_hdr);
1370 	}
1371 }
1372 
1373 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1374 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1375 			  struct rt6_info *rt)
1376 {
1377 	struct ipv6_pinfo *np = inet6_sk(sk);
1378 	unsigned int mtu;
1379 	struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1380 
1381 	/* callers pass dst together with a reference, set it first so
1382 	 * ip6_cork_release() can put it down even in case of an error.
1383 	 */
1384 	cork->base.dst = &rt->dst;
1385 
1386 	/*
1387 	 * setup for corking
1388 	 */
1389 	if (opt) {
1390 		if (WARN_ON(v6_cork->opt))
1391 			return -EINVAL;
1392 
1393 		nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1394 		if (unlikely(!nopt))
1395 			return -ENOBUFS;
1396 
1397 		nopt->tot_len = sizeof(*opt);
1398 		nopt->opt_flen = opt->opt_flen;
1399 		nopt->opt_nflen = opt->opt_nflen;
1400 
1401 		nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1402 		if (opt->dst0opt && !nopt->dst0opt)
1403 			return -ENOBUFS;
1404 
1405 		nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1406 		if (opt->dst1opt && !nopt->dst1opt)
1407 			return -ENOBUFS;
1408 
1409 		nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1410 		if (opt->hopopt && !nopt->hopopt)
1411 			return -ENOBUFS;
1412 
1413 		nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1414 		if (opt->srcrt && !nopt->srcrt)
1415 			return -ENOBUFS;
1416 
1417 		/* need source address above miyazawa*/
1418 	}
1419 	v6_cork->hop_limit = ipc6->hlimit;
1420 	v6_cork->tclass = ipc6->tclass;
1421 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1422 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1423 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1424 	else
1425 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1426 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1427 	if (np->frag_size < mtu) {
1428 		if (np->frag_size)
1429 			mtu = np->frag_size;
1430 	}
1431 	cork->base.fragsize = mtu;
1432 	cork->base.gso_size = ipc6->gso_size;
1433 	cork->base.tx_flags = 0;
1434 	cork->base.mark = ipc6->sockc.mark;
1435 	sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1436 
1437 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1438 		cork->base.flags |= IPCORK_ALLFRAG;
1439 	cork->base.length = 0;
1440 
1441 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1442 
1443 	return 0;
1444 }
1445 
1446 static int __ip6_append_data(struct sock *sk,
1447 			     struct sk_buff_head *queue,
1448 			     struct inet_cork_full *cork_full,
1449 			     struct inet6_cork *v6_cork,
1450 			     struct page_frag *pfrag,
1451 			     int getfrag(void *from, char *to, int offset,
1452 					 int len, int odd, struct sk_buff *skb),
1453 			     void *from, size_t length, int transhdrlen,
1454 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1455 {
1456 	struct sk_buff *skb, *skb_prev = NULL;
1457 	struct inet_cork *cork = &cork_full->base;
1458 	struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1459 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1460 	struct ubuf_info *uarg = NULL;
1461 	int exthdrlen = 0;
1462 	int dst_exthdrlen = 0;
1463 	int hh_len;
1464 	int copy;
1465 	int err;
1466 	int offset = 0;
1467 	bool zc = false;
1468 	u32 tskey = 0;
1469 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1470 	struct ipv6_txoptions *opt = v6_cork->opt;
1471 	int csummode = CHECKSUM_NONE;
1472 	unsigned int maxnonfragsize, headersize;
1473 	unsigned int wmem_alloc_delta = 0;
1474 	bool paged, extra_uref = false;
1475 
1476 	skb = skb_peek_tail(queue);
1477 	if (!skb) {
1478 		exthdrlen = opt ? opt->opt_flen : 0;
1479 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1480 	}
1481 
1482 	paged = !!cork->gso_size;
1483 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1484 	orig_mtu = mtu;
1485 
1486 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1487 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1488 		tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1489 
1490 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1491 
1492 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1493 			(opt ? opt->opt_nflen : 0);
1494 
1495 	headersize = sizeof(struct ipv6hdr) +
1496 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1497 		     (dst_allfrag(&rt->dst) ?
