xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 5fdaa88dfefa87ee1ea92750e99950dca182ea41)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/net/ipv4/ip_output.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  *
15  *	Changes:
16  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
17  *				extension headers are implemented.
18  *				route changes now work.
19  *				ip6_forward does not confuse sniffers.
20  *				etc.
21  *
22  *      H. von Brand    :       Added missing #include <linux/string.h>
23  *	Imran Patel	:	frag id should be in NBO
24  *      Kazunori MIYAZAWA @USAGI
25  *			:       add ip6_append_data and related functions
26  *				for datagram xmit
27  */
28 
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
41 
42 #include <linux/bpf-cgroup.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
61 
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 {
64 	struct dst_entry *dst = skb_dst(skb);
65 	struct net_device *dev = dst->dev;
66 	struct neighbour *neigh;
67 	struct in6_addr *nexthop;
68 	int ret;
69 
70 	if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
72 
73 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
74 		    ((mroute6_is_socket(net, skb) &&
75 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 		     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 					 &ipv6_hdr(skb)->saddr))) {
78 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
79 
80 			/* Do not check for IFF_ALLMULTI; multicast routing
81 			   is not supported in any case.
82 			 */
83 			if (newskb)
84 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 					net, sk, newskb, NULL, newskb->dev,
86 					dev_loopback_xmit);
87 
88 			if (ipv6_hdr(skb)->hop_limit == 0) {
89 				IP6_INC_STATS(net, idev,
90 					      IPSTATS_MIB_OUTDISCARDS);
91 				kfree_skb(skb);
92 				return 0;
93 			}
94 		}
95 
96 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
97 
98 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
99 		    IPV6_ADDR_SCOPE_NODELOCAL &&
100 		    !(dev->flags & IFF_LOOPBACK)) {
101 			kfree_skb(skb);
102 			return 0;
103 		}
104 	}
105 
106 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
107 		int res = lwtunnel_xmit(skb);
108 
109 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
110 			return res;
111 	}
112 
113 	rcu_read_lock_bh();
114 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
115 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
116 	if (unlikely(!neigh))
117 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
118 	if (!IS_ERR(neigh)) {
119 		sock_confirm_neigh(skb, neigh);
120 		ret = neigh_output(neigh, skb);
121 		rcu_read_unlock_bh();
122 		return ret;
123 	}
124 	rcu_read_unlock_bh();
125 
126 	IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
127 	kfree_skb(skb);
128 	return -EINVAL;
129 }
130 
131 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
132 {
133 	int ret;
134 
135 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
136 	if (ret) {
137 		kfree_skb(skb);
138 		return ret;
139 	}
140 
141 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
142 	/* Policy lookup after SNAT yielded a new policy */
143 	if (skb_dst(skb)->xfrm) {
144 		IPCB(skb)->flags |= IPSKB_REROUTED;
145 		return dst_output(net, sk, skb);
146 	}
147 #endif
148 
149 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
150 	    dst_allfrag(skb_dst(skb)) ||
151 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
152 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
153 	else
154 		return ip6_finish_output2(net, sk, skb);
155 }
156 
157 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
158 {
159 	struct net_device *dev = skb_dst(skb)->dev;
160 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
161 
162 	skb->protocol = htons(ETH_P_IPV6);
163 	skb->dev = dev;
164 
165 	if (unlikely(idev->cnf.disable_ipv6)) {
166 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
167 		kfree_skb(skb);
168 		return 0;
169 	}
170 
171 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
172 			    net, sk, skb, NULL, dev,
173 			    ip6_finish_output,
174 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
175 }
176 
177 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
178 {
179 	if (!np->autoflowlabel_set)
180 		return ip6_default_np_autolabel(net);
181 	else
182 		return np->autoflowlabel;
183 }
184 
185 /*
186  * xmit an sk_buff (used by TCP, SCTP and DCCP)
187  * Note : socket lock is not held for SYNACK packets, but might be modified
188  * by calls to skb_set_owner_w() and ipv6_local_error(),
189  * which are using proper atomic operations or spinlocks.
190  */
191 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
192 	     __u32 mark, struct ipv6_txoptions *opt, int tclass)
193 {
194 	struct net *net = sock_net(sk);
195 	const struct ipv6_pinfo *np = inet6_sk(sk);
196 	struct in6_addr *first_hop = &fl6->daddr;
197 	struct dst_entry *dst = skb_dst(skb);
198 	struct ipv6hdr *hdr;
199 	u8  proto = fl6->flowi6_proto;
200 	int seg_len = skb->len;
201 	int hlimit = -1;
202 	u32 mtu;
203 
204 	if (opt) {
205 		unsigned int head_room;
206 
207 		/* First: exthdrs may take lots of space (~8K for now)
208 		   MAX_HEADER is not enough.
209 		 */
210 		head_room = opt->opt_nflen + opt->opt_flen;
211 		seg_len += head_room;
212 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
213 
214 		if (skb_headroom(skb) < head_room) {
215 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
216 			if (!skb2) {
217 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 					      IPSTATS_MIB_OUTDISCARDS);
219 				kfree_skb(skb);
220 				return -ENOBUFS;
221 			}
222 			consume_skb(skb);
223 			skb = skb2;
224 			/* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
225 			 * it is safe to call in our context (socket lock not held)
226 			 */
227 			skb_set_owner_w(skb, (struct sock *)sk);
228 		}
229 		if (opt->opt_flen)
230 			ipv6_push_frag_opts(skb, opt, &proto);
231 		if (opt->opt_nflen)
232 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
233 					     &fl6->saddr);
234 	}
235 
236 	skb_push(skb, sizeof(struct ipv6hdr));
237 	skb_reset_network_header(skb);
238 	hdr = ipv6_hdr(skb);
239 
240 	/*
241 	 *	Fill in the IPv6 header
242 	 */
243 	if (np)
244 		hlimit = np->hop_limit;
245 	if (hlimit < 0)
246 		hlimit = ip6_dst_hoplimit(dst);
247 
248 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
249 				ip6_autoflowlabel(net, np), fl6));
250 
251 	hdr->payload_len = htons(seg_len);
252 	hdr->nexthdr = proto;
253 	hdr->hop_limit = hlimit;
254 
255 	hdr->saddr = fl6->saddr;
256 	hdr->daddr = *first_hop;
257 
258 	skb->protocol = htons(ETH_P_IPV6);
259 	skb->priority = sk->sk_priority;
260 	skb->mark = mark;
261 
262 	mtu = dst_mtu(dst);
263 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
264 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
265 			      IPSTATS_MIB_OUT, skb->len);
266 
267 		/* if egress device is enslaved to an L3 master device pass the
268 		 * skb to its handler for processing
269 		 */
270 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
271 		if (unlikely(!skb))
272 			return 0;
273 
274 		/* hooks should never assume socket lock is held.
