xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 22f01029)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	IPv6 output functions
4  *	Linux INET6 implementation
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Based on linux/net/ipv4/ip_output.c
10  *
11  *	Changes:
12  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
13  *				extension headers are implemented.
14  *				route changes now work.
15  *				ip6_forward does not confuse sniffers.
16  *				etc.
17  *
18  *      H. von Brand    :       Added missing #include <linux/string.h>
19  *	Imran Patel	:	frag id should be in NBO
20  *      Kazunori MIYAZAWA @USAGI
21  *			:       add ip6_append_data and related functions
22  *				for datagram xmit
23  */
24 
25 #include <linux/errno.h>
26 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/socket.h>
29 #include <linux/net.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/in6.h>
33 #include <linux/tcp.h>
34 #include <linux/route.h>
35 #include <linux/module.h>
36 #include <linux/slab.h>
37 
38 #include <linux/bpf-cgroup.h>
39 #include <linux/netfilter.h>
40 #include <linux/netfilter_ipv6.h>
41 
42 #include <net/sock.h>
43 #include <net/snmp.h>
44 
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 #include <net/protocol.h>
48 #include <net/ip6_route.h>
49 #include <net/addrconf.h>
50 #include <net/rawv6.h>
51 #include <net/icmp.h>
52 #include <net/xfrm.h>
53 #include <net/checksum.h>
54 #include <linux/mroute6.h>
55 #include <net/l3mdev.h>
56 #include <net/lwtunnel.h>
57 #include <net/ip_tunnels.h>
58 
59 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
60 {
61 	struct dst_entry *dst = skb_dst(skb);
62 	struct net_device *dev = dst->dev;
63 	struct inet6_dev *idev = ip6_dst_idev(dst);
64 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
65 	const struct in6_addr *daddr, *nexthop;
66 	struct ipv6hdr *hdr;
67 	struct neighbour *neigh;
68 	int ret;
69 
70 	/* Be paranoid, rather than too clever. */
71 	if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
72 		skb = skb_expand_head(skb, hh_len);
73 		if (!skb) {
74 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
75 			return -ENOMEM;
76 		}
77 	}
78 
79 	hdr = ipv6_hdr(skb);
80 	daddr = &hdr->daddr;
81 	if (ipv6_addr_is_multicast(daddr)) {
82 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
83 		    ((mroute6_is_socket(net, skb) &&
84 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
85 		     ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
86 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
87 
88 			/* Do not check for IFF_ALLMULTI; multicast routing
89 			   is not supported in any case.
90 			 */
91 			if (newskb)
92 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
93 					net, sk, newskb, NULL, newskb->dev,
94 					dev_loopback_xmit);
95 
96 			if (hdr->hop_limit == 0) {
97 				IP6_INC_STATS(net, idev,
98 					      IPSTATS_MIB_OUTDISCARDS);
99 				kfree_skb(skb);
100 				return 0;
101 			}
102 		}
103 
104 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
105 		if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
106 		    !(dev->flags & IFF_LOOPBACK)) {
107 			kfree_skb(skb);
108 			return 0;
109 		}
110 	}
111 
112 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
113 		int res = lwtunnel_xmit(skb);
114 
115 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
116 			return res;
117 	}
118 
119 	rcu_read_lock_bh();
120 	nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
121 	neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
122 	if (unlikely(!neigh))
123 		neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
124 	if (!IS_ERR(neigh)) {
125 		sock_confirm_neigh(skb, neigh);
126 		ret = neigh_output(neigh, skb, false);
127 		rcu_read_unlock_bh();
128 		return ret;
129 	}
130 	rcu_read_unlock_bh();
131 
132 	IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
133 	kfree_skb(skb);
134 	return -EINVAL;
135 }
136 
137 static int
138 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
139 				    struct sk_buff *skb, unsigned int mtu)
140 {
141 	struct sk_buff *segs, *nskb;
142 	netdev_features_t features;
143 	int ret = 0;
144 
145 	/* Please see corresponding comment in ip_finish_output_gso
146 	 * describing the cases where GSO segment length exceeds the
147 	 * egress MTU.
148 	 */
149 	features = netif_skb_features(skb);
150 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
151 	if (IS_ERR_OR_NULL(segs)) {
152 		kfree_skb(skb);
153 		return -ENOMEM;
154 	}
155 
156 	consume_skb(skb);
157 
158 	skb_list_walk_safe(segs, segs, nskb) {
159 		int err;
160 
161 		skb_mark_not_on_list(segs);
162 		err = ip6_fragment(net, sk, segs, ip6_finish_output2);
163 		if (err && ret == 0)
164 			ret = err;
165 	}
166 
167 	return ret;
168 }
169 
170 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
171 {
172 	unsigned int mtu;
173 
174 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
175 	/* Policy lookup after SNAT yielded a new policy */
176 	if (skb_dst(skb)->xfrm) {
177 		IP6CB(skb)->flags |= IP6SKB_REROUTED;
178 		return dst_output(net, sk, skb);
179 	}
180 #endif
181 
182 	mtu = ip6_skb_dst_mtu(skb);
183 	if (skb_is_gso(skb) && !skb_gso_validate_network_len(skb, mtu))
184 		return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
185 
186 	if ((skb->len > mtu && !skb_is_gso(skb)) ||
187 	    dst_allfrag(skb_dst(skb)) ||
188 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
189 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
190 	else
191 		return ip6_finish_output2(net, sk, skb);
192 }
193 
194 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
195 {
196 	int ret;
197 
198 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
199 	switch (ret) {
200 	case NET_XMIT_SUCCESS:
201 		return __ip6_finish_output(net, sk, skb);
202 	case NET_XMIT_CN:
203 		return __ip6_finish_output(net, sk, skb) ? : ret;
204 	default:
205 		kfree_skb(skb);
206 		return ret;
207 	}
208 }
209 
210 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
211 {
212 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
213 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
214 
215 	skb->protocol = htons(ETH_P_IPV6);
216 	skb->dev = dev;
217 
218 	if (unlikely(idev->cnf.disable_ipv6)) {
219 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
220 		kfree_skb(skb);
221 		return 0;
222 	}
223 
224 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
225 			    net, sk, skb, indev, dev,
226 			    ip6_finish_output,
227 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
228 }
229 EXPORT_SYMBOL(ip6_output);
230 
231 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
232 {
233 	if (!np->autoflowlabel_set)
234 		return ip6_default_np_autolabel(net);
235 	else
236 		return np->autoflowlabel;
237 }
238 
239 /*
240  * xmit an sk_buff (used by TCP, SCTP and DCCP)
241  * Note : socket lock is not held for SYNACK packets, but might be modified
242  * by calls to skb_set_owner_w() and ipv6_local_error(),
243  * which are using proper atomic operations or spinlocks.
