xref: /linux/net/ipv6/ip6_output.c (revision 66a0e2d579dbec5c676cfe446234ffebb267c564)
1 /*
2  *	IPv6 output functions
3  *	Linux INET6 implementation
4  *
5  *	Authors:
6  *	Pedro Roque		<roque@di.fc.ul.pt>
7  *
8  *	Based on linux/net/ipv4/ip_output.c
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  *
15  *	Changes:
16  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
17  *				extension headers are implemented.
18  *				route changes now work.
19  *				ip6_forward does not confuse sniffers.
20  *				etc.
21  *
22  *      H. von Brand    :       Added missing #include <linux/string.h>
23  *	Imran Patel	:	frag id should be in NBO
24  *      Kazunori MIYAZAWA @USAGI
25  *			:       add ip6_append_data and related functions
26  *				for datagram xmit
27  */
28 
29 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/string.h>
32 #include <linux/socket.h>
33 #include <linux/net.h>
34 #include <linux/netdevice.h>
35 #include <linux/if_arp.h>
36 #include <linux/in6.h>
37 #include <linux/tcp.h>
38 #include <linux/route.h>
39 #include <linux/module.h>
40 #include <linux/slab.h>
41 
42 #include <linux/bpf-cgroup.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45 
46 #include <net/sock.h>
47 #include <net/snmp.h>
48 
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
61 
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 {
64 	struct dst_entry *dst = skb_dst(skb);
65 	struct net_device *dev = dst->dev;
66 	struct neighbour *neigh;
67 	struct in6_addr *nexthop;
68 	int ret;
69 
70 	skb->protocol = htons(ETH_P_IPV6);
71 	skb->dev = dev;
72 
73 	if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
74 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
75 
76 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
77 		    ((mroute6_socket(net, skb) &&
78 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
79 		     ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
80 					 &ipv6_hdr(skb)->saddr))) {
81 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
82 
83 			/* Do not check for IFF_ALLMULTI; multicast routing
84 			   is not supported in any case.
85 			 */
86 			if (newskb)
87 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
88 					net, sk, newskb, NULL, newskb->dev,
89 					dev_loopback_xmit);
90 
91 			if (ipv6_hdr(skb)->hop_limit == 0) {
92 				IP6_INC_STATS(net, idev,
93 					      IPSTATS_MIB_OUTDISCARDS);
94 				kfree_skb(skb);
95 				return 0;
96 			}
97 		}
98 
99 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
100 
101 		if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
102 		    IPV6_ADDR_SCOPE_NODELOCAL &&
103 		    !(dev->flags & IFF_LOOPBACK)) {
104 			kfree_skb(skb);
105 			return 0;
106 		}
107 	}
108 
109 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
110 		int res = lwtunnel_xmit(skb);
111 
112 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
113 			return res;
114 	}
115 
116 	rcu_read_lock_bh();
117 	nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
118 	neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
119 	if (unlikely(!neigh))
120 		neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
121 	if (!IS_ERR(neigh)) {
122 		ret = dst_neigh_output(dst, neigh, skb);
123 		rcu_read_unlock_bh();
124 		return ret;
125 	}
126 	rcu_read_unlock_bh();
127 
128 	IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
129 	kfree_skb(skb);
130 	return -EINVAL;
131 }
132 
133 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
134 {
135 	int ret;
136 
137 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
138 	if (ret) {
139 		kfree_skb(skb);
140 		return ret;
141 	}
142 
143 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
144 	    dst_allfrag(skb_dst(skb)) ||
145 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
146 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
147 	else
148 		return ip6_finish_output2(net, sk, skb);
149 }
150 
151 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
152 {
153 	struct net_device *dev = skb_dst(skb)->dev;
154 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
155 
156 	if (unlikely(idev->cnf.disable_ipv6)) {
157 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
158 		kfree_skb(skb);
159 		return 0;
160 	}
161 
162 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
163 			    net, sk, skb, NULL, dev,
164 			    ip6_finish_output,
165 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
166 }
167 
168 /*
169  * xmit an sk_buff (used by TCP, SCTP and DCCP)
170  * Note : socket lock is not held for SYNACK packets, but might be modified
171  * by calls to skb_set_owner_w() and ipv6_local_error(),
172  * which are using proper atomic operations or spinlocks.
173  */
174 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
175 	     __u32 mark, struct ipv6_txoptions *opt, int tclass)
176 {
177 	struct net *net = sock_net(sk);
178 	const struct ipv6_pinfo *np = inet6_sk(sk);
179 	struct in6_addr *first_hop = &fl6->daddr;
180 	struct dst_entry *dst = skb_dst(skb);
181 	struct ipv6hdr *hdr;
182 	u8  proto = fl6->flowi6_proto;
183 	int seg_len = skb->len;
184 	int hlimit = -1;
185 	u32 mtu;
186 
187 	if (opt) {
188 		unsigned int head_room;
189 
190 		/* First: exthdrs may take lots of space (~8K for now)
191 		   MAX_HEADER is not enough.
192 		 */
193 		head_room = opt->opt_nflen + opt->opt_flen;
194 		seg_len += head_room;
195 		head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
196 
197 		if (skb_headroom(skb) < head_room) {
198 			struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
199 			if (!skb2) {
200 				IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
201 					      IPSTATS_MIB_OUTDISCARDS);
202 				kfree_skb(skb);
203 				return -ENOBUFS;
204 			}
205 			consume_skb(skb);
206 			skb = skb2;
207 			/* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
208 			 * it is safe to call in our context (socket lock not held)
209 			 */
210 			skb_set_owner_w(skb, (struct sock *)sk);
211 		}
212 		if (opt->opt_flen)
213 			ipv6_push_frag_opts(skb, opt, &proto);
214 		if (opt->opt_nflen)
215 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
216 					     &fl6->saddr);
217 	}
218 
219 	skb_push(skb, sizeof(struct ipv6hdr));
220 	skb_reset_network_header(skb);
221 	hdr = ipv6_hdr(skb);
222 
223 	/*
224 	 *	Fill in the IPv6 header
225 	 */
226 	if (np)
227 		hlimit = np->hop_limit;
228 	if (hlimit < 0)
229 		hlimit = ip6_dst_hoplimit(dst);
230 
231 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
232 						     np->autoflowlabel, fl6));
233 
234 	hdr->payload_len = htons(seg_len);
235 	hdr->nexthdr = proto;
236 	hdr->hop_limit = hlimit;
237 
238 	hdr->saddr = fl6->saddr;
239 	hdr->daddr = *first_hop;
240 
241 	skb->protocol = htons(ETH_P_IPV6);
242 	skb->priority = sk->sk_priority;
243 	skb->mark = mark;
244 
245 	mtu = dst_mtu(dst);
246 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
247 		IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
248 			      IPSTATS_MIB_OUT, skb->len);
249 
250 		/* if egress device is enslaved to an L3 master device pass the
251 		 * skb to its handler for processing
252 		 */
253 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
254 		if (unlikely(!skb))
255 			return 0;
256 
257 		/* hooks should never assume socket lock is held.