275 		 * we promote our socket to non const
276 		 */
277 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
278 			       net, (struct sock *)sk, skb, NULL, dst->dev,
279 			       dst_output);
280 	}
281 
282 	skb->dev = dst->dev;
283 	/* ipv6_local_error() does not require socket lock,
284 	 * we promote our socket to non const
285 	 */
286 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
287 
288 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
289 	kfree_skb(skb);
290 	return -EMSGSIZE;
291 }
292 EXPORT_SYMBOL(ip6_xmit);
293 
294 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
295 {
296 	struct ip6_ra_chain *ra;
297 	struct sock *last = NULL;
298 
299 	read_lock(&ip6_ra_lock);
300 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
301 		struct sock *sk = ra->sk;
302 		if (sk && ra->sel == sel &&
303 		    (!sk->sk_bound_dev_if ||
304 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
305 			if (last) {
306 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
307 				if (skb2)
308 					rawv6_rcv(last, skb2);
309 			}
310 			last = sk;
311 		}
312 	}
313 
314 	if (last) {
315 		rawv6_rcv(last, skb);
316 		read_unlock(&ip6_ra_lock);
317 		return 1;
318 	}
319 	read_unlock(&ip6_ra_lock);
320 	return 0;
321 }
322 
323 static int ip6_forward_proxy_check(struct sk_buff *skb)
324 {
325 	struct ipv6hdr *hdr = ipv6_hdr(skb);
326 	u8 nexthdr = hdr->nexthdr;
327 	__be16 frag_off;
328 	int offset;
329 
330 	if (ipv6_ext_hdr(nexthdr)) {
331 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
332 		if (offset < 0)
333 			return 0;
334 	} else
335 		offset = sizeof(struct ipv6hdr);
336 
337 	if (nexthdr == IPPROTO_ICMPV6) {
338 		struct icmp6hdr *icmp6;
339 
340 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
341 					 offset + 1 - skb->data)))
342 			return 0;
343 
344 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
345 
346 		switch (icmp6->icmp6_type) {
347 		case NDISC_ROUTER_SOLICITATION:
348 		case NDISC_ROUTER_ADVERTISEMENT:
349 		case NDISC_NEIGHBOUR_SOLICITATION:
350 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
351 		case NDISC_REDIRECT:
352 			/* For reaction involving unicast neighbor discovery
353 			 * message destined to the proxied address, pass it to
354 			 * input function.
355 			 */
356 			return 1;
357 		default:
358 			break;
359 		}
360 	}
361 
362 	/*
363 	 * The proxying router can't forward traffic sent to a link-local
364 	 * address, so signal the sender and discard the packet. This
365 	 * behavior is clarified by the MIPv6 specification.
366 	 */
367 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
368 		dst_link_failure(skb);
369 		return -1;
370 	}
371 
372 	return 0;
373 }
374 
375 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
376 				     struct sk_buff *skb)
377 {
378 	struct dst_entry *dst = skb_dst(skb);
379 
380 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
381 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
382 
383 	return dst_output(net, sk, skb);
384 }
385 
386 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
387 {
388 	if (skb->len <= mtu)
389 		return false;
390 
391 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
392 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
393 		return true;
394 
395 	if (skb->ignore_df)
396 		return false;
397 
398 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
399 		return false;
400 
401 	return true;
402 }
403 
404 int ip6_forward(struct sk_buff *skb)
405 {
406 	struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
407 	struct dst_entry *dst = skb_dst(skb);
408 	struct ipv6hdr *hdr = ipv6_hdr(skb);
409 	struct inet6_skb_parm *opt = IP6CB(skb);
410 	struct net *net = dev_net(dst->dev);
411 	u32 mtu;
412 
413 	if (net->ipv6.devconf_all->forwarding == 0)
414 		goto error;
415 
416 	if (skb->pkt_type != PACKET_HOST)
417 		goto drop;
418 
419 	if (unlikely(skb->sk))
420 		goto drop;
421 
422 	if (skb_warn_if_lro(skb))
423 		goto drop;
424 
425 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
426 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
427 		goto drop;
428 	}
429 
430 	skb_forward_csum(skb);
431 
432 	/*
433 	 *	We DO NOT make any processing on
434 	 *	RA packets, pushing them to user level AS IS
435 	 *	without ane WARRANTY that application will be able
436 	 *	to interpret them. The reason is that we
437 	 *	cannot make anything clever here.
438 	 *
439 	 *	We are not end-node, so that if packet contains
440 	 *	AH/ESP, we cannot make anything.
441 	 *	Defragmentation also would be mistake, RA packets
442 	 *	cannot be fragmented, because there is no warranty
443 	 *	that different fragments will go along one path. --ANK
444 	 */
445 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
446 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
447 			return 0;
448 	}
449 
450 	/*
451 	 *	check and decrement ttl
452 	 */
453 	if (hdr->hop_limit <= 1) {
454 		/* Force OUTPUT device used as source address */
455 		skb->dev = dst->dev;
456 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
457 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
458 
459 		kfree_skb(skb);
460 		return -ETIMEDOUT;
461 	}
462 
463 	/* XXX: idev->cnf.proxy_ndp? */
464 	if (net->ipv6.devconf_all->proxy_ndp &&
465 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
466 		int proxied = ip6_forward_proxy_check(skb);
467 		if (proxied > 0)
468 			return ip6_input(skb);
469 		else if (proxied < 0) {
470 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
471 			goto drop;
472 		}
473 	}
474 
475 	if (!xfrm6_route_forward(skb)) {
476 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
477 		goto drop;
478 	}
479 	dst = skb_dst(skb);
480 
481 	/* IPv6 specs say nothing about it, but it is clear that we cannot
482 	   send redirects to source routed frames.
483 	   We don't send redirects to frames decapsulated from IPsec.
484 	 */
485 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
486 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
487 		struct in6_addr *target = NULL;
488 		struct inet_peer *peer;
489 		struct rt6_info *rt;
490 
491 		/*
492 		 *	incoming and outgoing devices are the same
493 		 *	send a redirect.