244  */
245 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
246 	     __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
247 {
248 	struct net *net = sock_net(sk);
249 	const struct ipv6_pinfo *np = inet6_sk(sk);
250 	struct in6_addr *first_hop = &fl6->daddr;
251 	struct dst_entry *dst = skb_dst(skb);
252 	struct net_device *dev = dst->dev;
253 	struct inet6_dev *idev = ip6_dst_idev(dst);
254 	unsigned int head_room;
255 	struct ipv6hdr *hdr;
256 	u8  proto = fl6->flowi6_proto;
257 	int seg_len = skb->len;
258 	int hlimit = -1;
259 	u32 mtu;
260 
261 	head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dev);
262 	if (opt)
263 		head_room += opt->opt_nflen + opt->opt_flen;
264 
265 	if (unlikely(head_room > skb_headroom(skb))) {
266 		skb = skb_expand_head(skb, head_room);
267 		if (!skb) {
268 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
269 			return -ENOBUFS;
270 		}
271 	}
272 
273 	if (opt) {
274 		seg_len += opt->opt_nflen + opt->opt_flen;
275 
276 		if (opt->opt_flen)
277 			ipv6_push_frag_opts(skb, opt, &proto);
278 
279 		if (opt->opt_nflen)
280 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
281 					     &fl6->saddr);
282 	}
283 
284 	skb_push(skb, sizeof(struct ipv6hdr));
285 	skb_reset_network_header(skb);
286 	hdr = ipv6_hdr(skb);
287 
288 	/*
289 	 *	Fill in the IPv6 header
290 	 */
291 	if (np)
292 		hlimit = np->hop_limit;
293 	if (hlimit < 0)
294 		hlimit = ip6_dst_hoplimit(dst);
295 
296 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
297 				ip6_autoflowlabel(net, np), fl6));
298 
299 	hdr->payload_len = htons(seg_len);
300 	hdr->nexthdr = proto;
301 	hdr->hop_limit = hlimit;
302 
303 	hdr->saddr = fl6->saddr;
304 	hdr->daddr = *first_hop;
305 
306 	skb->protocol = htons(ETH_P_IPV6);
307 	skb->priority = priority;
308 	skb->mark = mark;
309 
310 	mtu = dst_mtu(dst);
311 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
312 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
313 
314 		/* if egress device is enslaved to an L3 master device pass the
315 		 * skb to its handler for processing
316 		 */
317 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
318 		if (unlikely(!skb))
319 			return 0;
320 
321 		/* hooks should never assume socket lock is held.
322 		 * we promote our socket to non const
323 		 */
324 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
325 			       net, (struct sock *)sk, skb, NULL, dev,
326 			       dst_output);
327 	}
328 
329 	skb->dev = dev;
330 	/* ipv6_local_error() does not require socket lock,
331 	 * we promote our socket to non const
332 	 */
333 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
334 
335 	IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
336 	kfree_skb(skb);
337 	return -EMSGSIZE;
338 }
339 EXPORT_SYMBOL(ip6_xmit);
340 
341 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
342 {
343 	struct ip6_ra_chain *ra;
344 	struct sock *last = NULL;
345 
346 	read_lock(&ip6_ra_lock);
347 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
348 		struct sock *sk = ra->sk;
349 		if (sk && ra->sel == sel &&
350 		    (!sk->sk_bound_dev_if ||
351 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
352 			struct ipv6_pinfo *np = inet6_sk(sk);
353 
354 			if (np && np->rtalert_isolate &&
355 			    !net_eq(sock_net(sk), dev_net(skb->dev))) {
356 				continue;
357 			}
358 			if (last) {
359 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
360 				if (skb2)
361 					rawv6_rcv(last, skb2);
362 			}
363 			last = sk;
364 		}
365 	}
366 
367 	if (last) {
368 		rawv6_rcv(last, skb);
369 		read_unlock(&ip6_ra_lock);
370 		return 1;
371 	}
372 	read_unlock(&ip6_ra_lock);
373 	return 0;
374 }
375 
376 static int ip6_forward_proxy_check(struct sk_buff *skb)
377 {
378 	struct ipv6hdr *hdr = ipv6_hdr(skb);
379 	u8 nexthdr = hdr->nexthdr;
380 	__be16 frag_off;
381 	int offset;
382 
383 	if (ipv6_ext_hdr(nexthdr)) {
384 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
385 		if (offset < 0)
386 			return 0;
387 	} else
388 		offset = sizeof(struct ipv6hdr);
389 
390 	if (nexthdr == IPPROTO_ICMPV6) {
391 		struct icmp6hdr *icmp6;
392 
393 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
394 					 offset + 1 - skb->data)))
395 			return 0;
396 
397 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
398 
399 		switch (icmp6->icmp6_type) {
400 		case NDISC_ROUTER_SOLICITATION:
401 		case NDISC_ROUTER_ADVERTISEMENT:
402 		case NDISC_NEIGHBOUR_SOLICITATION:
403 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
404 		case NDISC_REDIRECT:
405 			/* For reaction involving unicast neighbor discovery
406 			 * message destined to the proxied address, pass it to
407 			 * input function.