258 		 * we promote our socket to non const
259 		 */
260 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
261 			       net, (struct sock *)sk, skb, NULL, dst->dev,
262 			       dst_output);
263 	}
264 
265 	skb->dev = dst->dev;
266 	/* ipv6_local_error() does not require socket lock,
267 	 * we promote our socket to non const
268 	 */
269 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
270 
271 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
272 	kfree_skb(skb);
273 	return -EMSGSIZE;
274 }
275 EXPORT_SYMBOL(ip6_xmit);
276 
277 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
278 {
279 	struct ip6_ra_chain *ra;
280 	struct sock *last = NULL;
281 
282 	read_lock(&ip6_ra_lock);
283 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
284 		struct sock *sk = ra->sk;
285 		if (sk && ra->sel == sel &&
286 		    (!sk->sk_bound_dev_if ||
287 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
288 			if (last) {
289 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
290 				if (skb2)
291 					rawv6_rcv(last, skb2);
292 			}
293 			last = sk;
294 		}
295 	}
296 
297 	if (last) {
298 		rawv6_rcv(last, skb);
299 		read_unlock(&ip6_ra_lock);
300 		return 1;
301 	}
302 	read_unlock(&ip6_ra_lock);
303 	return 0;
304 }
305 
306 static int ip6_forward_proxy_check(struct sk_buff *skb)
307 {
308 	struct ipv6hdr *hdr = ipv6_hdr(skb);
309 	u8 nexthdr = hdr->nexthdr;
310 	__be16 frag_off;
311 	int offset;
312 
313 	if (ipv6_ext_hdr(nexthdr)) {
314 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
315 		if (offset < 0)
316 			return 0;
317 	} else
318 		offset = sizeof(struct ipv6hdr);
319 
320 	if (nexthdr == IPPROTO_ICMPV6) {
321 		struct icmp6hdr *icmp6;
322 
323 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
324 					 offset + 1 - skb->data)))
325 			return 0;
326 
327 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
328 
329 		switch (icmp6->icmp6_type) {
330 		case NDISC_ROUTER_SOLICITATION:
331 		case NDISC_ROUTER_ADVERTISEMENT:
332 		case NDISC_NEIGHBOUR_SOLICITATION:
333 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
334 		case NDISC_REDIRECT:
335 			/* For reaction involving unicast neighbor discovery
336 			 * message destined to the proxied address, pass it to
337 			 * input function.
338 			 */
339 			return 1;
340 		default:
341 			break;
342 		}
343 	}
344 
345 	/*
346 	 * The proxying router can't forward traffic sent to a link-local
347 	 * address, so signal the sender and discard the packet. This
348 	 * behavior is clarified by the MIPv6 specification.
349 	 */
350 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
351 		dst_link_failure(skb);
352 		return -1;
353 	}
354 
355 	return 0;
356 }
357 
358 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
359 				     struct sk_buff *skb)
360 {
361 	return dst_output(net, sk, skb);
362 }
363 
364 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
365 {
366 	unsigned int mtu;
367 	struct inet6_dev *idev;
368 
369 	if (dst_metric_locked(dst, RTAX_MTU)) {
370 		mtu = dst_metric_raw(dst, RTAX_MTU);
371 		if (mtu)
372 			return mtu;
373 	}
374 
375 	mtu = IPV6_MIN_MTU;
376 	rcu_read_lock();
377 	idev = __in6_dev_get(dst->dev);
378 	if (idev)
379 		mtu = idev->cnf.mtu6;
380 	rcu_read_unlock();
381 
382 	return mtu;
383 }
384 
385 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
386 {
387 	if (skb->len <= mtu)
388 		return false;
389 
390 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
391 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
392 		return true;
393 
394 	if (skb->ignore_df)
395 		return false;
396 
397 	if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
398 		return false;
399 
400 	return true;
401 }
402 
403 int ip6_forward(struct sk_buff *skb)
404 {
405 	struct dst_entry *dst = skb_dst(skb);
406 	struct ipv6hdr *hdr = ipv6_hdr(skb);
407 	struct inet6_skb_parm *opt = IP6CB(skb);
408 	struct net *net = dev_net(dst->dev);
409 	u32 mtu;
410 
411 	if (net->ipv6.devconf_all->forwarding == 0)
412 		goto error;
413 
414 	if (skb->pkt_type != PACKET_HOST)
415 		goto drop;
416 
417 	if (unlikely(skb->sk))
418 		goto drop;
419 
420 	if (skb_warn_if_lro(skb))
421 		goto drop;
422 
423 	if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
424 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
425 				IPSTATS_MIB_INDISCARDS);
426 		goto drop;
427 	}
428 
429 	skb_forward_csum(skb);
430 
431 	/*
432 	 *	We DO NOT make any processing on
433 	 *	RA packets, pushing them to user level AS IS
434 	 *	without ane WARRANTY that application will be able
435 	 *	to interpret them. The reason is that we
436 	 *	cannot make anything clever here.
437 	 *
438 	 *	We are not end-node, so that if packet contains
439 	 *	AH/ESP, we cannot make anything.
440 	 *	Defragmentation also would be mistake, RA packets
441 	 *	cannot be fragmented, because there is no warranty
442 	 *	that different fragments will go along one path. --ANK
443 	 */
444 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
445 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
446 			return 0;
447 	}
448 
449 	/*
450 	 *	check and decrement ttl
451 	 */
452 	if (hdr->hop_limit <= 1) {
453 		/* Force OUTPUT device used as source address */
454 		skb->dev = dst->dev;
455 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
456 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
457 				IPSTATS_MIB_INHDRERRORS);
458 
459 		kfree_skb(skb);
460 		return -ETIMEDOUT;
461 	}
462 
463 	/* XXX: idev->cnf.proxy_ndp? */
464 	if (net->ipv6.devconf_all->proxy_ndp &&
465 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
466 		int proxied = ip6_forward_proxy_check(skb);
467 		if (proxied > 0)
468 			return ip6_input(skb);
469 		else if (proxied < 0) {
470 			__IP6_INC_STATS(net, ip6_dst_idev(dst),
471 					IPSTATS_MIB_INDISCARDS);
472 			goto drop;
473 		}
474 	}
475 
476 	if (!xfrm6_route_forward(skb)) {
477 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
478 				IPSTATS_MIB_INDISCARDS);
479 		goto drop;
480 	}
481 	dst = skb_dst(skb);
482 
483 	/* IPv6 specs say nothing about it, but it is clear that we cannot
484 	   send redirects to source routed frames.
485 	   We don't send redirects to frames decapsulated from IPsec.
486 	 */
487 	if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
488 		struct in6_addr *target = NULL;
489 		struct inet_peer *peer;
490 		struct rt6_info *rt;
491 
492 		/*
493 		 *	incoming and outgoing devices are the same
494 		 *	send a redirect.