494 		 */
495 
496 		rt = (struct rt6_info *) dst;
497 		if (rt->rt6i_flags & RTF_GATEWAY)
498 			target = &rt->rt6i_gateway;
499 		else
500 			target = &hdr->daddr;
501 
502 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
503 
504 		/* Limit redirects both by destination (here)
505 		   and by source (inside ndisc_send_redirect)
506 		 */
507 		if (inet_peer_xrlim_allow(peer, 1*HZ))
508 			ndisc_send_redirect(skb, target);
509 		if (peer)
510 			inet_putpeer(peer);
511 	} else {
512 		int addrtype = ipv6_addr_type(&hdr->saddr);
513 
514 		/* This check is security critical. */
515 		if (addrtype == IPV6_ADDR_ANY ||
516 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
517 			goto error;
518 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
519 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
520 				    ICMPV6_NOT_NEIGHBOUR, 0);
521 			goto error;
522 		}
523 	}
524 
525 	mtu = ip6_dst_mtu_forward(dst);
526 	if (mtu < IPV6_MIN_MTU)
527 		mtu = IPV6_MIN_MTU;
528 
529 	if (ip6_pkt_too_big(skb, mtu)) {
530 		/* Again, force OUTPUT device used as source address */
531 		skb->dev = dst->dev;
532 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
533 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
534 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
535 				IPSTATS_MIB_FRAGFAILS);
536 		kfree_skb(skb);
537 		return -EMSGSIZE;
538 	}
539 
540 	if (skb_cow(skb, dst->dev->hard_header_len)) {
541 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
542 				IPSTATS_MIB_OUTDISCARDS);
543 		goto drop;
544 	}
545 
546 	hdr = ipv6_hdr(skb);
547 
548 	/* Mangling hops number delayed to point after skb COW */
549 
550 	hdr->hop_limit--;
551 
552 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
553 		       net, NULL, skb, skb->dev, dst->dev,
554 		       ip6_forward_finish);
555 
556 error:
557 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
558 drop:
559 	kfree_skb(skb);
560 	return -EINVAL;
561 }
562 
563 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
564 {
565 	to->pkt_type = from->pkt_type;
566 	to->priority = from->priority;
567 	to->protocol = from->protocol;
568 	skb_dst_drop(to);
569 	skb_dst_set(to, dst_clone(skb_dst(from)));
570 	to->dev = from->dev;
571 	to->mark = from->mark;
572 
573 #ifdef CONFIG_NET_SCHED
574 	to->tc_index = from->tc_index;
575 #endif
576 	nf_copy(to, from);
577 	skb_copy_secmark(to, from);
578 }
579 
580 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
581 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
582 {
583 	struct sk_buff *frag;
584 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
585 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
586 				inet6_sk(skb->sk) : NULL;
587 	struct ipv6hdr *tmp_hdr;
588 	struct frag_hdr *fh;
589 	unsigned int mtu, hlen, left, len;
590 	int hroom, troom;
591 	__be32 frag_id;
592 	int ptr, offset = 0, err = 0;
593 	u8 *prevhdr, nexthdr = 0;
594 
595 	err = ip6_find_1stfragopt(skb, &prevhdr);
596 	if (err < 0)
597 		goto fail;
598 	hlen = err;
599 	nexthdr = *prevhdr;
600 
601 	mtu = ip6_skb_dst_mtu(skb);
602 
603 	/* We must not fragment if the socket is set to force MTU discovery
604 	 * or if the skb it not generated by a local socket.
605 	 */
606 	if (unlikely(!skb->ignore_df && skb->len > mtu))
607 		goto fail_toobig;
608 
609 	if (IP6CB(skb)->frag_max_size) {
610 		if (IP6CB(skb)->frag_max_size > mtu)
611 			goto fail_toobig;
612 
613 		/* don't send fragments larger than what we received */
614 		mtu = IP6CB(skb)->frag_max_size;
615 		if (mtu < IPV6_MIN_MTU)
616 			mtu = IPV6_MIN_MTU;
617 	}
618 
619 	if (np && np->frag_size < mtu) {
620 		if (np->frag_size)
621 			mtu = np->frag_size;
622 	}
623 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
624 		goto fail_toobig;
625 	mtu -= hlen + sizeof(struct frag_hdr);
626 
627 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
628 				    &ipv6_hdr(skb)->saddr);
629 
630 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
631 	    (err = skb_checksum_help(skb)))
632 		goto fail;
633 
634 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
635 	if (skb_has_frag_list(skb)) {
636 		unsigned int first_len = skb_pagelen(skb);
637 		struct sk_buff *frag2;
638 
639 		if (first_len - hlen > mtu ||
640 		    ((first_len - hlen) & 7) ||
641 		    skb_cloned(skb) ||
642 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
643 			goto slow_path;
644 
645 		skb_walk_frags(skb, frag) {
646 			/* Correct geometry. */
647 			if (frag->len > mtu ||
648 			    ((frag->len & 7) && frag->next) ||
649 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
650 				goto slow_path_clean;
651 
652 			/* Partially cloned skb? */
653 			if (skb_shared(frag))
654 				goto slow_path_clean;
655 
656 			BUG_ON(frag->sk);
657 			if (skb->sk) {
658 				frag->sk = skb->sk;
659 				frag->destructor = sock_wfree;
660 			}
661 			skb->truesize -= frag->truesize;
662 		}
663 
664 		err = 0;
665 		offset = 0;
666 		/* BUILD HEADER */
667 
668 		*prevhdr = NEXTHDR_FRAGMENT;
669 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
670 		if (!tmp_hdr) {
671 			err = -ENOMEM;
672 			goto fail;
673 		}
674 		frag = skb_shinfo(skb)->frag_list;
675 		skb_frag_list_init(skb);
676 
677 		__skb_pull(skb, hlen);
678 		fh = __skb_push(skb, sizeof(struct frag_hdr));
679 		__skb_push(skb, hlen);
680 		skb_reset_network_header(skb);
681 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
682 
683 		fh->nexthdr = nexthdr;
684 		fh->reserved = 0;
685 		fh->frag_off = htons(IP6_MF);
686 		fh->identification = frag_id;
687 
688 		first_len = skb_pagelen(skb);
689 		skb->data_len = first_len - skb_headlen(skb);
690 		skb->len = first_len;
691 		ipv6_hdr(skb)->payload_len = htons(first_len -
692 						   sizeof(struct ipv6hdr));
693 
694 		for (;;) {
695 			/* Prepare header of the next frame,
696 			 * before previous one went down. */
697 			if (frag) {
698 				frag->ip_summed = CHECKSUM_NONE;
699 				skb_reset_transport_header(frag);
700 				fh = __skb_push(frag, sizeof(struct frag_hdr));
701 				__skb_push(frag, hlen);
702 				skb_reset_network_header(frag);
703 				memcpy(skb_network_header(frag), tmp_hdr,
704 				       hlen);
705 				offset += skb->len - hlen - sizeof(struct frag_hdr);
706 				fh->nexthdr = nexthdr;
707 				fh->reserved = 0;
708 				fh->frag_off = htons(offset);
709 				if (frag->next)
710 					fh->frag_off |= htons(IP6_MF);
711 				fh->identification = frag_id;
712 				ipv6_hdr(frag)->payload_len =
713 						htons(frag->len -
714 						      sizeof(struct ipv6hdr));
715 				ip6_copy_metadata(frag, skb);
716 			}
717 
718 			err = output(net, sk, skb);
719 			if (!err)
720 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
721 					      IPSTATS_MIB_FRAGCREATES);
722 
723 			if (err || !frag)
724 				break;
725 
726 			skb = frag;
727 			frag = skb->next;
728 			skb->next = NULL;
729 		}
730 
731 		kfree(tmp_hdr);
732 
733 		if (err == 0) {
734 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
735 				      IPSTATS_MIB_FRAGOKS);
736 			return 0;
737 		}
738 
739 		kfree_skb_list(frag);
740 
741 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
742 			      IPSTATS_MIB_FRAGFAILS);
743 		return err;
744 
745 slow_path_clean:
746 		skb_walk_frags(skb, frag2) {
747 			if (frag2 == frag)
748 				break;
749 			frag2->sk = NULL;
750 			frag2->destructor = NULL;
751 			skb->truesize += frag2->truesize;
752 		}
753 	}
754 
755 slow_path:
756 	left = skb->len - hlen;		/* Space per frame */
757 	ptr = hlen;			/* Where to start from */
758 
759 	/*
760 	 *	Fragment the datagram.