1498 		      sizeof(struct frag_hdr) : 0) +
1499 		     rt->rt6i_nfheader_len;
1500 
1501 	if (mtu <= fragheaderlen ||
1502 	    ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1503 		goto emsgsize;
1504 
1505 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1506 		     sizeof(struct frag_hdr);
1507 
1508 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1509 	 * the first fragment
1510 	 */
1511 	if (headersize + transhdrlen > mtu)
1512 		goto emsgsize;
1513 
1514 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1515 	    (sk->sk_protocol == IPPROTO_UDP ||
1516 	     sk->sk_protocol == IPPROTO_ICMPV6 ||
1517 	     sk->sk_protocol == IPPROTO_RAW)) {
1518 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1519 				sizeof(struct ipv6hdr));
1520 		goto emsgsize;
1521 	}
1522 
1523 	if (ip6_sk_ignore_df(sk))
1524 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1525 	else
1526 		maxnonfragsize = mtu;
1527 
1528 	if (cork->length + length > maxnonfragsize - headersize) {
1529 emsgsize:
1530 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1531 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1532 		return -EMSGSIZE;
1533 	}
1534 
1535 	/* CHECKSUM_PARTIAL only with no extension headers and when
1536 	 * we are not going to fragment
1537 	 */
1538 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1539 	    headersize == sizeof(struct ipv6hdr) &&
1540 	    length <= mtu - headersize &&
1541 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1542 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1543 		csummode = CHECKSUM_PARTIAL;
1544 
1545 	if ((flags & MSG_ZEROCOPY) && length) {
1546 		struct msghdr *msg = from;
1547 
1548 		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1549 			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1550 				return -EINVAL;
1551 
1552 			/* Leave uarg NULL if can't zerocopy, callers should
1553 			 * be able to handle it.
1554 			 */
1555 			if ((rt->dst.dev->features & NETIF_F_SG) &&
1556 			    csummode == CHECKSUM_PARTIAL) {
1557 				paged = true;
1558 				zc = true;
1559 				uarg = msg->msg_ubuf;
1560 			}
1561 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1562 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1563 			if (!uarg)
1564 				return -ENOBUFS;
1565 			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1566 			if (rt->dst.dev->features & NETIF_F_SG &&
1567 			    csummode == CHECKSUM_PARTIAL) {
1568 				paged = true;
1569 				zc = true;
1570 			} else {
1571 				uarg->zerocopy = 0;
1572 				skb_zcopy_set(skb, uarg, &extra_uref);
1573 			}
1574 		}
1575 	}
1576 
1577 	/*
1578 	 * Let's try using as much space as possible.
1579 	 * Use MTU if total length of the message fits into the MTU.
1580 	 * Otherwise, we need to reserve fragment header and
1581 	 * fragment alignment (= 8-15 octects, in total).
1582 	 *
1583 	 * Note that we may need to "move" the data from the tail
1584 	 * of the buffer to the new fragment when we split
1585 	 * the message.
1586 	 *
1587 	 * FIXME: It may be fragmented into multiple chunks
1588 	 *        at once if non-fragmentable extension headers
1589 	 *        are too large.
1590 	 * --yoshfuji
1591 	 */
1592 
1593 	cork->length += length;
1594 	if (!skb)
1595 		goto alloc_new_skb;
1596 
1597 	while (length > 0) {
1598 		/* Check if the remaining data fits into current packet. */
1599 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1600 		if (copy < length)
1601 			copy = maxfraglen - skb->len;
1602 
1603 		if (copy <= 0) {
1604 			char *data;
1605 			unsigned int datalen;
1606 			unsigned int fraglen;
1607 			unsigned int fraggap;
1608 			unsigned int alloclen, alloc_extra;
1609 			unsigned int pagedlen;
1610 alloc_new_skb:
1611 			/* There's no room in the current skb */
1612 			if (skb)
1613 				fraggap = skb->len - maxfraglen;
1614 			else
1615 				fraggap = 0;
1616 			/* update mtu and maxfraglen if necessary */
1617 			if (!skb || !skb_prev)
1618 				ip6_append_data_mtu(&mtu, &maxfraglen,
1619 						    fragheaderlen, skb, rt,
1620 						    orig_mtu);
1621 
1622 			skb_prev = skb;
1623 
1624 			/*
1625 			 * If remaining data exceeds the mtu,
1626 			 * we know we need more fragment(s).