495 		 */
496 
497 		rt = (struct rt6_info *) dst;
498 		if (rt->rt6i_flags & RTF_GATEWAY)
499 			target = &rt->rt6i_gateway;
500 		else
501 			target = &hdr->daddr;
502 
503 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
504 
505 		/* Limit redirects both by destination (here)
506 		   and by source (inside ndisc_send_redirect)
507 		 */
508 		if (inet_peer_xrlim_allow(peer, 1*HZ))
509 			ndisc_send_redirect(skb, target);
510 		if (peer)
511 			inet_putpeer(peer);
512 	} else {
513 		int addrtype = ipv6_addr_type(&hdr->saddr);
514 
515 		/* This check is security critical. */
516 		if (addrtype == IPV6_ADDR_ANY ||
517 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
518 			goto error;
519 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
520 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
521 				    ICMPV6_NOT_NEIGHBOUR, 0);
522 			goto error;
523 		}
524 	}
525 
526 	mtu = ip6_dst_mtu_forward(dst);
527 	if (mtu < IPV6_MIN_MTU)
528 		mtu = IPV6_MIN_MTU;
529 
530 	if (ip6_pkt_too_big(skb, mtu)) {
531 		/* Again, force OUTPUT device used as source address */
532 		skb->dev = dst->dev;
533 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
534 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
535 				IPSTATS_MIB_INTOOBIGERRORS);
536 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
537 				IPSTATS_MIB_FRAGFAILS);
538 		kfree_skb(skb);
539 		return -EMSGSIZE;
540 	}
541 
542 	if (skb_cow(skb, dst->dev->hard_header_len)) {
543 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
544 				IPSTATS_MIB_OUTDISCARDS);
545 		goto drop;
546 	}
547 
548 	hdr = ipv6_hdr(skb);
549 
550 	/* Mangling hops number delayed to point after skb COW */
551 
552 	hdr->hop_limit--;
553 
554 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
555 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
556 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
557 		       net, NULL, skb, skb->dev, dst->dev,
558 		       ip6_forward_finish);
559 
560 error:
561 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
562 drop:
563 	kfree_skb(skb);
564 	return -EINVAL;
565 }
566 
567 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
568 {
569 	to->pkt_type = from->pkt_type;
570 	to->priority = from->priority;
571 	to->protocol = from->protocol;
572 	skb_dst_drop(to);
573 	skb_dst_set(to, dst_clone(skb_dst(from)));
574 	to->dev = from->dev;
575 	to->mark = from->mark;
576 
577 #ifdef CONFIG_NET_SCHED
578 	to->tc_index = from->tc_index;
579 #endif
580 	nf_copy(to, from);
581 	skb_copy_secmark(to, from);
582 }
583 
584 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
585 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
586 {
587 	struct sk_buff *frag;
588 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
589 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
590 				inet6_sk(skb->sk) : NULL;
591 	struct ipv6hdr *tmp_hdr;
592 	struct frag_hdr *fh;
593 	unsigned int mtu, hlen, left, len;
594 	int hroom, troom;
595 	__be32 frag_id;
596 	int ptr, offset = 0, err = 0;
597 	u8 *prevhdr, nexthdr = 0;
598 
599 	hlen = ip6_find_1stfragopt(skb, &prevhdr);
600 	nexthdr = *prevhdr;
601 
602 	mtu = ip6_skb_dst_mtu(skb);
603 
604 	/* We must not fragment if the socket is set to force MTU discovery
605 	 * or if the skb it not generated by a local socket.
606 	 */
607 	if (unlikely(!skb->ignore_df && skb->len > mtu))
608 		goto fail_toobig;
609 
610 	if (IP6CB(skb)->frag_max_size) {
611 		if (IP6CB(skb)->frag_max_size > mtu)
612 			goto fail_toobig;
613 
614 		/* don't send fragments larger than what we received */
615 		mtu = IP6CB(skb)->frag_max_size;
616 		if (mtu < IPV6_MIN_MTU)
617 			mtu = IPV6_MIN_MTU;
618 	}
619 
620 	if (np && np->frag_size < mtu) {
621 		if (np->frag_size)
622 			mtu = np->frag_size;
623 	}
624 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
625 		goto fail_toobig;
626 	mtu -= hlen + sizeof(struct frag_hdr);
627 
628 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
629 				    &ipv6_hdr(skb)->saddr);
630 
631 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
632 	    (err = skb_checksum_help(skb)))
633 		goto fail;
634 
635 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
636 	if (skb_has_frag_list(skb)) {
637 		unsigned int first_len = skb_pagelen(skb);
638 		struct sk_buff *frag2;
639 
640 		if (first_len - hlen > mtu ||
641 		    ((first_len - hlen) & 7) ||
642 		    skb_cloned(skb) ||
643 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
644 			goto slow_path;
645 
646 		skb_walk_frags(skb, frag) {
647 			/* Correct geometry. */
648 			if (frag->len > mtu ||
649 			    ((frag->len & 7) && frag->next) ||
650 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
651 				goto slow_path_clean;
652 
653 			/* Partially cloned skb? */
654 			if (skb_shared(frag))
655 				goto slow_path_clean;
656 
657 			BUG_ON(frag->sk);
658 			if (skb->sk) {
659 				frag->sk = skb->sk;
660 				frag->destructor = sock_wfree;
661 			}
662 			skb->truesize -= frag->truesize;
663 		}
664 
665 		err = 0;
666 		offset = 0;
667 		/* BUILD HEADER */
668 
669 		*prevhdr = NEXTHDR_FRAGMENT;
670 		tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
671 		if (!tmp_hdr) {
672 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
673 				      IPSTATS_MIB_FRAGFAILS);
674 			err = -ENOMEM;
675 			goto fail;
676 		}
677 		frag = skb_shinfo(skb)->frag_list;
678 		skb_frag_list_init(skb);
679 
680 		__skb_pull(skb, hlen);
681 		fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
682 		__skb_push(skb, hlen);
683 		skb_reset_network_header(skb);
684 		memcpy(skb_network_header(skb), tmp_hdr, hlen);
685 
686 		fh->nexthdr = nexthdr;
687 		fh->reserved = 0;
688 		fh->frag_off = htons(IP6_MF);
689 		fh->identification = frag_id;
690 
691 		first_len = skb_pagelen(skb);
692 		skb->data_len = first_len - skb_headlen(skb);
693 		skb->len = first_len;
694 		ipv6_hdr(skb)->payload_len = htons(first_len -
695 						   sizeof(struct ipv6hdr));
696 
697 		dst_hold(&rt->dst);
698 
699 		for (;;) {
700 			/* Prepare header of the next frame,
701 			 * before previous one went down. */
702 			if (frag) {
703 				frag->ip_summed = CHECKSUM_NONE;
704 				skb_reset_transport_header(frag);
705 				fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
706 				__skb_push(frag, hlen);
707 				skb_reset_network_header(frag);
708 				memcpy(skb_network_header(frag), tmp_hdr,
709 				       hlen);
710 				offset += skb->len - hlen - sizeof(struct frag_hdr);
711 				fh->nexthdr = nexthdr;
712 				fh->reserved = 0;
713 				fh->frag_off = htons(offset);
714 				if (frag->next)
715 					fh->frag_off |= htons(IP6_MF);
716 				fh->identification = frag_id;
717 				ipv6_hdr(frag)->payload_len =
718 						htons(frag->len -
719 						      sizeof(struct ipv6hdr));
720 				ip6_copy_metadata(frag, skb);
721 			}
722 
723 			err = output(net, sk, skb);
724 			if (!err)
725 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
726 					      IPSTATS_MIB_FRAGCREATES);
727 
728 			if (err || !frag)
729 				break;
730 
731 			skb = frag;
732 			frag = skb->next;
733 			skb->next = NULL;
734 		}
735 
736 		kfree(tmp_hdr);
737 
738 		if (err == 0) {
739 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
740 				      IPSTATS_MIB_FRAGOKS);
741 			ip6_rt_put(rt);
742 			return 0;
743 		}
744 
745 		kfree_skb_list(frag);
746 
747 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
748 			      IPSTATS_MIB_FRAGFAILS);
749 		ip6_rt_put(rt);
750 		return err;
751 
752 slow_path_clean:
753 		skb_walk_frags(skb, frag2) {
754 			if (frag2 == frag)
755 				break;
756 			frag2->sk = NULL;
757 			frag2->destructor = NULL;
758 			skb->truesize += frag2->truesize;
759 		}
760 	}
761 
762 slow_path:
763 	left = skb->len - hlen;		/* Space per frame */
764 	ptr = hlen;			/* Where to start from */
765 
766 	/*
767 	 *	Fragment the datagram.