761 	 */
762 
763 	troom = rt->dst.dev->needed_tailroom;
764 
765 	/*
766 	 *	Keep copying data until we run out.
767 	 */
768 	while (left > 0)	{
769 		u8 *fragnexthdr_offset;
770 
771 		len = left;
772 		/* IF: it doesn't fit, use 'mtu' - the data space left */
773 		if (len > mtu)
774 			len = mtu;
775 		/* IF: we are not sending up to and including the packet end
776 		   then align the next start on an eight byte boundary */
777 		if (len < left)	{
778 			len &= ~7;
779 		}
780 
781 		/* Allocate buffer */
782 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
783 				 hroom + troom, GFP_ATOMIC);
784 		if (!frag) {
785 			err = -ENOMEM;
786 			goto fail;
787 		}
788 
789 		/*
790 		 *	Set up data on packet
791 		 */
792 
793 		ip6_copy_metadata(frag, skb);
794 		skb_reserve(frag, hroom);
795 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
796 		skb_reset_network_header(frag);
797 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
798 		frag->transport_header = (frag->network_header + hlen +
799 					  sizeof(struct frag_hdr));
800 
801 		/*
802 		 *	Charge the memory for the fragment to any owner
803 		 *	it might possess
804 		 */
805 		if (skb->sk)
806 			skb_set_owner_w(frag, skb->sk);
807 
808 		/*
809 		 *	Copy the packet header into the new buffer.
810 		 */
811 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
812 
813 		fragnexthdr_offset = skb_network_header(frag);
814 		fragnexthdr_offset += prevhdr - skb_network_header(skb);
815 		*fragnexthdr_offset = NEXTHDR_FRAGMENT;
816 
817 		/*
818 		 *	Build fragment header.
819 		 */
820 		fh->nexthdr = nexthdr;
821 		fh->reserved = 0;
822 		fh->identification = frag_id;
823 
824 		/*
825 		 *	Copy a block of the IP datagram.
826 		 */
827 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
828 				     len));
829 		left -= len;
830 
831 		fh->frag_off = htons(offset);
832 		if (left > 0)
833 			fh->frag_off |= htons(IP6_MF);
834 		ipv6_hdr(frag)->payload_len = htons(frag->len -
835 						    sizeof(struct ipv6hdr));
836 
837 		ptr += len;
838 		offset += len;
839 
840 		/*
841 		 *	Put this fragment into the sending queue.
842 		 */
843 		err = output(net, sk, frag);
844 		if (err)
845 			goto fail;
846 
847 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
848 			      IPSTATS_MIB_FRAGCREATES);
849 	}
850 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
851 		      IPSTATS_MIB_FRAGOKS);
852 	consume_skb(skb);
853 	return err;
854 
855 fail_toobig:
856 	if (skb->sk && dst_allfrag(skb_dst(skb)))
857 		sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
858 
859 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
860 	err = -EMSGSIZE;
861 
862 fail:
863 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
864 		      IPSTATS_MIB_FRAGFAILS);
865 	kfree_skb(skb);
866 	return err;
867 }
868 
869 static inline int ip6_rt_check(const struct rt6key *rt_key,
870 			       const struct in6_addr *fl_addr,
871 			       const struct in6_addr *addr_cache)
872 {
873 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
874 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
875 }
876 
877 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
878 					  struct dst_entry *dst,
879 					  const struct flowi6 *fl6)
880 {
881 	struct ipv6_pinfo *np = inet6_sk(sk);
882 	struct rt6_info *rt;
883 
884 	if (!dst)
885 		goto out;
886 
887 	if (dst->ops->family != AF_INET6) {
888 		dst_release(dst);
889 		return NULL;
890 	}
891 
892 	rt = (struct rt6_info *)dst;
893 	/* Yes, checking route validity in not connected
894 	 * case is not very simple. Take into account,
895 	 * that we do not support routing by source, TOS,
896 	 * and MSG_DONTROUTE		--ANK (980726)
897 	 *
898 	 * 1. ip6_rt_check(): If route was host route,
899 	 *    check that cached destination is current.
900 	 *    If it is network route, we still may
901 	 *    check its validity using saved pointer
902 	 *    to the last used address: daddr_cache.
903 	 *    We do not want to save whole address now,
904 	 *    (because main consumer of this service
905 	 *    is tcp, which has not this problem),
906 	 *    so that the last trick works only on connected
907 	 *    sockets.
908 	 * 2. oif also should be the same.
909 	 */
910 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
911 #ifdef CONFIG_IPV6_SUBTREES
912 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
913 #endif
914 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
915 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
916 		dst_release(dst);
917 		dst = NULL;
918 	}
919 
920 out:
921 	return dst;
922 }
923 
924 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
925 			       struct dst_entry **dst, struct flowi6 *fl6)
926 {
927 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
928 	struct neighbour *n;
929 	struct rt6_info *rt;
930 #endif
931 	int err;
932 	int flags = 0;
933 
934 	/* The correct way to handle this would be to do
935 	 * ip6_route_get_saddr, and then ip6_route_output; however,
936 	 * the route-specific preferred source forces the
937 	 * ip6_route_output call _before_ ip6_route_get_saddr.
938 	 *
939 	 * In source specific routing (no src=any default route),
940 	 * ip6_route_output will fail given src=any saddr, though, so
941 	 * that's why we try it again later.
942 	 */
943 	if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
944 		struct fib6_info *from;
945 		struct rt6_info *rt;
946 		bool had_dst = *dst != NULL;
947 
948 		if (!had_dst)
949 			*dst = ip6_route_output(net, sk, fl6);
950 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
951 
952 		rcu_read_lock();
953 		from = rt ? rcu_dereference(rt->from) : NULL;
954 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
955 					  sk ? inet6_sk(sk)->srcprefs : 0,
956 					  &fl6->saddr);
957 		rcu_read_unlock();
958 
959 		if (err)
960 			goto out_err_release;
961 
962 		/* If we had an erroneous initial result, pretend it
963 		 * never existed and let the SA-enabled version take
964 		 * over.