408 			 */
409 			return 1;
410 		default:
411 			break;
412 		}
413 	}
414 
415 	/*
416 	 * The proxying router can't forward traffic sent to a link-local
417 	 * address, so signal the sender and discard the packet. This
418 	 * behavior is clarified by the MIPv6 specification.
419 	 */
420 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
421 		dst_link_failure(skb);
422 		return -1;
423 	}
424 
425 	return 0;
426 }
427 
428 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
429 				     struct sk_buff *skb)
430 {
431 	struct dst_entry *dst = skb_dst(skb);
432 
433 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
434 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
435 
436 #ifdef CONFIG_NET_SWITCHDEV
437 	if (skb->offload_l3_fwd_mark) {
438 		consume_skb(skb);
439 		return 0;
440 	}
441 #endif
442 
443 	skb->tstamp = 0;
444 	return dst_output(net, sk, skb);
445 }
446 
447 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
448 {
449 	if (skb->len <= mtu)
450 		return false;
451 
452 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
453 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
454 		return true;
455 
456 	if (skb->ignore_df)
457 		return false;
458 
459 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
460 		return false;
461 
462 	return true;
463 }
464 
465 int ip6_forward(struct sk_buff *skb)
466 {
467 	struct dst_entry *dst = skb_dst(skb);
468 	struct ipv6hdr *hdr = ipv6_hdr(skb);
469 	struct inet6_skb_parm *opt = IP6CB(skb);
470 	struct net *net = dev_net(dst->dev);
471 	struct inet6_dev *idev;
472 	u32 mtu;
473 
474 	idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
475 	if (net->ipv6.devconf_all->forwarding == 0)
476 		goto error;
477 
478 	if (skb->pkt_type != PACKET_HOST)
479 		goto drop;
480 
481 	if (unlikely(skb->sk))
482 		goto drop;
483 
484 	if (skb_warn_if_lro(skb))
485 		goto drop;
486 
487 	if (!net->ipv6.devconf_all->disable_policy &&
488 	    !idev->cnf.disable_policy &&
489 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
490 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
491 		goto drop;
492 	}
493 
494 	skb_forward_csum(skb);
495 
496 	/*
497 	 *	We DO NOT make any processing on
498 	 *	RA packets, pushing them to user level AS IS
499 	 *	without ane WARRANTY that application will be able
500 	 *	to interpret them. The reason is that we
501 	 *	cannot make anything clever here.
502 	 *
503 	 *	We are not end-node, so that if packet contains
504 	 *	AH/ESP, we cannot make anything.
505 	 *	Defragmentation also would be mistake, RA packets
506 	 *	cannot be fragmented, because there is no warranty
507 	 *	that different fragments will go along one path. --ANK
508 	 */
509 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
510 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
511 			return 0;
512 	}
513 
514 	/*
515 	 *	check and decrement ttl
516 	 */
517 	if (hdr->hop_limit <= 1) {
518 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
519 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
520 
521 		kfree_skb(skb);
522 		return -ETIMEDOUT;
523 	}
524 
525 	/* XXX: idev->cnf.proxy_ndp? */
526 	if (net->ipv6.devconf_all->proxy_ndp &&
527 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
528 		int proxied = ip6_forward_proxy_check(skb);
529 		if (proxied > 0) {
530 			hdr->hop_limit--;
531 			return ip6_input(skb);
532 		} else if (proxied < 0) {
533 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
534 			goto drop;
535 		}
536 	}
537 
538 	if (!xfrm6_route_forward(skb)) {
539 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
540 		goto drop;
541 	}
542 	dst = skb_dst(skb);
543 
544 	/* IPv6 specs say nothing about it, but it is clear that we cannot
545 	   send redirects to source routed frames.
546 	   We don't send redirects to frames decapsulated from IPsec.
547 	 */
548 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
549 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
550 		struct in6_addr *target = NULL;
551 		struct inet_peer *peer;
552 		struct rt6_info *rt;
553 
554 		/*
555 		 *	incoming and outgoing devices are the same
556 		 *	send a redirect.