768 	 */
769 
770 	*prevhdr = NEXTHDR_FRAGMENT;
771 	troom = rt->dst.dev->needed_tailroom;
772 
773 	/*
774 	 *	Keep copying data until we run out.
775 	 */
776 	while (left > 0)	{
777 		len = left;
778 		/* IF: it doesn't fit, use 'mtu' - the data space left */
779 		if (len > mtu)
780 			len = mtu;
781 		/* IF: we are not sending up to and including the packet end
782 		   then align the next start on an eight byte boundary */
783 		if (len < left)	{
784 			len &= ~7;
785 		}
786 
787 		/* Allocate buffer */
788 		frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
789 				 hroom + troom, GFP_ATOMIC);
790 		if (!frag) {
791 			IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
792 				      IPSTATS_MIB_FRAGFAILS);
793 			err = -ENOMEM;
794 			goto fail;
795 		}
796 
797 		/*
798 		 *	Set up data on packet
799 		 */
800 
801 		ip6_copy_metadata(frag, skb);
802 		skb_reserve(frag, hroom);
803 		skb_put(frag, len + hlen + sizeof(struct frag_hdr));
804 		skb_reset_network_header(frag);
805 		fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
806 		frag->transport_header = (frag->network_header + hlen +
807 					  sizeof(struct frag_hdr));
808 
809 		/*
810 		 *	Charge the memory for the fragment to any owner
811 		 *	it might possess
812 		 */
813 		if (skb->sk)
814 			skb_set_owner_w(frag, skb->sk);
815 
816 		/*
817 		 *	Copy the packet header into the new buffer.
818 		 */
819 		skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
820 
821 		/*
822 		 *	Build fragment header.
823 		 */
824 		fh->nexthdr = nexthdr;
825 		fh->reserved = 0;
826 		fh->identification = frag_id;
827 
828 		/*
829 		 *	Copy a block of the IP datagram.
830 		 */
831 		BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
832 				     len));
833 		left -= len;
834 
835 		fh->frag_off = htons(offset);
836 		if (left > 0)
837 			fh->frag_off |= htons(IP6_MF);
838 		ipv6_hdr(frag)->payload_len = htons(frag->len -
839 						    sizeof(struct ipv6hdr));
840 
841 		ptr += len;
842 		offset += len;
843 
844 		/*
845 		 *	Put this fragment into the sending queue.
846 		 */
847 		err = output(net, sk, frag);
848 		if (err)
849 			goto fail;
850 
851 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
852 			      IPSTATS_MIB_FRAGCREATES);
853 	}
854 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
855 		      IPSTATS_MIB_FRAGOKS);
856 	consume_skb(skb);
857 	return err;
858 
859 fail_toobig:
860 	if (skb->sk && dst_allfrag(skb_dst(skb)))
861 		sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
862 
863 	skb->dev = skb_dst(skb)->dev;
864 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
865 	err = -EMSGSIZE;
866 
867 fail:
868 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
869 		      IPSTATS_MIB_FRAGFAILS);
870 	kfree_skb(skb);
871 	return err;
872 }
873 
874 static inline int ip6_rt_check(const struct rt6key *rt_key,
875 			       const struct in6_addr *fl_addr,
876 			       const struct in6_addr *addr_cache)
877 {
878 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
879 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
880 }
881 
882 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
883 					  struct dst_entry *dst,
884 					  const struct flowi6 *fl6)
885 {
886 	struct ipv6_pinfo *np = inet6_sk(sk);
887 	struct rt6_info *rt;
888 
889 	if (!dst)
890 		goto out;
891 
892 	if (dst->ops->family != AF_INET6) {
893 		dst_release(dst);
894 		return NULL;
895 	}
896 
897 	rt = (struct rt6_info *)dst;
898 	/* Yes, checking route validity in not connected
899 	 * case is not very simple. Take into account,
900 	 * that we do not support routing by source, TOS,
901 	 * and MSG_DONTROUTE		--ANK (980726)
902 	 *
903 	 * 1. ip6_rt_check(): If route was host route,
904 	 *    check that cached destination is current.
905 	 *    If it is network route, we still may
906 	 *    check its validity using saved pointer
907 	 *    to the last used address: daddr_cache.
908 	 *    We do not want to save whole address now,
909 	 *    (because main consumer of this service
910 	 *    is tcp, which has not this problem),
911 	 *    so that the last trick works only on connected
912 	 *    sockets.
913 	 * 2. oif also should be the same.
914 	 */
915 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
916 #ifdef CONFIG_IPV6_SUBTREES
917 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
918 #endif
919 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
920 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
921 		dst_release(dst);
922 		dst = NULL;
923 	}
924 
925 out:
926 	return dst;
927 }
928 
929 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
930 			       struct dst_entry **dst, struct flowi6 *fl6)
931 {
932 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
933 	struct neighbour *n;
934 	struct rt6_info *rt;
935 #endif
936 	int err;
937 	int flags = 0;
938 
939 	/* The correct way to handle this would be to do
940 	 * ip6_route_get_saddr, and then ip6_route_output; however,
941 	 * the route-specific preferred source forces the
942 	 * ip6_route_output call _before_ ip6_route_get_saddr.
943 	 *
944 	 * In source specific routing (no src=any default route),
945 	 * ip6_route_output will fail given src=any saddr, though, so
946 	 * that's why we try it again later.
947 	 */
948 	if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
949 		struct rt6_info *rt;
950 		bool had_dst = *dst != NULL;
951 
952 		if (!had_dst)
953 			*dst = ip6_route_output(net, sk, fl6);
954 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
955 		err = ip6_route_get_saddr(net, rt, &fl6->daddr,
956 					  sk ? inet6_sk(sk)->srcprefs : 0,
957 					  &fl6->saddr);
958 		if (err)
959 			goto out_err_release;
960 
961 		/* If we had an erroneous initial result, pretend it
962 		 * never existed and let the SA-enabled version take
963 		 * over.