965 		 */
966 		if (!had_dst && (*dst)->error) {
967 			dst_release(*dst);
968 			*dst = NULL;
969 		}
970 
971 		if (fl6->flowi6_oif)
972 			flags |= RT6_LOOKUP_F_IFACE;
973 	}
974 
975 	if (!*dst)
976 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
977 
978 	err = (*dst)->error;
979 	if (err)
980 		goto out_err_release;
981 
982 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
983 	/*
984 	 * Here if the dst entry we've looked up
985 	 * has a neighbour entry that is in the INCOMPLETE
986 	 * state and the src address from the flow is
987 	 * marked as OPTIMISTIC, we release the found
988 	 * dst entry and replace it instead with the
989 	 * dst entry of the nexthop router
990 	 */
991 	rt = (struct rt6_info *) *dst;
992 	rcu_read_lock_bh();
993 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
994 				      rt6_nexthop(rt, &fl6->daddr));
995 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
996 	rcu_read_unlock_bh();
997 
998 	if (err) {
999 		struct inet6_ifaddr *ifp;
1000 		struct flowi6 fl_gw6;
1001 		int redirect;
1002 
1003 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1004 				      (*dst)->dev, 1);
1005 
1006 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1007 		if (ifp)
1008 			in6_ifa_put(ifp);
1009 
1010 		if (redirect) {
1011 			/*
1012 			 * We need to get the dst entry for the
1013 			 * default router instead
1014 			 */
1015 			dst_release(*dst);
1016 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1017 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1018 			*dst = ip6_route_output(net, sk, &fl_gw6);
1019 			err = (*dst)->error;
1020 			if (err)
1021 				goto out_err_release;
1022 		}
1023 	}
1024 #endif
1025 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1026 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1027 		err = -EAFNOSUPPORT;
1028 		goto out_err_release;
1029 	}
1030 
1031 	return 0;
1032 
1033 out_err_release:
1034 	dst_release(*dst);
1035 	*dst = NULL;
1036 
1037 	if (err == -ENETUNREACH)
1038 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1039 	return err;
1040 }
1041 
1042 /**
1043  *	ip6_dst_lookup - perform route lookup on flow
1044  *	@sk: socket which provides route info
1045  *	@dst: pointer to dst_entry * for result
1046  *	@fl6: flow to lookup
1047  *
1048  *	This function performs a route lookup on the given flow.
1049  *
1050  *	It returns zero on success, or a standard errno code on error.
1051  */
1052 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1053 		   struct flowi6 *fl6)
1054 {
1055 	*dst = NULL;
1056 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1057 }
1058 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1059 
1060 /**
1061  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1062  *	@sk: socket which provides route info
1063  *	@fl6: flow to lookup
1064  *	@final_dst: final destination address for ipsec lookup
1065  *
1066  *	This function performs a route lookup on the given flow.
1067  *
1068  *	It returns a valid dst pointer on success, or a pointer encoded
1069  *	error code.
1070  */
1071 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1072 				      const struct in6_addr *final_dst)
1073 {
1074 	struct dst_entry *dst = NULL;
1075 	int err;
1076 
1077 	err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1078 	if (err)
1079 		return ERR_PTR(err);
1080 	if (final_dst)
1081 		fl6->daddr = *final_dst;
1082 
1083 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1084 }
1085 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1086 
1087 /**
1088  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1089  *	@sk: socket which provides the dst cache and route info
1090  *	@fl6: flow to lookup
1091  *	@final_dst: final destination address for ipsec lookup
1092  *	@connected: whether @sk is connected or not
1093  *
1094  *	This function performs a route lookup on the given flow with the
1095  *	possibility of using the cached route in the socket if it is valid.
1096  *	It will take the socket dst lock when operating on the dst cache.
1097  *	As a result, this function can only be used in process context.
1098  *
1099  *	In addition, for a connected socket, cache the dst in the socket
1100  *	if the current cache is not valid.
1101  *
1102  *	It returns a valid dst pointer on success, or a pointer encoded
1103  *	error code.
1104  */
1105 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1106 					 const struct in6_addr *final_dst,
1107 					 bool connected)
1108 {
1109 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1110 
1111 	dst = ip6_sk_dst_check(sk, dst, fl6);
1112 	if (dst)
1113 		return dst;
1114 
1115 	dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1116 	if (connected && !IS_ERR(dst))
1117 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1118 
1119 	return dst;
1120 }
1121 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1122 
1123 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1124 					       gfp_t gfp)
1125 {
1126 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1127 }
1128 
1129 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1130 						gfp_t gfp)
1131 {
1132 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1133 }
1134 
1135 static void ip6_append_data_mtu(unsigned int *mtu,
1136 				int *maxfraglen,
1137 				unsigned int fragheaderlen,
1138 				struct sk_buff *skb,
1139 				struct rt6_info *rt,
1140 				unsigned int orig_mtu)
1141 {
1142 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1143 		if (!skb) {
1144 			/* first fragment, reserve header_len */
1145 			*mtu = orig_mtu - rt->dst.header_len;
1146 
1147 		} else {
1148 			/*
1149 			 * this fragment is not first, the headers
1150 			 * space is regarded as data space.