1627 			 */
1628 			datalen = length + fraggap;
1629 
1630 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1631 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1632 			fraglen = datalen + fragheaderlen;
1633 			pagedlen = 0;
1634 
1635 			alloc_extra = hh_len;
1636 			alloc_extra += dst_exthdrlen;
1637 			alloc_extra += rt->dst.trailer_len;
1638 
1639 			/* We just reserve space for fragment header.
1640 			 * Note: this may be overallocation if the message
1641 			 * (without MSG_MORE) fits into the MTU.
1642 			 */
1643 			alloc_extra += sizeof(struct frag_hdr);
1644 
1645 			if ((flags & MSG_MORE) &&
1646 			    !(rt->dst.dev->features&NETIF_F_SG))
1647 				alloclen = mtu;
1648 			else if (!paged &&
1649 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1650 				  !(rt->dst.dev->features & NETIF_F_SG)))
1651 				alloclen = fraglen;
1652 			else if (!zc) {
1653 				alloclen = min_t(int, fraglen, MAX_HEADER);
1654 				pagedlen = fraglen - alloclen;
1655 			} else {
1656 				alloclen = fragheaderlen + transhdrlen;
1657 				pagedlen = datalen - transhdrlen;
1658 			}
1659 			alloclen += alloc_extra;
1660 
1661 			if (datalen != length + fraggap) {
1662 				/*
1663 				 * this is not the last fragment, the trailer
1664 				 * space is regarded as data space.
1665 				 */
1666 				datalen += rt->dst.trailer_len;
1667 			}
1668 
1669 			fraglen = datalen + fragheaderlen;
1670 
1671 			copy = datalen - transhdrlen - fraggap - pagedlen;
1672 			if (copy < 0) {
1673 				err = -EINVAL;
1674 				goto error;
1675 			}
1676 			if (transhdrlen) {
1677 				skb = sock_alloc_send_skb(sk, alloclen,
1678 						(flags & MSG_DONTWAIT), &err);
1679 			} else {
1680 				skb = NULL;
1681 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1682 				    2 * sk->sk_sndbuf)
1683 					skb = alloc_skb(alloclen,
1684 							sk->sk_allocation);
1685 				if (unlikely(!skb))
1686 					err = -ENOBUFS;
1687 			}
1688 			if (!skb)
1689 				goto error;
1690 			/*
1691 			 *	Fill in the control structures
1692 			 */
1693 			skb->protocol = htons(ETH_P_IPV6);
1694 			skb->ip_summed = csummode;
1695 			skb->csum = 0;
1696 			/* reserve for fragmentation and ipsec header */
1697 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1698 				    dst_exthdrlen);
1699 
1700 			/*
1701 			 *	Find where to start putting bytes
1702 			 */
1703 			data = skb_put(skb, fraglen - pagedlen);
1704 			skb_set_network_header(skb, exthdrlen);
1705 			data += fragheaderlen;
1706 			skb->transport_header = (skb->network_header +
1707 						 fragheaderlen);
1708 			if (fraggap) {
1709 				skb->csum = skb_copy_and_csum_bits(
1710 					skb_prev, maxfraglen,
1711 					data + transhdrlen, fraggap);
1712 				skb_prev->csum = csum_sub(skb_prev->csum,
1713 							  skb->csum);
1714 				data += fraggap;
1715 				pskb_trim_unique(skb_prev, maxfraglen);
1716 			}
1717 			if (copy > 0 &&
1718 			    getfrag(from, data + transhdrlen, offset,
1719 				    copy, fraggap, skb) < 0) {
1720 				err = -EFAULT;
1721 				kfree_skb(skb);
1722 				goto error;
1723 			}
1724 
1725 			offset += copy;
1726 			length -= copy + transhdrlen;
1727 			transhdrlen = 0;
1728 			exthdrlen = 0;
1729 			dst_exthdrlen = 0;
1730 
1731 			/* Only the initial fragment is time stamped */