964 		 */
965 		if (!had_dst && (*dst)->error) {
966 			dst_release(*dst);
967 			*dst = NULL;
968 		}
969 
970 		if (fl6->flowi6_oif)
971 			flags |= RT6_LOOKUP_F_IFACE;
972 	}
973 
974 	if (!*dst)
975 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
976 
977 	err = (*dst)->error;
978 	if (err)
979 		goto out_err_release;
980 
981 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
982 	/*
983 	 * Here if the dst entry we've looked up
984 	 * has a neighbour entry that is in the INCOMPLETE
985 	 * state and the src address from the flow is
986 	 * marked as OPTIMISTIC, we release the found
987 	 * dst entry and replace it instead with the
988 	 * dst entry of the nexthop router
989 	 */
990 	rt = (struct rt6_info *) *dst;
991 	rcu_read_lock_bh();
992 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
993 				      rt6_nexthop(rt, &fl6->daddr));
994 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
995 	rcu_read_unlock_bh();
996 
997 	if (err) {
998 		struct inet6_ifaddr *ifp;
999 		struct flowi6 fl_gw6;
1000 		int redirect;
1001 
1002 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1003 				      (*dst)->dev, 1);
1004 
1005 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1006 		if (ifp)
1007 			in6_ifa_put(ifp);
1008 
1009 		if (redirect) {
1010 			/*
1011 			 * We need to get the dst entry for the
1012 			 * default router instead
1013 			 */
1014 			dst_release(*dst);
1015 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1016 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1017 			*dst = ip6_route_output(net, sk, &fl_gw6);
1018 			err = (*dst)->error;
1019 			if (err)
1020 				goto out_err_release;
1021 		}
1022 	}
1023 #endif
1024 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1025 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1026 		err = -EAFNOSUPPORT;
1027 		goto out_err_release;
1028 	}
1029 
1030 	return 0;
1031 
1032 out_err_release:
1033 	dst_release(*dst);
1034 	*dst = NULL;
1035 
1036 	if (err == -ENETUNREACH)
1037 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1038 	return err;
1039 }
1040 
1041 /**
1042  *	ip6_dst_lookup - perform route lookup on flow
1043  *	@sk: socket which provides route info
1044  *	@dst: pointer to dst_entry * for result
1045  *	@fl6: flow to lookup
1046  *
1047  *	This function performs a route lookup on the given flow.
1048  *
1049  *	It returns zero on success, or a standard errno code on error.
1050  */
1051 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1052 		   struct flowi6 *fl6)
1053 {
1054 	*dst = NULL;
1055 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1056 }
1057 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1058 
1059 /**
1060  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1061  *	@sk: socket which provides route info
1062  *	@fl6: flow to lookup
1063  *	@final_dst: final destination address for ipsec lookup
1064  *
1065  *	This function performs a route lookup on the given flow.
1066  *
1067  *	It returns a valid dst pointer on success, or a pointer encoded
1068  *	error code.
1069  */
1070 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1071 				      const struct in6_addr *final_dst)
1072 {
1073 	struct dst_entry *dst = NULL;
1074 	int err;
1075 
1076 	err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1077 	if (err)
1078 		return ERR_PTR(err);
1079 	if (final_dst)
1080 		fl6->daddr = *final_dst;
1081 
1082 	return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1083 }
1084 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1085 
1086 /**
1087  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1088  *	@sk: socket which provides the dst cache and route info
1089  *	@fl6: flow to lookup
1090  *	@final_dst: final destination address for ipsec lookup
1091  *
1092  *	This function performs a route lookup on the given flow with the
1093  *	possibility of using the cached route in the socket if it is valid.
1094  *	It will take the socket dst lock when operating on the dst cache.
1095  *	As a result, this function can only be used in process context.
1096  *
1097  *	It returns a valid dst pointer on success, or a pointer encoded
1098  *	error code.
1099  */
1100 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1101 					 const struct in6_addr *final_dst)
1102 {
1103 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1104 
1105 	dst = ip6_sk_dst_check(sk, dst, fl6);
1106 	if (!dst)
1107 		dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1108 
1109 	return dst;
1110 }
1111 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1112 
1113 static inline int ip6_ufo_append_data(struct sock *sk,
1114 			struct sk_buff_head *queue,
1115 			int getfrag(void *from, char *to, int offset, int len,
1116 			int odd, struct sk_buff *skb),
1117 			void *from, int length, int hh_len, int fragheaderlen,
1118 			int exthdrlen, int transhdrlen, int mtu,
1119 			unsigned int flags, const struct flowi6 *fl6)
1120 
1121 {
1122 	struct sk_buff *skb;
1123 	int err;
1124 
1125 	/* There is support for UDP large send offload by network
1126 	 * device, so create one single skb packet containing complete
1127 	 * udp datagram
1128 	 */
1129 	skb = skb_peek_tail(queue);
1130 	if (!skb) {
1131 		skb = sock_alloc_send_skb(sk,
1132 			hh_len + fragheaderlen + transhdrlen + 20,
1133 			(flags & MSG_DONTWAIT), &err);
1134 		if (!skb)
1135 			return err;
1136 
1137 		/* reserve space for Hardware header */
1138 		skb_reserve(skb, hh_len);
1139 
1140 		/* create space for UDP/IP header */
1141 		skb_put(skb, fragheaderlen + transhdrlen);
1142 
1143 		/* initialize network header pointer */
1144 		skb_set_network_header(skb, exthdrlen);
1145 
1146 		/* initialize protocol header pointer */
1147 		skb->transport_header = skb->network_header + fragheaderlen;
1148 
1149 		skb->protocol = htons(ETH_P_IPV6);
1150 		skb->csum = 0;
1151 
1152 		__skb_queue_tail(queue, skb);
1153 	} else if (skb_is_gso(skb)) {
1154 		goto append;
1155 	}
1156 
1157 	skb->ip_summed = CHECKSUM_PARTIAL;
1158 	/* Specify the length of each IPv6 datagram fragment.
1159 	 * It has to be a multiple of 8.
1160 	 */
1161 	skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1162 				     sizeof(struct frag_hdr)) & ~7;
1163 	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1164 	skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1165 							 &fl6->daddr,
1166 							 &fl6->saddr);
1167 
1168 append:
1169 	return skb_append_datato_frags(sk, skb, getfrag, from,
1170 				       (length - transhdrlen));
1171 }
1172 
1173 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1174 					       gfp_t gfp)
1175 {
1176 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1177 }
1178 
1179 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1180 						gfp_t gfp)
1181 {
1182 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1183 }
1184 
1185 static void ip6_append_data_mtu(unsigned int *mtu,
1186 				int *maxfraglen,
1187 				unsigned int fragheaderlen,
1188 				struct sk_buff *skb,
1189 				struct rt6_info *rt,
1190 				unsigned int orig_mtu)
1191 {
1192 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1193 		if (!skb) {
1194 			/* first fragment, reserve header_len */
1195 			*mtu = orig_mtu - rt->dst.header_len;
1196 
1197 		} else {
1198 			/*
1199 			 * this fragment is not first, the headers
1200 			 * space is regarded as data space.