1151 			 */
1152 			*mtu = orig_mtu;
1153 		}
1154 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1155 			      + fragheaderlen - sizeof(struct frag_hdr);
1156 	}
1157 }
1158 
1159 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1160 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1161 			  struct rt6_info *rt, struct flowi6 *fl6)
1162 {
1163 	struct ipv6_pinfo *np = inet6_sk(sk);
1164 	unsigned int mtu;
1165 	struct ipv6_txoptions *opt = ipc6->opt;
1166 
1167 	/*
1168 	 * setup for corking
1169 	 */
1170 	if (opt) {
1171 		if (WARN_ON(v6_cork->opt))
1172 			return -EINVAL;
1173 
1174 		v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1175 		if (unlikely(!v6_cork->opt))
1176 			return -ENOBUFS;
1177 
1178 		v6_cork->opt->tot_len = sizeof(*opt);
1179 		v6_cork->opt->opt_flen = opt->opt_flen;
1180 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1181 
1182 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1183 						    sk->sk_allocation);
1184 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1185 			return -ENOBUFS;
1186 
1187 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1188 						    sk->sk_allocation);
1189 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1190 			return -ENOBUFS;
1191 
1192 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1193 						   sk->sk_allocation);
1194 		if (opt->hopopt && !v6_cork->opt->hopopt)
1195 			return -ENOBUFS;
1196 
1197 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1198 						    sk->sk_allocation);
1199 		if (opt->srcrt && !v6_cork->opt->srcrt)
1200 			return -ENOBUFS;
1201 
1202 		/* need source address above miyazawa*/
1203 	}
1204 	dst_hold(&rt->dst);
1205 	cork->base.dst = &rt->dst;
1206 	cork->fl.u.ip6 = *fl6;
1207 	v6_cork->hop_limit = ipc6->hlimit;
1208 	v6_cork->tclass = ipc6->tclass;
1209 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1210 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1211 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1212 	else
1213 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1214 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1215 	if (np->frag_size < mtu) {
1216 		if (np->frag_size)
1217 			mtu = np->frag_size;
1218 	}
1219 	if (mtu < IPV6_MIN_MTU)
1220 		return -EINVAL;
1221 	cork->base.fragsize = mtu;
1222 	cork->base.gso_size = sk->sk_type == SOCK_DGRAM &&
1223 			      sk->sk_protocol == IPPROTO_UDP ? ipc6->gso_size : 0;
1224 
1225 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1226 		cork->base.flags |= IPCORK_ALLFRAG;
1227 	cork->base.length = 0;
1228 
1229 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1230 
1231 	return 0;
1232 }
1233 
1234 static int __ip6_append_data(struct sock *sk,
1235 			     struct flowi6 *fl6,
1236 			     struct sk_buff_head *queue,
1237 			     struct inet_cork *cork,
1238 			     struct inet6_cork *v6_cork,
1239 			     struct page_frag *pfrag,
1240 			     int getfrag(void *from, char *to, int offset,
1241 					 int len, int odd, struct sk_buff *skb),
1242 			     void *from, int length, int transhdrlen,
1243 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1244 {
1245 	struct sk_buff *skb, *skb_prev = NULL;
1246 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1247 	int exthdrlen = 0;
1248 	int dst_exthdrlen = 0;
1249 	int hh_len;
1250 	int copy;
1251 	int err;
1252 	int offset = 0;
1253 	__u8 tx_flags = 0;
1254 	u32 tskey = 0;
1255 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1256 	struct ipv6_txoptions *opt = v6_cork->opt;
1257 	int csummode = CHECKSUM_NONE;
1258 	unsigned int maxnonfragsize, headersize;
1259 	unsigned int wmem_alloc_delta = 0;
1260 	bool paged;
1261 
1262 	skb = skb_peek_tail(queue);
1263 	if (!skb) {
1264 		exthdrlen = opt ? opt->opt_flen : 0;
1265 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1266 	}
1267 
1268 	paged = !!cork->gso_size;
1269 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1270 	orig_mtu = mtu;
1271 
1272 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1273 
1274 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1275 			(opt ? opt->opt_nflen : 0);
1276 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1277 		     sizeof(struct frag_hdr);
1278 
1279 	headersize = sizeof(struct ipv6hdr) +
1280 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1281 		     (dst_allfrag(&rt->dst) ?
1282 		      sizeof(struct frag_hdr) : 0) +
1283 		     rt->rt6i_nfheader_len;
1284 
1285 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1286 	 * the first fragment
1287 	 */
1288 	if (headersize + transhdrlen > mtu)
1289 		goto emsgsize;
1290 
1291 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1292 	    (sk->sk_protocol == IPPROTO_UDP ||
1293 	     sk->sk_protocol == IPPROTO_RAW)) {
1294 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1295 				sizeof(struct ipv6hdr));
1296 		goto emsgsize;
1297 	}
1298 
1299 	if (ip6_sk_ignore_df(sk))
1300 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1301 	else
1302 		maxnonfragsize = mtu;
1303 
1304 	if (cork->length + length > maxnonfragsize - headersize) {
1305 emsgsize:
1306 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1307 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1308 		return -EMSGSIZE;
1309 	}
1310 
1311 	/* CHECKSUM_PARTIAL only with no extension headers and when
1312 	 * we are not going to fragment
1313 	 */
1314 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1315 	    headersize == sizeof(struct ipv6hdr) &&
1316 	    length <= mtu - headersize &&
1317 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1318 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1319 		csummode = CHECKSUM_PARTIAL;
1320 
1321 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1322 		sock_tx_timestamp(sk, ipc6->sockc.tsflags, &tx_flags);
1323 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1324 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1325 			tskey = sk->sk_tskey++;
1326 	}
1327 
1328 	/*
1329 	 * Let's try using as much space as possible.
1330 	 * Use MTU if total length of the message fits into the MTU.
1331 	 * Otherwise, we need to reserve fragment header and
1332 	 * fragment alignment (= 8-15 octects, in total).
1333 	 *
1334 	 * Note that we may need to "move" the data from the tail of
1335 	 * of the buffer to the new fragment when we split
1336 	 * the message.
1337 	 *
1338 	 * FIXME: It may be fragmented into multiple chunks
1339 	 *        at once if non-fragmentable extension headers
1340 	 *        are too large.
1341 	 * --yoshfuji
1342 	 */
1343 
1344 	cork->length += length;
1345 	if (!skb)
1346 		goto alloc_new_skb;
1347 
1348 	while (length > 0) {
1349 		/* Check if the remaining data fits into current packet. */
1350 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1351 		if (copy < length)
1352 			copy = maxfraglen - skb->len;
1353 
1354 		if (copy <= 0) {
1355 			char *data;
1356 			unsigned int datalen;
1357 			unsigned int fraglen;
1358 			unsigned int fraggap;
1359 			unsigned int alloclen;
1360 			unsigned int pagedlen = 0;
1361 alloc_new_skb:
1362 			/* There's no room in the current skb */
1363 			if (skb)
1364 				fraggap = skb->len - maxfraglen;
1365 			else
1366 				fraggap = 0;
1367 			/* update mtu and maxfraglen if necessary */
1368 			if (!skb || !skb_prev)
1369 				ip6_append_data_mtu(&mtu, &maxfraglen,
1370 						    fragheaderlen, skb, rt,
1371 						    orig_mtu);
1372 
1373 			skb_prev = skb;
1374 
1375 			/*
1376 			 * If remaining data exceeds the mtu,
1377 			 * we know we need more fragment(s).
1378 			 */
1379 			datalen = length + fraggap;
1380 
1381 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1382 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1383 			fraglen = datalen + fragheaderlen;
1384 
1385 			if ((flags & MSG_MORE) &&
1386 			    !(rt->dst.dev->features&NETIF_F_SG))
1387 				alloclen = mtu;
1388 			else if (!paged)
1389 				alloclen = fraglen;
1390 			else {
1391 				alloclen = min_t(int, fraglen, MAX_HEADER);
1392 				pagedlen = fraglen - alloclen;
1393 			}
1394 
1395 			alloclen += dst_exthdrlen;
1396 
1397 			if (datalen != length + fraggap) {
1398 				/*
1399 				 * this is not the last fragment, the trailer
1400 				 * space is regarded as data space.