557 		 */
558 
559 		rt = (struct rt6_info *) dst;
560 		if (rt->rt6i_flags & RTF_GATEWAY)
561 			target = &rt->rt6i_gateway;
562 		else
563 			target = &hdr->daddr;
564 
565 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
566 
567 		/* Limit redirects both by destination (here)
568 		   and by source (inside ndisc_send_redirect)
569 		 */
570 		if (inet_peer_xrlim_allow(peer, 1*HZ))
571 			ndisc_send_redirect(skb, target);
572 		if (peer)
573 			inet_putpeer(peer);
574 	} else {
575 		int addrtype = ipv6_addr_type(&hdr->saddr);
576 
577 		/* This check is security critical. */
578 		if (addrtype == IPV6_ADDR_ANY ||
579 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
580 			goto error;
581 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
582 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
583 				    ICMPV6_NOT_NEIGHBOUR, 0);
584 			goto error;
585 		}
586 	}
587 
588 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
589 	if (mtu < IPV6_MIN_MTU)
590 		mtu = IPV6_MIN_MTU;
591 
592 	if (ip6_pkt_too_big(skb, mtu)) {
593 		/* Again, force OUTPUT device used as source address */
594 		skb->dev = dst->dev;
595 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
596 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
597 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
598 				IPSTATS_MIB_FRAGFAILS);
599 		kfree_skb(skb);
600 		return -EMSGSIZE;
601 	}
602 
603 	if (skb_cow(skb, dst->dev->hard_header_len)) {
604 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
605 				IPSTATS_MIB_OUTDISCARDS);
606 		goto drop;
607 	}
608 
609 	hdr = ipv6_hdr(skb);
610 
611 	/* Mangling hops number delayed to point after skb COW */
612 
613 	hdr->hop_limit--;
614 
615 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
616 		       net, NULL, skb, skb->dev, dst->dev,
617 		       ip6_forward_finish);
618 
619 error:
620 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
621 drop:
622 	kfree_skb(skb);
623 	return -EINVAL;
624 }
625 
626 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
627 {
628 	to->pkt_type = from->pkt_type;
629 	to->priority = from->priority;
630 	to->protocol = from->protocol;
631 	skb_dst_drop(to);
632 	skb_dst_set(to, dst_clone(skb_dst(from)));
633 	to->dev = from->dev;
634 	to->mark = from->mark;
635 
636 	skb_copy_hash(to, from);
637 
638 #ifdef CONFIG_NET_SCHED
639 	to->tc_index = from->tc_index;
640 #endif
641 	nf_copy(to, from);
642 	skb_ext_copy(to, from);
643 	skb_copy_secmark(to, from);
644 }
645 
646 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
647 		      u8 nexthdr, __be32 frag_id,
648 		      struct ip6_fraglist_iter *iter)
649 {
650 	unsigned int first_len;
651 	struct frag_hdr *fh;
652 
653 	/* BUILD HEADER */
654 	*prevhdr = NEXTHDR_FRAGMENT;
655 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
656 	if (!iter->tmp_hdr)
657 		return -ENOMEM;
658 
659 	iter->frag = skb_shinfo(skb)->frag_list;
660 	skb_frag_list_init(skb);
661 
662 	iter->offset = 0;
663 	iter->hlen = hlen;
664 	iter->frag_id = frag_id;
665 	iter->nexthdr = nexthdr;
666 
667 	__skb_pull(skb, hlen);
668 	fh = __skb_push(skb, sizeof(struct frag_hdr));
669 	__skb_push(skb, hlen);
670 	skb_reset_network_header(skb);
671 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
672 
673 	fh->nexthdr = nexthdr;
674 	fh->reserved = 0;
675 	fh->frag_off = htons(IP6_MF);
676 	fh->identification = frag_id;
677 
678 	first_len = skb_pagelen(skb);
679 	skb->data_len = first_len - skb_headlen(skb);
680 	skb->len = first_len;
681 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
682 
683 	return 0;
684 }
685 EXPORT_SYMBOL(ip6_fraglist_init);
686 
687 void ip6_fraglist_prepare(struct sk_buff *skb,
688 			  struct ip6_fraglist_iter *iter)
689 {
690 	struct sk_buff *frag = iter->frag;
691 	unsigned int hlen = iter->hlen;
692 	struct frag_hdr *fh;
693 
694 	frag->ip_summed = CHECKSUM_NONE;
695 	skb_reset_transport_header(frag);
696 	fh = __skb_push(frag, sizeof(struct frag_hdr));
697 	__skb_push(frag, hlen);
698 	skb_reset_network_header(frag);
699 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
700 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
701 	fh->nexthdr = iter->nexthdr;
702 	fh->reserved = 0;
703 	fh->frag_off = htons(iter->offset);
704 	if (frag->next)
705 		fh->frag_off |= htons(IP6_MF);
706 	fh->identification = iter->frag_id;
707 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
708 	ip6_copy_metadata(frag, skb);
709 }
710 EXPORT_SYMBOL(ip6_fraglist_prepare);
711 
712 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
713 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
714 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
715 {
716 	state->prevhdr = prevhdr;
717 	state->nexthdr = nexthdr;
718 	state->frag_id = frag_id;
719 
720 	state->hlen = hlen;
721 	state->mtu = mtu;
722 
723 	state->left = skb->len - hlen;	/* Space per frame */
724 	state->ptr = hlen;		/* Where to start from */
725 
726 	state->hroom = hdr_room;
727 	state->troom = needed_tailroom;
728 
729 	state->offset = 0;
730 }
731 EXPORT_SYMBOL(ip6_frag_init);
732 
733 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
734 {
735 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
736 	struct sk_buff *frag;
737 	struct frag_hdr *fh;
738 	unsigned int len;
739 
740 	len = state->left;
741 	/* IF: it doesn't fit, use 'mtu' - the data space left */
742 	if (len > state->mtu)
743 		len = state->mtu;
744 	/* IF: we are not sending up to and including the packet end
745 	   then align the next start on an eight byte boundary */
746 	if (len < state->left)
747 		len &= ~7;
748 
749 	/* Allocate buffer */
750 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
751 			 state->hroom + state->troom, GFP_ATOMIC);
752 	if (!frag)
753 		return ERR_PTR(-ENOMEM);
754 
755 	/*
756 	 *	Set up data on packet
757 	 */
758 
759 	ip6_copy_metadata(frag, skb);
760 	skb_reserve(frag, state->hroom);
761 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
762 	skb_reset_network_header(frag);
763 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
764 	frag->transport_header = (frag->network_header + state->hlen +
765 				  sizeof(struct frag_hdr));
766 
767 	/*
768 	 *	Charge the memory for the fragment to any owner
769 	 *	it might possess
770 	 */
771 	if (skb->sk)
772 		skb_set_owner_w(frag, skb->sk);
773 
774 	/*
775 	 *	Copy the packet header into the new buffer.