1201 			 */
1202 			*mtu = orig_mtu;
1203 		}
1204 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1205 			      + fragheaderlen - sizeof(struct frag_hdr);
1206 	}
1207 }
1208 
1209 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1210 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1211 			  struct rt6_info *rt, struct flowi6 *fl6)
1212 {
1213 	struct ipv6_pinfo *np = inet6_sk(sk);
1214 	unsigned int mtu;
1215 	struct ipv6_txoptions *opt = ipc6->opt;
1216 
1217 	/*
1218 	 * setup for corking
1219 	 */
1220 	if (opt) {
1221 		if (WARN_ON(v6_cork->opt))
1222 			return -EINVAL;
1223 
1224 		v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1225 		if (unlikely(!v6_cork->opt))
1226 			return -ENOBUFS;
1227 
1228 		v6_cork->opt->tot_len = opt->tot_len;
1229 		v6_cork->opt->opt_flen = opt->opt_flen;
1230 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1231 
1232 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1233 						    sk->sk_allocation);
1234 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1235 			return -ENOBUFS;
1236 
1237 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1238 						    sk->sk_allocation);
1239 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1240 			return -ENOBUFS;
1241 
1242 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1243 						   sk->sk_allocation);
1244 		if (opt->hopopt && !v6_cork->opt->hopopt)
1245 			return -ENOBUFS;
1246 
1247 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1248 						    sk->sk_allocation);
1249 		if (opt->srcrt && !v6_cork->opt->srcrt)
1250 			return -ENOBUFS;
1251 
1252 		/* need source address above miyazawa*/
1253 	}
1254 	dst_hold(&rt->dst);
1255 	cork->base.dst = &rt->dst;
1256 	cork->fl.u.ip6 = *fl6;
1257 	v6_cork->hop_limit = ipc6->hlimit;
1258 	v6_cork->tclass = ipc6->tclass;
1259 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1260 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1261 		      rt->dst.dev->mtu : dst_mtu(&rt->dst);
1262 	else
1263 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1264 		      rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1265 	if (np->frag_size < mtu) {
1266 		if (np->frag_size)
1267 			mtu = np->frag_size;
1268 	}
1269 	cork->base.fragsize = mtu;
1270 	if (dst_allfrag(rt->dst.path))
1271 		cork->base.flags |= IPCORK_ALLFRAG;
1272 	cork->base.length = 0;
1273 
1274 	return 0;
1275 }
1276 
1277 static int __ip6_append_data(struct sock *sk,
1278 			     struct flowi6 *fl6,
1279 			     struct sk_buff_head *queue,
1280 			     struct inet_cork *cork,
1281 			     struct inet6_cork *v6_cork,
1282 			     struct page_frag *pfrag,
1283 			     int getfrag(void *from, char *to, int offset,
1284 					 int len, int odd, struct sk_buff *skb),
1285 			     void *from, int length, int transhdrlen,
1286 			     unsigned int flags, struct ipcm6_cookie *ipc6,
1287 			     const struct sockcm_cookie *sockc)
1288 {
1289 	struct sk_buff *skb, *skb_prev = NULL;
1290 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1291 	int exthdrlen = 0;
1292 	int dst_exthdrlen = 0;
1293 	int hh_len;
1294 	int copy;
1295 	int err;
1296 	int offset = 0;
1297 	__u8 tx_flags = 0;
1298 	u32 tskey = 0;
1299 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1300 	struct ipv6_txoptions *opt = v6_cork->opt;
1301 	int csummode = CHECKSUM_NONE;
1302 	unsigned int maxnonfragsize, headersize;
1303 
1304 	skb = skb_peek_tail(queue);
1305 	if (!skb) {
1306 		exthdrlen = opt ? opt->opt_flen : 0;
1307 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1308 	}
1309 
1310 	mtu = cork->fragsize;
1311 	orig_mtu = mtu;
1312 
1313 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1314 
1315 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1316 			(opt ? opt->opt_nflen : 0);
1317 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1318 		     sizeof(struct frag_hdr);
1319 
1320 	headersize = sizeof(struct ipv6hdr) +
1321 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1322 		     (dst_allfrag(&rt->dst) ?
1323 		      sizeof(struct frag_hdr) : 0) +
1324 		     rt->rt6i_nfheader_len;
1325 
1326 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1327 	    (sk->sk_protocol == IPPROTO_UDP ||
1328 	     sk->sk_protocol == IPPROTO_RAW)) {
1329 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1330 				sizeof(struct ipv6hdr));
1331 		goto emsgsize;
1332 	}
1333 
1334 	if (ip6_sk_ignore_df(sk))
1335 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1336 	else
1337 		maxnonfragsize = mtu;
1338 
1339 	if (cork->length + length > maxnonfragsize - headersize) {
1340 emsgsize:
1341 		ipv6_local_error(sk, EMSGSIZE, fl6,
1342 				 mtu - headersize +
1343 				 sizeof(struct ipv6hdr));
1344 		return -EMSGSIZE;
1345 	}
1346 
1347 	/* CHECKSUM_PARTIAL only with no extension headers and when
1348 	 * we are not going to fragment
1349 	 */
1350 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1351 	    headersize == sizeof(struct ipv6hdr) &&
1352 	    length <= mtu - headersize &&
1353 	    !(flags & MSG_MORE) &&
1354 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1355 		csummode = CHECKSUM_PARTIAL;
1356 
1357 	if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1358 		sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1359 		if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1360 		    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1361 			tskey = sk->sk_tskey++;
1362 	}
1363 
1364 	/*
1365 	 * Let's try using as much space as possible.
1366 	 * Use MTU if total length of the message fits into the MTU.
1367 	 * Otherwise, we need to reserve fragment header and
1368 	 * fragment alignment (= 8-15 octects, in total).
1369 	 *
1370 	 * Note that we may need to "move" the data from the tail of
1371 	 * of the buffer to the new fragment when we split
1372 	 * the message.
1373 	 *
1374 	 * FIXME: It may be fragmented into multiple chunks
1375 	 *        at once if non-fragmentable extension headers
1376 	 *        are too large.
1377 	 * --yoshfuji
1378 	 */
1379 
1380 	cork->length += length;
1381 	if ((((length + fragheaderlen) > mtu) ||
1382 	     (skb && skb_is_gso(skb))) &&
1383 	    (sk->sk_protocol == IPPROTO_UDP) &&
1384 	    (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
1385 	    (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) {
1386 		err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1387 					  hh_len, fragheaderlen, exthdrlen,
1388 					  transhdrlen, mtu, flags, fl6);
1389 		if (err)
1390 			goto error;
1391 		return 0;
1392 	}
1393 
1394 	if (!skb)
1395 		goto alloc_new_skb;
1396 
1397 	while (length > 0) {
1398 		/* Check if the remaining data fits into current packet. */
1399 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1400 		if (copy < length)
1401 			copy = maxfraglen - skb->len;
1402 
1403 		if (copy <= 0) {
1404 			char *data;
1405 			unsigned int datalen;
1406 			unsigned int fraglen;
1407 			unsigned int fraggap;
1408 			unsigned int alloclen;
1409 alloc_new_skb:
1410 			/* There's no room in the current skb */
1411 			if (skb)
1412 				fraggap = skb->len - maxfraglen;
1413 			else
1414 				fraggap = 0;
1415 			/* update mtu and maxfraglen if necessary */
1416 			if (!skb || !skb_prev)
1417 				ip6_append_data_mtu(&mtu, &maxfraglen,
1418 						    fragheaderlen, skb, rt,
1419 						    orig_mtu);
1420 
1421 			skb_prev = skb;
1422 
1423 			/*
1424 			 * If remaining data exceeds the mtu,
1425 			 * we know we need more fragment(s).