1401 				 */
1402 				datalen += rt->dst.trailer_len;
1403 			}
1404 
1405 			alloclen += rt->dst.trailer_len;
1406 			fraglen = datalen + fragheaderlen;
1407 
1408 			/*
1409 			 * We just reserve space for fragment header.
1410 			 * Note: this may be overallocation if the message
1411 			 * (without MSG_MORE) fits into the MTU.
1412 			 */
1413 			alloclen += sizeof(struct frag_hdr);
1414 
1415 			copy = datalen - transhdrlen - fraggap - pagedlen;
1416 			if (copy < 0) {
1417 				err = -EINVAL;
1418 				goto error;
1419 			}
1420 			if (transhdrlen) {
1421 				skb = sock_alloc_send_skb(sk,
1422 						alloclen + hh_len,
1423 						(flags & MSG_DONTWAIT), &err);
1424 			} else {
1425 				skb = NULL;
1426 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1427 				    2 * sk->sk_sndbuf)
1428 					skb = alloc_skb(alloclen + hh_len,
1429 							sk->sk_allocation);
1430 				if (unlikely(!skb))
1431 					err = -ENOBUFS;
1432 			}
1433 			if (!skb)
1434 				goto error;
1435 			/*
1436 			 *	Fill in the control structures
1437 			 */
1438 			skb->protocol = htons(ETH_P_IPV6);
1439 			skb->ip_summed = csummode;
1440 			skb->csum = 0;
1441 			/* reserve for fragmentation and ipsec header */
1442 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1443 				    dst_exthdrlen);
1444 
1445 			/* Only the initial fragment is time stamped */
1446 			skb_shinfo(skb)->tx_flags = tx_flags;
1447 			tx_flags = 0;
1448 			skb_shinfo(skb)->tskey = tskey;
1449 			tskey = 0;
1450 
1451 			/*
1452 			 *	Find where to start putting bytes
1453 			 */
1454 			data = skb_put(skb, fraglen - pagedlen);
1455 			skb_set_network_header(skb, exthdrlen);
1456 			data += fragheaderlen;
1457 			skb->transport_header = (skb->network_header +
1458 						 fragheaderlen);
1459 			if (fraggap) {
1460 				skb->csum = skb_copy_and_csum_bits(
1461 					skb_prev, maxfraglen,
1462 					data + transhdrlen, fraggap, 0);
1463 				skb_prev->csum = csum_sub(skb_prev->csum,
1464 							  skb->csum);
1465 				data += fraggap;
1466 				pskb_trim_unique(skb_prev, maxfraglen);
1467 			}
1468 			if (copy > 0 &&
1469 			    getfrag(from, data + transhdrlen, offset,
1470 				    copy, fraggap, skb) < 0) {
1471 				err = -EFAULT;
1472 				kfree_skb(skb);
1473 				goto error;
1474 			}
1475 
1476 			offset += copy;
1477 			length -= copy + transhdrlen;
1478 			transhdrlen = 0;
1479 			exthdrlen = 0;
1480 			dst_exthdrlen = 0;
1481 
1482 			if ((flags & MSG_CONFIRM) && !skb_prev)
1483 				skb_set_dst_pending_confirm(skb, 1);
1484 
1485 			/*
1486 			 * Put the packet on the pending queue
1487 			 */
1488 			if (!skb->destructor) {
1489 				skb->destructor = sock_wfree;
1490 				skb->sk = sk;
1491 				wmem_alloc_delta += skb->truesize;
1492 			}
1493 			__skb_queue_tail(queue, skb);
1494 			continue;
1495 		}
1496 
1497 		if (copy > length)
1498 			copy = length;
1499 
1500 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1501 		    skb_tailroom(skb) >= copy) {
1502 			unsigned int off;
1503 
1504 			off = skb->len;
1505 			if (getfrag(from, skb_put(skb, copy),
1506 						offset, copy, off, skb) < 0) {
1507 				__skb_trim(skb, off);
1508 				err = -EFAULT;
1509 				goto error;
1510 			}
1511 		} else {
1512 			int i = skb_shinfo(skb)->nr_frags;
1513 
1514 			err = -ENOMEM;
1515 			if (!sk_page_frag_refill(sk, pfrag))
1516 				goto error;
1517 
1518 			if (!skb_can_coalesce(skb, i, pfrag->page,
1519 					      pfrag->offset)) {
1520 				err = -EMSGSIZE;
1521 				if (i == MAX_SKB_FRAGS)
1522 					goto error;
1523 
1524 				__skb_fill_page_desc(skb, i, pfrag->page,
1525 						     pfrag->offset, 0);
1526 				skb_shinfo(skb)->nr_frags = ++i;
1527 				get_page(pfrag->page);
1528 			}
1529 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1530 			if (getfrag(from,
1531 				    page_address(pfrag->page) + pfrag->offset,
1532 				    offset, copy, skb->len, skb) < 0)
1533 				goto error_efault;
1534 
1535 			pfrag->offset += copy;
1536 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1537 			skb->len += copy;
1538 			skb->data_len += copy;
1539 			skb->truesize += copy;
1540 			wmem_alloc_delta += copy;
1541 		}
1542 		offset += copy;
1543 		length -= copy;
1544 	}
1545 
1546 	if (wmem_alloc_delta)
1547 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1548 	return 0;
1549 
1550 error_efault:
1551 	err = -EFAULT;
1552 error:
1553 	cork->length -= length;
1554 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1555 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1556 	return err;
1557 }
1558 
1559 int ip6_append_data(struct sock *sk,
1560 		    int getfrag(void *from, char *to, int offset, int len,
1561 				int odd, struct sk_buff *skb),
1562 		    void *from, int length, int transhdrlen,
1563 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1564 		    struct rt6_info *rt, unsigned int flags)
1565 {
1566 	struct inet_sock *inet = inet_sk(sk);
1567 	struct ipv6_pinfo *np = inet6_sk(sk);
1568 	int exthdrlen;
1569 	int err;
1570 
1571 	if (flags&MSG_PROBE)
1572 		return 0;
1573 	if (skb_queue_empty(&sk->sk_write_queue)) {
1574 		/*
1575 		 * setup for corking
1576 		 */
1577 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1578 				     ipc6, rt, fl6);
1579 		if (err)
1580 			return err;
1581 
1582 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1583 		length += exthdrlen;
1584 		transhdrlen += exthdrlen;
1585 	} else {
1586 		fl6 = &inet->cork.fl.u.ip6;
1587 		transhdrlen = 0;
1588 	}
1589 
1590 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1591 				 &np->cork, sk_page_frag(sk), getfrag,
1592 				 from, length, transhdrlen, flags, ipc6);
1593 }
1594 EXPORT_SYMBOL_GPL(ip6_append_data);
1595 
1596 static void ip6_cork_release(struct inet_cork_full *cork,
1597 			     struct inet6_cork *v6_cork)
1598 {
1599 	if (v6_cork->opt) {