776 	 */
777 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
778 
779 	fragnexthdr_offset = skb_network_header(frag);
780 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
781 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
782 
783 	/*
784 	 *	Build fragment header.
785 	 */
786 	fh->nexthdr = state->nexthdr;
787 	fh->reserved = 0;
788 	fh->identification = state->frag_id;
789 
790 	/*
791 	 *	Copy a block of the IP datagram.
792 	 */
793 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
794 			     len));
795 	state->left -= len;
796 
797 	fh->frag_off = htons(state->offset);
798 	if (state->left > 0)
799 		fh->frag_off |= htons(IP6_MF);
800 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
801 
802 	state->ptr += len;
803 	state->offset += len;
804 
805 	return frag;
806 }
807 EXPORT_SYMBOL(ip6_frag_next);
808 
809 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
810 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
811 {
812 	struct sk_buff *frag;
813 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
814 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
815 				inet6_sk(skb->sk) : NULL;
816 	struct ip6_frag_state state;
817 	unsigned int mtu, hlen, nexthdr_offset;
818 	ktime_t tstamp = skb->tstamp;
819 	int hroom, err = 0;
820 	__be32 frag_id;
821 	u8 *prevhdr, nexthdr = 0;
822 
823 	err = ip6_find_1stfragopt(skb, &prevhdr);
824 	if (err < 0)
825 		goto fail;
826 	hlen = err;
827 	nexthdr = *prevhdr;
828 	nexthdr_offset = prevhdr - skb_network_header(skb);
829 
830 	mtu = ip6_skb_dst_mtu(skb);
831 
832 	/* We must not fragment if the socket is set to force MTU discovery
833 	 * or if the skb it not generated by a local socket.
834 	 */
835 	if (unlikely(!skb->ignore_df && skb->len > mtu))
836 		goto fail_toobig;
837 
838 	if (IP6CB(skb)->frag_max_size) {
839 		if (IP6CB(skb)->frag_max_size > mtu)
840 			goto fail_toobig;
841 
842 		/* don't send fragments larger than what we received */
843 		mtu = IP6CB(skb)->frag_max_size;
844 		if (mtu < IPV6_MIN_MTU)
845 			mtu = IPV6_MIN_MTU;
846 	}
847 
848 	if (np && np->frag_size < mtu) {
849 		if (np->frag_size)
850 			mtu = np->frag_size;
851 	}
852 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
853 		goto fail_toobig;
854 	mtu -= hlen + sizeof(struct frag_hdr);
855 
856 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
857 				    &ipv6_hdr(skb)->saddr);
858 
859 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
860 	    (err = skb_checksum_help(skb)))
861 		goto fail;
862 
863 	prevhdr = skb_network_header(skb) + nexthdr_offset;
864 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
865 	if (skb_has_frag_list(skb)) {
866 		unsigned int first_len = skb_pagelen(skb);
867 		struct ip6_fraglist_iter iter;
868 		struct sk_buff *frag2;
869 
870 		if (first_len - hlen > mtu ||
871 		    ((first_len - hlen) & 7) ||
872 		    skb_cloned(skb) ||
873 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
874 			goto slow_path;
875 
876 		skb_walk_frags(skb, frag) {
877 			/* Correct geometry. */
878 			if (frag->len > mtu ||
879 			    ((frag->len & 7) && frag->next) ||
880 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
881 				goto slow_path_clean;
882 
883 			/* Partially cloned skb? */
884 			if (skb_shared(frag))
885 				goto slow_path_clean;
886 
887 			BUG_ON(frag->sk);
888 			if (skb->sk) {
889 				frag->sk = skb->sk;
890 				frag->destructor = sock_wfree;
891 			}
892 			skb->truesize -= frag->truesize;
893 		}
894 
895 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
896 					&iter);
897 		if (err < 0)
898 			goto fail;
899 
900 		for (;;) {
901 			/* Prepare header of the next frame,
902 			 * before previous one went down. */
903 			if (iter.frag)
904 				ip6_fraglist_prepare(skb, &iter);
905 
906 			skb->tstamp = tstamp;
907 			err = output(net, sk, skb);
908 			if (!err)
909 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
910 					      IPSTATS_MIB_FRAGCREATES);
911 
912 			if (err || !iter.frag)
913 				break;
914 
915 			skb = ip6_fraglist_next(&iter);
916 		}
917 
918 		kfree(iter.tmp_hdr);
919 
920 		if (err == 0) {
921 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
922 				      IPSTATS_MIB_FRAGOKS);
923 			return 0;
924 		}
925 
926 		kfree_skb_list(iter.frag);
927 
928 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
929 			      IPSTATS_MIB_FRAGFAILS);
930 		return err;
931 
932 slow_path_clean:
933 		skb_walk_frags(skb, frag2) {
934 			if (frag2 == frag)
935 				break;
936 			frag2->sk = NULL;
937 			frag2->destructor = NULL;
938 			skb->truesize += frag2->truesize;
939 		}
940 	}
941 
942 slow_path:
943 	/*
944 	 *	Fragment the datagram.