1426 			 */
1427 			datalen = length + fraggap;
1428 
1429 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1430 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1431 			if ((flags & MSG_MORE) &&
1432 			    !(rt->dst.dev->features&NETIF_F_SG))
1433 				alloclen = mtu;
1434 			else
1435 				alloclen = datalen + fragheaderlen;
1436 
1437 			alloclen += dst_exthdrlen;
1438 
1439 			if (datalen != length + fraggap) {
1440 				/*
1441 				 * this is not the last fragment, the trailer
1442 				 * space is regarded as data space.
1443 				 */
1444 				datalen += rt->dst.trailer_len;
1445 			}
1446 
1447 			alloclen += rt->dst.trailer_len;
1448 			fraglen = datalen + fragheaderlen;
1449 
1450 			/*
1451 			 * We just reserve space for fragment header.
1452 			 * Note: this may be overallocation if the message
1453 			 * (without MSG_MORE) fits into the MTU.
1454 			 */
1455 			alloclen += sizeof(struct frag_hdr);
1456 
1457 			if (transhdrlen) {
1458 				skb = sock_alloc_send_skb(sk,
1459 						alloclen + hh_len,
1460 						(flags & MSG_DONTWAIT), &err);
1461 			} else {
1462 				skb = NULL;
1463 				if (atomic_read(&sk->sk_wmem_alloc) <=
1464 				    2 * sk->sk_sndbuf)
1465 					skb = sock_wmalloc(sk,
1466 							   alloclen + hh_len, 1,
1467 							   sk->sk_allocation);
1468 				if (unlikely(!skb))
1469 					err = -ENOBUFS;
1470 			}
1471 			if (!skb)
1472 				goto error;
1473 			/*
1474 			 *	Fill in the control structures
1475 			 */
1476 			skb->protocol = htons(ETH_P_IPV6);
1477 			skb->ip_summed = csummode;
1478 			skb->csum = 0;
1479 			/* reserve for fragmentation and ipsec header */
1480 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1481 				    dst_exthdrlen);
1482 
1483 			/* Only the initial fragment is time stamped */
1484 			skb_shinfo(skb)->tx_flags = tx_flags;
1485 			tx_flags = 0;
1486 			skb_shinfo(skb)->tskey = tskey;
1487 			tskey = 0;
1488 
1489 			/*
1490 			 *	Find where to start putting bytes
1491 			 */
1492 			data = skb_put(skb, fraglen);
1493 			skb_set_network_header(skb, exthdrlen);
1494 			data += fragheaderlen;
1495 			skb->transport_header = (skb->network_header +
1496 						 fragheaderlen);
1497 			if (fraggap) {
1498 				skb->csum = skb_copy_and_csum_bits(
1499 					skb_prev, maxfraglen,
1500 					data + transhdrlen, fraggap, 0);
1501 				skb_prev->csum = csum_sub(skb_prev->csum,
1502 							  skb->csum);
1503 				data += fraggap;
1504 				pskb_trim_unique(skb_prev, maxfraglen);
1505 			}
1506 			copy = datalen - transhdrlen - fraggap;
1507 
1508 			if (copy < 0) {
1509 				err = -EINVAL;
1510 				kfree_skb(skb);
1511 				goto error;
1512 			} else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1513 				err = -EFAULT;
1514 				kfree_skb(skb);
1515 				goto error;
1516 			}
1517 
1518 			offset += copy;
1519 			length -= datalen - fraggap;
1520 			transhdrlen = 0;
1521 			exthdrlen = 0;
1522 			dst_exthdrlen = 0;
1523 
1524 			/*
1525 			 * Put the packet on the pending queue
1526 			 */
1527 			__skb_queue_tail(queue, skb);
1528 			continue;
1529 		}
1530 
1531 		if (copy > length)
1532 			copy = length;
1533 
1534 		if (!(rt->dst.dev->features&NETIF_F_SG)) {
1535 			unsigned int off;
1536 
1537 			off = skb->len;
1538 			if (getfrag(from, skb_put(skb, copy),
1539 						offset, copy, off, skb) < 0) {
1540 				__skb_trim(skb, off);
1541 				err = -EFAULT;
1542 				goto error;
1543 			}
1544 		} else {
1545 			int i = skb_shinfo(skb)->nr_frags;
1546 
1547 			err = -ENOMEM;
1548 			if (!sk_page_frag_refill(sk, pfrag))
1549 				goto error;
1550 
1551 			if (!skb_can_coalesce(skb, i, pfrag->page,
1552 					      pfrag->offset)) {
1553 				err = -EMSGSIZE;
1554 				if (i == MAX_SKB_FRAGS)
1555 					goto error;
1556 
1557 				__skb_fill_page_desc(skb, i, pfrag->page,
1558 						     pfrag->offset, 0);
1559 				skb_shinfo(skb)->nr_frags = ++i;
1560 				get_page(pfrag->page);
1561 			}
1562 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1563 			if (getfrag(from,
1564 				    page_address(pfrag->page) + pfrag->offset,
1565 				    offset, copy, skb->len, skb) < 0)
1566 				goto error_efault;
1567 
1568 			pfrag->offset += copy;
1569 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1570 			skb->len += copy;
1571 			skb->data_len += copy;
1572 			skb->truesize += copy;
1573 			atomic_add(copy, &sk->sk_wmem_alloc);
1574 		}
1575 		offset += copy;
1576 		length -= copy;
1577 	}
1578 
1579 	return 0;
1580 
1581 error_efault:
1582 	err = -EFAULT;
1583 error:
1584 	cork->length -= length;
1585 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1586 	return err;
1587 }
1588 
1589 int ip6_append_data(struct sock *sk,
1590 		    int getfrag(void *from, char *to, int offset, int len,
1591 				int odd, struct sk_buff *skb),
1592 		    void *from, int length, int transhdrlen,
1593 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1594 		    struct rt6_info *rt, unsigned int flags,
1595 		    const struct sockcm_cookie *sockc)
1596 {
1597 	struct inet_sock *inet = inet_sk(sk);
1598 	struct ipv6_pinfo *np = inet6_sk(sk);
1599 	int exthdrlen;
1600 	int err;
1601 
1602 	if (flags&MSG_PROBE)
1603 		return 0;
1604 	if (skb_queue_empty(&sk->sk_write_queue)) {
1605 		/*
1606 		 * setup for corking
1607 		 */
1608 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1609 				     ipc6, rt, fl6);
1610 		if (err)
1611 			return err;
1612 
1613 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1614 		length += exthdrlen;
1615 		transhdrlen += exthdrlen;
1616 	} else {
1617 		fl6 = &inet->cork.fl.u.ip6;
1618 		transhdrlen = 0;
1619 	}
1620 
1621 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1622 				 &np->cork, sk_page_frag(sk), getfrag,
1623 				 from, length, transhdrlen, flags, ipc6, sockc);
1624 }
1625 EXPORT_SYMBOL_GPL(ip6_append_data);
1626 
1627 static void ip6_cork_release(struct inet_cork_full *cork,