1732 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1733 			cork->tx_flags = 0;
1734 			skb_shinfo(skb)->tskey = tskey;
1735 			tskey = 0;
1736 			skb_zcopy_set(skb, uarg, &extra_uref);
1737 
1738 			if ((flags & MSG_CONFIRM) && !skb_prev)
1739 				skb_set_dst_pending_confirm(skb, 1);
1740 
1741 			/*
1742 			 * Put the packet on the pending queue
1743 			 */
1744 			if (!skb->destructor) {
1745 				skb->destructor = sock_wfree;
1746 				skb->sk = sk;
1747 				wmem_alloc_delta += skb->truesize;
1748 			}
1749 			__skb_queue_tail(queue, skb);
1750 			continue;
1751 		}
1752 
1753 		if (copy > length)
1754 			copy = length;
1755 
1756 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1757 		    skb_tailroom(skb) >= copy) {
1758 			unsigned int off;
1759 
1760 			off = skb->len;
1761 			if (getfrag(from, skb_put(skb, copy),
1762 						offset, copy, off, skb) < 0) {
1763 				__skb_trim(skb, off);
1764 				err = -EFAULT;
1765 				goto error;
1766 			}
1767 		} else if (!zc) {
1768 			int i = skb_shinfo(skb)->nr_frags;
1769 
1770 			err = -ENOMEM;
1771 			if (!sk_page_frag_refill(sk, pfrag))
1772 				goto error;
1773 
1774 			skb_zcopy_downgrade_managed(skb);
1775 			if (!skb_can_coalesce(skb, i, pfrag->page,
1776 					      pfrag->offset)) {
1777 				err = -EMSGSIZE;
1778 				if (i == MAX_SKB_FRAGS)
1779 					goto error;
1780 
1781 				__skb_fill_page_desc(skb, i, pfrag->page,
1782 						     pfrag->offset, 0);
1783 				skb_shinfo(skb)->nr_frags = ++i;
1784 				get_page(pfrag->page);
1785 			}
1786 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1787 			if (getfrag(from,
1788 				    page_address(pfrag->page) + pfrag->offset,
1789 				    offset, copy, skb->len, skb) < 0)
1790 				goto error_efault;
1791 
1792 			pfrag->offset += copy;
1793 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1794 			skb->len += copy;
1795 			skb->data_len += copy;
1796 			skb->truesize += copy;
1797 			wmem_alloc_delta += copy;
1798 		} else {
1799 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1800 			if (err < 0)
1801 				goto error;
1802 		}
1803 		offset += copy;
1804 		length -= copy;
1805 	}
1806 
1807 	if (wmem_alloc_delta)
1808 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1809 	return 0;
1810 
1811 error_efault:
1812 	err = -EFAULT;
1813 error:
1814 	net_zcopy_put_abort(uarg, extra_uref);
1815 	cork->length -= length;
1816 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1817 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1818 	return err;
1819 }
1820 
1821 int ip6_append_data(struct sock *sk,
1822 		    int getfrag(void *from, char *to, int offset, int len,
1823 				int odd, struct sk_buff *skb),
1824 		    void *from, size_t length, int transhdrlen,
1825 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1826 		    struct rt6_info *rt, unsigned int flags)
1827 {
1828 	struct inet_sock *inet = inet_sk(sk);
1829 	struct ipv6_pinfo *np = inet6_sk(sk);
1830 	int exthdrlen;
1831 	int err;
1832 
1833 	if (flags&MSG_PROBE)
1834 		return 0;
1835 	if (skb_queue_empty(&sk->sk_write_queue)) {
1836 		/*
1837 		 * setup for corking
1838 		 */
1839 		dst_hold(&rt->dst);
1840 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1841 				     ipc6, rt);
1842 		if (err)
1843 			return err;
1844 