945 	 */
946 
947 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
948 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
949 		      &state);
950 
951 	/*
952 	 *	Keep copying data until we run out.
953 	 */
954 
955 	while (state.left > 0) {
956 		frag = ip6_frag_next(skb, &state);
957 		if (IS_ERR(frag)) {
958 			err = PTR_ERR(frag);
959 			goto fail;
960 		}
961 
962 		/*
963 		 *	Put this fragment into the sending queue.
964 		 */
965 		frag->tstamp = tstamp;
966 		err = output(net, sk, frag);
967 		if (err)
968 			goto fail;
969 
970 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
971 			      IPSTATS_MIB_FRAGCREATES);
972 	}
973 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
974 		      IPSTATS_MIB_FRAGOKS);
975 	consume_skb(skb);
976 	return err;
977 
978 fail_toobig:
979 	if (skb->sk && dst_allfrag(skb_dst(skb)))
980 		sk_gso_disable(skb->sk);
981 
982 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
983 	err = -EMSGSIZE;
984 
985 fail:
986 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
987 		      IPSTATS_MIB_FRAGFAILS);
988 	kfree_skb(skb);
989 	return err;
990 }
991 
992 static inline int ip6_rt_check(const struct rt6key *rt_key,
993 			       const struct in6_addr *fl_addr,
994 			       const struct in6_addr *addr_cache)
995 {
996 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
997 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
998 }
999 
1000 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1001 					  struct dst_entry *dst,
1002 					  const struct flowi6 *fl6)
1003 {
1004 	struct ipv6_pinfo *np = inet6_sk(sk);
1005 	struct rt6_info *rt;
1006 
1007 	if (!dst)
1008 		goto out;
1009 
1010 	if (dst->ops->family != AF_INET6) {
1011 		dst_release(dst);
1012 		return NULL;
1013 	}
1014 
1015 	rt = (struct rt6_info *)dst;
1016 	/* Yes, checking route validity in not connected
1017 	 * case is not very simple. Take into account,
1018 	 * that we do not support routing by source, TOS,
1019 	 * and MSG_DONTROUTE		--ANK (980726)
1020 	 *
1021 	 * 1. ip6_rt_check(): If route was host route,
1022 	 *    check that cached destination is current.
1023 	 *    If it is network route, we still may
1024 	 *    check its validity using saved pointer
1025 	 *    to the last used address: daddr_cache.
1026 	 *    We do not want to save whole address now,
1027 	 *    (because main consumer of this service
1028 	 *    is tcp, which has not this problem),
1029 	 *    so that the last trick works only on connected
1030 	 *    sockets.
1031 	 * 2. oif also should be the same.
1032 	 */
1033 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1034 #ifdef CONFIG_IPV6_SUBTREES
1035 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1036 #endif
1037 	   (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
1038 	      (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
1039 		dst_release(dst);
1040 		dst = NULL;
1041 	}
1042 
1043 out:
1044 	return dst;
1045 }
1046 
1047 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1048 			       struct dst_entry **dst, struct flowi6 *fl6)
1049 {
1050 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1051 	struct neighbour *n;
1052 	struct rt6_info *rt;
1053 #endif
1054 	int err;
1055 	int flags = 0;
1056 
1057 	/* The correct way to handle this would be to do
1058 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1059 	 * the route-specific preferred source forces the
1060 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1061 	 *
1062 	 * In source specific routing (no src=any default route),
1063 	 * ip6_route_output will fail given src=any saddr, though, so
1064 	 * that's why we try it again later.
1065 	 */
1066 	if (ipv6_addr_any(&fl6->saddr)) {
1067 		struct fib6_info *from;
1068 		struct rt6_info *rt;
1069 
1070 		*dst = ip6_route_output(net, sk, fl6);
1071 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1072 
1073 		rcu_read_lock();
1074 		from = rt ? rcu_dereference(rt->from) : NULL;
1075 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1076 					  sk ? inet6_sk(sk)->srcprefs : 0,
1077 					  &fl6->saddr);
1078 		rcu_read_unlock();
1079 
1080 		if (err)
1081 			goto out_err_release;
1082 
1083 		/* If we had an erroneous initial result, pretend it
1084 		 * never existed and let the SA-enabled version take
1085 		 * over.
1086 		 */
1087 		if ((*dst)->error) {
1088 			dst_release(*dst);
1089 			*dst = NULL;
1090 		}
1091 
1092 		if (fl6->flowi6_oif)
1093 			flags |= RT6_LOOKUP_F_IFACE;
1094 	}
1095 
1096 	if (!*dst)
1097 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1098 
1099 	err = (*dst)->error;
1100 	if (err)
1101 		goto out_err_release;
1102 
1103 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1104 	/*
1105 	 * Here if the dst entry we've looked up
1106 	 * has a neighbour entry that is in the INCOMPLETE
1107 	 * state and the src address from the flow is
1108 	 * marked as OPTIMISTIC, we release the found
1109 	 * dst entry and replace it instead with the
1110 	 * dst entry of the nexthop router
1111 	 */
1112 	rt = (struct rt6_info *) *dst;
1113 	rcu_read_lock_bh();
1114 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1115 				      rt6_nexthop(rt, &fl6->daddr));
1116 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1117 	rcu_read_unlock_bh();
1118 
1119 	if (err) {
1120 		struct inet6_ifaddr *ifp;
1121 		struct flowi6 fl_gw6;
1122 		int redirect;
1123 
1124 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1125 				      (*dst)->dev, 1);
1126 
1127 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1128 		if (ifp)
1129 			in6_ifa_put(ifp);
1130 
1131 		if (redirect) {
1132 			/*
1133 			 * We need to get the dst entry for the
1134 			 * default router instead
1135 			 */
1136 			dst_release(*dst);
1137 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1138 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1139 			*dst = ip6_route_output(net, sk, &fl_gw6);
1140 			err = (*dst)->error;
1141 			if (err)
1142 				goto out_err_release;
1143 		}
1144 	}
1145 #endif
1146 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1147 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1148 		err = -EAFNOSUPPORT;
1149 		goto out_err_release;
1150 	}
1151 
1152 	return 0;
1153 
1154 out_err_release:
1155 	dst_release(*dst);
1156 	*dst = NULL;
1157 
1158 	if (err == -ENETUNREACH)
1159 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1160 	return err;
1161 }
1162 
1163 /**
1164  *	ip6_dst_lookup - perform route lookup on flow
1165  *	@net: Network namespace to perform lookup in
1166  *	@sk: socket which provides route info
1167  *	@dst: pointer to dst_entry * for result
1168  *	@fl6: flow to lookup
1169  *
1170  *	This function performs a route lookup on the given flow.