1600 		kfree(v6_cork->opt->dst0opt);
1601 		kfree(v6_cork->opt->dst1opt);
1602 		kfree(v6_cork->opt->hopopt);
1603 		kfree(v6_cork->opt->srcrt);
1604 		kfree(v6_cork->opt);
1605 		v6_cork->opt = NULL;
1606 	}
1607 
1608 	if (cork->base.dst) {
1609 		dst_release(cork->base.dst);
1610 		cork->base.dst = NULL;
1611 		cork->base.flags &= ~IPCORK_ALLFRAG;
1612 	}
1613 	memset(&cork->fl, 0, sizeof(cork->fl));
1614 }
1615 
1616 struct sk_buff *__ip6_make_skb(struct sock *sk,
1617 			       struct sk_buff_head *queue,
1618 			       struct inet_cork_full *cork,
1619 			       struct inet6_cork *v6_cork)
1620 {
1621 	struct sk_buff *skb, *tmp_skb;
1622 	struct sk_buff **tail_skb;
1623 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1624 	struct ipv6_pinfo *np = inet6_sk(sk);
1625 	struct net *net = sock_net(sk);
1626 	struct ipv6hdr *hdr;
1627 	struct ipv6_txoptions *opt = v6_cork->opt;
1628 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1629 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1630 	unsigned char proto = fl6->flowi6_proto;
1631 
1632 	skb = __skb_dequeue(queue);
1633 	if (!skb)
1634 		goto out;
1635 	tail_skb = &(skb_shinfo(skb)->frag_list);
1636 
1637 	/* move skb->data to ip header from ext header */
1638 	if (skb->data < skb_network_header(skb))
1639 		__skb_pull(skb, skb_network_offset(skb));
1640 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1641 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1642 		*tail_skb = tmp_skb;
1643 		tail_skb = &(tmp_skb->next);
1644 		skb->len += tmp_skb->len;
1645 		skb->data_len += tmp_skb->len;
1646 		skb->truesize += tmp_skb->truesize;
1647 		tmp_skb->destructor = NULL;
1648 		tmp_skb->sk = NULL;
1649 	}
1650 
1651 	/* Allow local fragmentation. */
1652 	skb->ignore_df = ip6_sk_ignore_df(sk);
1653 
1654 	*final_dst = fl6->daddr;
1655 	__skb_pull(skb, skb_network_header_len(skb));
1656 	if (opt && opt->opt_flen)
1657 		ipv6_push_frag_opts(skb, opt, &proto);
1658 	if (opt && opt->opt_nflen)
1659 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1660 
1661 	skb_push(skb, sizeof(struct ipv6hdr));
1662 	skb_reset_network_header(skb);
1663 	hdr = ipv6_hdr(skb);
1664 
1665 	ip6_flow_hdr(hdr, v6_cork->tclass,
1666 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1667 					ip6_autoflowlabel(net, np), fl6));
1668 	hdr->hop_limit = v6_cork->hop_limit;
1669 	hdr->nexthdr = proto;
1670 	hdr->saddr = fl6->saddr;
1671 	hdr->daddr = *final_dst;
1672 
1673 	skb->priority = sk->sk_priority;
1674 	skb->mark = sk->sk_mark;
1675 
1676 	skb->tstamp = cork->base.transmit_time;
1677 
1678 	skb_dst_set(skb, dst_clone(&rt->dst));
1679 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1680 	if (proto == IPPROTO_ICMPV6) {
1681 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1682 
1683 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1684 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1685 	}
1686 
1687 	ip6_cork_release(cork, v6_cork);
1688 out:
1689 	return skb;
1690 }
1691 
1692 int ip6_send_skb(struct sk_buff *skb)
1693 {
1694 	struct net *net = sock_net(skb->sk);
1695 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1696 	int err;
1697 
1698 	err = ip6_local_out(net, skb->sk, skb);
1699 	if (err) {
1700 		if (err > 0)
1701 			err = net_xmit_errno(err);
1702 		if (err)
1703 			IP6_INC_STATS(net, rt->rt6i_idev,
1704 				      IPSTATS_MIB_OUTDISCARDS);
1705 	}
1706 
1707 	return err;
1708 }
1709 
1710 int ip6_push_pending_frames(struct sock *sk)
1711 {
1712 	struct sk_buff *skb;
1713 
1714 	skb = ip6_finish_skb(sk);
1715 	if (!skb)
1716 		return 0;
1717 
1718 	return ip6_send_skb(skb);
1719 }
1720 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1721 
1722 static void __ip6_flush_pending_frames(struct sock *sk,
1723 				       struct sk_buff_head *queue,
1724 				       struct inet_cork_full *cork,
1725 				       struct inet6_cork *v6_cork)
1726 {
1727 	struct sk_buff *skb;
1728 
1729 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1730 		if (skb_dst(skb))
1731 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1732 				      IPSTATS_MIB_OUTDISCARDS);
1733 		kfree_skb(skb);
1734 	}
1735 
1736 	ip6_cork_release(cork, v6_cork);
1737 }
1738 
1739 void ip6_flush_pending_frames(struct sock *sk)
1740 {
1741 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1742 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1743 }
1744 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1745 
1746 struct sk_buff *ip6_make_skb(struct sock *sk,
1747 			     int getfrag(void *from, char *to, int offset,
1748 					 int len, int odd, struct sk_buff *skb),
1749 			     void *from, int length, int transhdrlen,
1750 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1751 			     struct rt6_info *rt, unsigned int flags,
1752 			     struct inet_cork_full *cork)
1753 {
1754 	struct inet6_cork v6_cork;
1755 	struct sk_buff_head queue;
1756 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1757 	int err;
1758 
1759 	if (flags & MSG_PROBE)
1760 		return NULL;
1761 
1762 	__skb_queue_head_init(&queue);
1763 
1764 	cork->base.flags = 0;
1765 	cork->base.addr = 0;
1766 	cork->base.opt = NULL;
1767 	cork->base.dst = NULL;
1768 	v6_cork.opt = NULL;
1769 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1770 	if (err) {
1771 		ip6_cork_release(cork, &v6_cork);
1772 		return ERR_PTR(err);
1773 	}
1774 	if (ipc6->dontfrag < 0)
1775 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1776 
1777 	err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1778 				&current->task_frag, getfrag, from,
1779 				length + exthdrlen, transhdrlen + exthdrlen,
1780 				flags, ipc6);
1781 	if (err) {
1782 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1783 		return ERR_PTR(err);
1784 	}
1785 
1786 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
1787 }
1788