1845 		inet->cork.fl.u.ip6 = *fl6;
1846 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1847 		length += exthdrlen;
1848 		transhdrlen += exthdrlen;
1849 	} else {
1850 		transhdrlen = 0;
1851 	}
1852 
1853 	return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1854 				 &np->cork, sk_page_frag(sk), getfrag,
1855 				 from, length, transhdrlen, flags, ipc6);
1856 }
1857 EXPORT_SYMBOL_GPL(ip6_append_data);
1858 
1859 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1860 {
1861 	struct dst_entry *dst = cork->base.dst;
1862 
1863 	cork->base.dst = NULL;
1864 	cork->base.flags &= ~IPCORK_ALLFRAG;
1865 	skb_dst_set(skb, dst);
1866 }
1867 
1868 static void ip6_cork_release(struct inet_cork_full *cork,
1869 			     struct inet6_cork *v6_cork)
1870 {
1871 	if (v6_cork->opt) {
1872 		struct ipv6_txoptions *opt = v6_cork->opt;
1873 
1874 		kfree(opt->dst0opt);
1875 		kfree(opt->dst1opt);
1876 		kfree(opt->hopopt);
1877 		kfree(opt->srcrt);
1878 		kfree(opt);
1879 		v6_cork->opt = NULL;
1880 	}
1881 
1882 	if (cork->base.dst) {
1883 		dst_release(cork->base.dst);
1884 		cork->base.dst = NULL;
1885 		cork->base.flags &= ~IPCORK_ALLFRAG;
1886 	}
1887 }
1888 
1889 struct sk_buff *__ip6_make_skb(struct sock *sk,
1890 			       struct sk_buff_head *queue,
1891 			       struct inet_cork_full *cork,
1892 			       struct inet6_cork *v6_cork)
1893 {
1894 	struct sk_buff *skb, *tmp_skb;
1895 	struct sk_buff **tail_skb;
1896 	struct in6_addr *final_dst;
1897 	struct ipv6_pinfo *np = inet6_sk(sk);
1898 	struct net *net = sock_net(sk);
1899 	struct ipv6hdr *hdr;
1900 	struct ipv6_txoptions *opt = v6_cork->opt;
1901 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1902 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1903 	unsigned char proto = fl6->flowi6_proto;
1904 
1905 	skb = __skb_dequeue(queue);
1906 	if (!skb)
1907 		goto out;
1908 	tail_skb = &(skb_shinfo(skb)->frag_list);
1909 
1910 	/* move skb->data to ip header from ext header */
1911 	if (skb->data < skb_network_header(skb))
1912 		__skb_pull(skb, skb_network_offset(skb));
1913 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1914 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1915 		*tail_skb = tmp_skb;
1916 		tail_skb = &(tmp_skb->next);
1917 		skb->len += tmp_skb->len;
1918 		skb->data_len += tmp_skb->len;
1919 		skb->truesize += tmp_skb->truesize;
1920 		tmp_skb->destructor = NULL;
1921 		tmp_skb->sk = NULL;
1922 	}
1923 
1924 	/* Allow local fragmentation. */
1925 	skb->ignore_df = ip6_sk_ignore_df(sk);
1926 	__skb_pull(skb, skb_network_header_len(skb));
1927 
1928 	final_dst = &fl6->daddr;
1929 	if (opt && opt->opt_flen)
1930 		ipv6_push_frag_opts(skb, opt, &proto);
1931 	if (opt && opt->opt_nflen)
1932 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1933 
1934 	skb_push(skb, sizeof(struct ipv6hdr));
1935 	skb_reset_network_header(skb);
1936 	hdr = ipv6_hdr(skb);
1937 
1938 	ip6_flow_hdr(hdr, v6_cork->tclass,
1939 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1940 					ip6_autoflowlabel(net, np), fl6));
1941 	hdr->hop_limit = v6_cork->hop_limit;
1942 	hdr->nexthdr = proto;
1943 	hdr->saddr = fl6->saddr;
1944 	hdr->daddr = *final_dst;
1945 
1946 	skb->priority = sk->sk_priority;
1947 	skb->mark = cork->base.mark;