1171  *
1172  *	It returns zero on success, or a standard errno code on error.
1173  */
1174 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1175 		   struct flowi6 *fl6)
1176 {
1177 	*dst = NULL;
1178 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1179 }
1180 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1181 
1182 /**
1183  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1184  *	@net: Network namespace to perform lookup in
1185  *	@sk: socket which provides route info
1186  *	@fl6: flow to lookup
1187  *	@final_dst: final destination address for ipsec lookup
1188  *
1189  *	This function performs a route lookup on the given flow.
1190  *
1191  *	It returns a valid dst pointer on success, or a pointer encoded
1192  *	error code.
1193  */
1194 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1195 				      const struct in6_addr *final_dst)
1196 {
1197 	struct dst_entry *dst = NULL;
1198 	int err;
1199 
1200 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1201 	if (err)
1202 		return ERR_PTR(err);
1203 	if (final_dst)
1204 		fl6->daddr = *final_dst;
1205 
1206 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1207 }
1208 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1209 
1210 /**
1211  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1212  *	@sk: socket which provides the dst cache and route info
1213  *	@fl6: flow to lookup
1214  *	@final_dst: final destination address for ipsec lookup
1215  *	@connected: whether @sk is connected or not
1216  *
1217  *	This function performs a route lookup on the given flow with the
1218  *	possibility of using the cached route in the socket if it is valid.
1219  *	It will take the socket dst lock when operating on the dst cache.
1220  *	As a result, this function can only be used in process context.
1221  *
1222  *	In addition, for a connected socket, cache the dst in the socket
1223  *	if the current cache is not valid.
1224  *
1225  *	It returns a valid dst pointer on success, or a pointer encoded
1226  *	error code.
1227  */
1228 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1229 					 const struct in6_addr *final_dst,
1230 					 bool connected)
1231 {
1232 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1233 
1234 	dst = ip6_sk_dst_check(sk, dst, fl6);
1235 	if (dst)
1236 		return dst;
1237 
1238 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1239 	if (connected && !IS_ERR(dst))
1240 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1241 
1242 	return dst;
1243 }
1244 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1245 
1246 /**
1247  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1248  *      @skb: Packet for which lookup is done
1249  *      @dev: Tunnel device
1250  *      @net: Network namespace of tunnel device
1251  *      @sock: Socket which provides route info
1252  *      @saddr: Memory to store the src ip address
1253  *      @info: Tunnel information
1254  *      @protocol: IP protocol
1255  *      @use_cache: Flag to enable cache usage
1256  *      This function performs a route lookup on a tunnel
1257  *
1258  *      It returns a valid dst pointer and stores src address to be used in
1259  *      tunnel in param saddr on success, else a pointer encoded error code.
1260  */
1261 
1262 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1263 					struct net_device *dev,
1264 					struct net *net,
1265 					struct socket *sock,
1266 					struct in6_addr *saddr,
1267 					const struct ip_tunnel_info *info,
1268 					u8 protocol,
1269 					bool use_cache)
1270 {
1271 	struct dst_entry *dst = NULL;
1272 #ifdef CONFIG_DST_CACHE
1273 	struct dst_cache *dst_cache;
1274 #endif
1275 	struct flowi6 fl6;
1276 	__u8 prio;
1277 
1278 #ifdef CONFIG_DST_CACHE
1279 	dst_cache = (struct dst_cache *)&info->dst_cache;
1280 	if (use_cache) {
1281 		dst = dst_cache_get_ip6(dst_cache, saddr);
1282 		if (dst)
1283 			return dst;
1284 	}
1285 #endif
1286 	memset(&fl6, 0, sizeof(fl6));
1287 	fl6.flowi6_mark = skb->mark;
1288 	fl6.flowi6_proto = protocol;
1289 	fl6.daddr = info->key.u.ipv6.dst;
1290 	fl6.saddr = info->key.u.ipv6.src;
1291 	prio = info->key.tos;
1292 	fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1293 					  info->key.label);
1294 
1295 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1296 					      NULL);
1297 	if (IS_ERR(dst)) {
1298 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1299 		return ERR_PTR(-ENETUNREACH);
1300 	}
1301 	if (dst->dev == dev) { /* is this necessary? */
1302 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1303 		dst_release(dst);
1304 		return ERR_PTR(-ELOOP);
1305 	}
1306 #ifdef CONFIG_DST_CACHE
1307 	if (use_cache)
1308 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1309 #endif
1310 	*saddr = fl6.saddr;
1311 	return dst;
1312 }
1313 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1314 
1315 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1316 					       gfp_t gfp)
1317 {
1318 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1319 }
1320 
1321 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1322 						gfp_t gfp)
1323 {
1324 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1325 }
1326 
1327 static void ip6_append_data_mtu(unsigned int *mtu,
1328 				int *maxfraglen,
1329 				unsigned int fragheaderlen,
1330 				struct sk_buff *skb,
1331 				struct rt6_info *rt,
1332 				unsigned int orig_mtu)
1333 {
1334 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1335 		if (!skb) {
1336 			/* first fragment, reserve header_len */
1337 			*mtu = orig_mtu - rt->dst.header_len;
1338 
1339 		} else {
1340 			/*
1341 			 * this fragment is not first, the headers
1342 			 * space is regarded as data space.