1628 			     struct inet6_cork *v6_cork)
1629 {
1630 	if (v6_cork->opt) {
1631 		kfree(v6_cork->opt->dst0opt);
1632 		kfree(v6_cork->opt->dst1opt);
1633 		kfree(v6_cork->opt->hopopt);
1634 		kfree(v6_cork->opt->srcrt);
1635 		kfree(v6_cork->opt);
1636 		v6_cork->opt = NULL;
1637 	}
1638 
1639 	if (cork->base.dst) {
1640 		dst_release(cork->base.dst);
1641 		cork->base.dst = NULL;
1642 		cork->base.flags &= ~IPCORK_ALLFRAG;
1643 	}
1644 	memset(&cork->fl, 0, sizeof(cork->fl));
1645 }
1646 
1647 struct sk_buff *__ip6_make_skb(struct sock *sk,
1648 			       struct sk_buff_head *queue,
1649 			       struct inet_cork_full *cork,
1650 			       struct inet6_cork *v6_cork)
1651 {
1652 	struct sk_buff *skb, *tmp_skb;
1653 	struct sk_buff **tail_skb;
1654 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1655 	struct ipv6_pinfo *np = inet6_sk(sk);
1656 	struct net *net = sock_net(sk);
1657 	struct ipv6hdr *hdr;
1658 	struct ipv6_txoptions *opt = v6_cork->opt;
1659 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1660 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1661 	unsigned char proto = fl6->flowi6_proto;
1662 
1663 	skb = __skb_dequeue(queue);
1664 	if (!skb)
1665 		goto out;
1666 	tail_skb = &(skb_shinfo(skb)->frag_list);
1667 
1668 	/* move skb->data to ip header from ext header */
1669 	if (skb->data < skb_network_header(skb))
1670 		__skb_pull(skb, skb_network_offset(skb));
1671 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1672 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1673 		*tail_skb = tmp_skb;
1674 		tail_skb = &(tmp_skb->next);
1675 		skb->len += tmp_skb->len;
1676 		skb->data_len += tmp_skb->len;
1677 		skb->truesize += tmp_skb->truesize;
1678 		tmp_skb->destructor = NULL;
1679 		tmp_skb->sk = NULL;
1680 	}
1681 
1682 	/* Allow local fragmentation. */
1683 	skb->ignore_df = ip6_sk_ignore_df(sk);
1684 
1685 	*final_dst = fl6->daddr;
1686 	__skb_pull(skb, skb_network_header_len(skb));
1687 	if (opt && opt->opt_flen)
1688 		ipv6_push_frag_opts(skb, opt, &proto);
1689 	if (opt && opt->opt_nflen)
1690 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1691 
1692 	skb_push(skb, sizeof(struct ipv6hdr));
1693 	skb_reset_network_header(skb);
1694 	hdr = ipv6_hdr(skb);
1695 
1696 	ip6_flow_hdr(hdr, v6_cork->tclass,
1697 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1698 					np->autoflowlabel, fl6));
1699 	hdr->hop_limit = v6_cork->hop_limit;
1700 	hdr->nexthdr = proto;
1701 	hdr->saddr = fl6->saddr;
1702 	hdr->daddr = *final_dst;
1703 
1704 	skb->priority = sk->sk_priority;
1705 	skb->mark = sk->sk_mark;
1706 
1707 	skb_dst_set(skb, dst_clone(&rt->dst));
1708 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1709 	if (proto == IPPROTO_ICMPV6) {
1710 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1711 
1712 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1713 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1714 	}
1715 
1716 	ip6_cork_release(cork, v6_cork);
1717 out:
1718 	return skb;
1719 }
1720 
1721 int ip6_send_skb(struct sk_buff *skb)
1722 {
1723 	struct net *net = sock_net(skb->sk);
1724 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1725 	int err;
1726 
1727 	err = ip6_local_out(net, skb->sk, skb);
1728 	if (err) {
1729 		if (err > 0)
1730 			err = net_xmit_errno(err);
1731 		if (err)
1732 			IP6_INC_STATS(net, rt->rt6i_idev,
1733 				      IPSTATS_MIB_OUTDISCARDS);
1734 	}
1735 
1736 	return err;
1737 }
1738 
1739 int ip6_push_pending_frames(struct sock *sk)
1740 {
1741 	struct sk_buff *skb;
1742 
1743 	skb = ip6_finish_skb(sk);
1744 	if (!skb)
1745 		return 0;
1746 
1747 	return ip6_send_skb(skb);
1748 }
1749 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1750 
1751 static void __ip6_flush_pending_frames(struct sock *sk,
1752 				       struct sk_buff_head *queue,
1753 				       struct inet_cork_full *cork,
1754 				       struct inet6_cork *v6_cork)
1755 {
1756 	struct sk_buff *skb;
1757 
1758 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1759 		if (skb_dst(skb))
1760 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1761 				      IPSTATS_MIB_OUTDISCARDS);
1762 		kfree_skb(skb);
1763 	}
1764 
1765 	ip6_cork_release(cork, v6_cork);
1766 }
1767 
1768 void ip6_flush_pending_frames(struct sock *sk)
1769 {
1770 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1771 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1772 }
1773 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1774 
1775 struct sk_buff *ip6_make_skb(struct sock *sk,
1776 			     int getfrag(void *from, char *to, int offset,
1777 					 int len, int odd, struct sk_buff *skb),
1778 			     void *from, int length, int transhdrlen,
1779 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1780 			     struct rt6_info *rt, unsigned int flags,
1781 			     const struct sockcm_cookie *sockc)
1782 {
1783 	struct inet_cork_full cork;
1784 	struct inet6_cork v6_cork;
1785 	struct sk_buff_head queue;
1786 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1787 	int err;
1788 
1789 	if (flags & MSG_PROBE)
1790 		return NULL;
1791 
1792 	__skb_queue_head_init(&queue);
1793 
1794 	cork.base.flags = 0;
1795 	cork.base.addr = 0;
1796 	cork.base.opt = NULL;
1797 	v6_cork.opt = NULL;
1798 	err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
1799 	if (err)
1800 		return ERR_PTR(err);
1801 
1802 	if (ipc6->dontfrag < 0)
1803 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1804 
1805 	err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1806 				&current->task_frag, getfrag, from,
1807 				length + exthdrlen, transhdrlen + exthdrlen,
1808 				flags, ipc6, sockc);
1809 	if (err) {
1810 		__ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1811 		return ERR_PTR(err);
1812 	}
1813 
1814 	return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1815 }
1816