1948 	skb->tstamp = cork->base.transmit_time;
1949 
1950 	ip6_cork_steal_dst(skb, cork);
1951 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1952 	if (proto == IPPROTO_ICMPV6) {
1953 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1954 
1955 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1956 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1957 	}
1958 
1959 	ip6_cork_release(cork, v6_cork);
1960 out:
1961 	return skb;
1962 }
1963 
1964 int ip6_send_skb(struct sk_buff *skb)
1965 {
1966 	struct net *net = sock_net(skb->sk);
1967 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1968 	int err;
1969 
1970 	err = ip6_local_out(net, skb->sk, skb);
1971 	if (err) {
1972 		if (err > 0)
1973 			err = net_xmit_errno(err);
1974 		if (err)
1975 			IP6_INC_STATS(net, rt->rt6i_idev,
1976 				      IPSTATS_MIB_OUTDISCARDS);
1977 	}
1978 
1979 	return err;
1980 }
1981 
1982 int ip6_push_pending_frames(struct sock *sk)
1983 {
1984 	struct sk_buff *skb;
1985 
1986 	skb = ip6_finish_skb(sk);
1987 	if (!skb)
1988 		return 0;
1989 
1990 	return ip6_send_skb(skb);
1991 }
1992 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1993 
1994 static void __ip6_flush_pending_frames(struct sock *sk,
1995 				       struct sk_buff_head *queue,
1996 				       struct inet_cork_full *cork,
1997 				       struct inet6_cork *v6_cork)
1998 {
1999 	struct sk_buff *skb;
2000 
2001 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2002 		if (skb_dst(skb))
2003 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2004 				      IPSTATS_MIB_OUTDISCARDS);
2005 		kfree_skb(skb);
2006 	}
2007 
2008 	ip6_cork_release(cork, v6_cork);
2009 }
2010 
2011 void ip6_flush_pending_frames(struct sock *sk)
2012 {
2013 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2014 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2015 }
2016 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2017 
2018 struct sk_buff *ip6_make_skb(struct sock *sk,
2019 			     int getfrag(void *from, char *to, int offset,
2020 					 int len, int odd, struct sk_buff *skb),
2021 			     void *from, size_t length, int transhdrlen,
2022 			     struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2023 			     unsigned int flags, struct inet_cork_full *cork)
2024 {
2025 	struct inet6_cork v6_cork;
2026 	struct sk_buff_head queue;
2027 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2028 	int err;
2029 
2030 	if (flags & MSG_PROBE) {
2031 		dst_release(&rt->dst);
2032 		return NULL;
2033 	}
2034 
2035 	__skb_queue_head_init(&queue);
2036 
2037 	cork->base.flags = 0;
2038 	cork->base.addr = 0;
2039 	cork->base.opt = NULL;
2040 	v6_cork.opt = NULL;
2041 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2042 	if (err) {
2043 		ip6_cork_release(cork, &v6_cork);
2044 		return ERR_PTR(err);
2045 	}
2046 	if (ipc6->dontfrag < 0)
2047 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2048 
2049 	err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2050 				&current->task_frag, getfrag, from,
2051 				length + exthdrlen, transhdrlen + exthdrlen,
2052 				flags, ipc6);
2053 	if (err) {
2054 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2055 		return ERR_PTR(err);
2056 	}
2057 
2058 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2059 }
2060