1343 			 */
1344 			*mtu = orig_mtu;
1345 		}
1346 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1347 			      + fragheaderlen - sizeof(struct frag_hdr);
1348 	}
1349 }
1350 
1351 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1352 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1353 			  struct rt6_info *rt, struct flowi6 *fl6)
1354 {
1355 	struct ipv6_pinfo *np = inet6_sk(sk);
1356 	unsigned int mtu;
1357 	struct ipv6_txoptions *opt = ipc6->opt;
1358 
1359 	/*
1360 	 * setup for corking
1361 	 */
1362 	if (opt) {
1363 		if (WARN_ON(v6_cork->opt))
1364 			return -EINVAL;
1365 
1366 		v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1367 		if (unlikely(!v6_cork->opt))
1368 			return -ENOBUFS;
1369 
1370 		v6_cork->opt->tot_len = sizeof(*opt);
1371 		v6_cork->opt->opt_flen = opt->opt_flen;
1372 		v6_cork->opt->opt_nflen = opt->opt_nflen;
1373 
1374 		v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1375 						    sk->sk_allocation);
1376 		if (opt->dst0opt && !v6_cork->opt->dst0opt)
1377 			return -ENOBUFS;
1378 
1379 		v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1380 						    sk->sk_allocation);
1381 		if (opt->dst1opt && !v6_cork->opt->dst1opt)
1382 			return -ENOBUFS;
1383 
1384 		v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1385 						   sk->sk_allocation);
1386 		if (opt->hopopt && !v6_cork->opt->hopopt)
1387 			return -ENOBUFS;
1388 
1389 		v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1390 						    sk->sk_allocation);
1391 		if (opt->srcrt && !v6_cork->opt->srcrt)
1392 			return -ENOBUFS;
1393 
1394 		/* need source address above miyazawa*/
1395 	}
1396 	dst_hold(&rt->dst);
1397 	cork->base.dst = &rt->dst;
1398 	cork->fl.u.ip6 = *fl6;
1399 	v6_cork->hop_limit = ipc6->hlimit;
1400 	v6_cork->tclass = ipc6->tclass;
1401 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1402 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1403 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1404 	else
1405 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1406 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1407 	if (np->frag_size < mtu) {
1408 		if (np->frag_size)
1409 			mtu = np->frag_size;
1410 	}
1411 	if (mtu < IPV6_MIN_MTU)
1412 		return -EINVAL;
1413 	cork->base.fragsize = mtu;
1414 	cork->base.gso_size = ipc6->gso_size;
1415 	cork->base.tx_flags = 0;
1416 	cork->base.mark = ipc6->sockc.mark;
1417 	sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1418 
1419 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1420 		cork->base.flags |= IPCORK_ALLFRAG;
1421 	cork->base.length = 0;
1422 
1423 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1424 
1425 	return 0;
1426 }
1427 
1428 static int __ip6_append_data(struct sock *sk,
1429 			     struct flowi6 *fl6,
1430 			     struct sk_buff_head *queue,
1431 			     struct inet_cork *cork,
1432 			     struct inet6_cork *v6_cork,
1433 			     struct page_frag *pfrag,
1434 			     int getfrag(void *from, char *to, int offset,
1435 					 int len, int odd, struct sk_buff *skb),
1436 			     void *from, int length, int transhdrlen,
1437 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1438 {
1439 	struct sk_buff *skb, *skb_prev = NULL;
1440 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1441 	struct ubuf_info *uarg = NULL;
1442 	int exthdrlen = 0;
1443 	int dst_exthdrlen = 0;
1444 	int hh_len;
1445 	int copy;
1446 	int err;
1447 	int offset = 0;
1448 	u32 tskey = 0;
1449 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1450 	struct ipv6_txoptions *opt = v6_cork->opt;
1451 	int csummode = CHECKSUM_NONE;
1452 	unsigned int maxnonfragsize, headersize;
1453 	unsigned int wmem_alloc_delta = 0;
1454 	bool paged, extra_uref = false;
1455 
1456 	skb = skb_peek_tail(queue);
1457 	if (!skb) {
1458 		exthdrlen = opt ? opt->opt_flen : 0;
1459 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1460 	}
1461 
1462 	paged = !!cork->gso_size;
1463 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1464 	orig_mtu = mtu;
1465 
1466 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1467 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1468 		tskey = sk->sk_tskey++;
1469 
1470 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1471 
1472 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1473 			(opt ? opt->opt_nflen : 0);
1474 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1475 		     sizeof(struct frag_hdr);
1476 
1477 	headersize = sizeof(struct ipv6hdr) +
1478 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1479 		     (dst_allfrag(&rt->dst) ?
1480 		      sizeof(struct frag_hdr) : 0) +
1481 		     rt->rt6i_nfheader_len;
1482 
1483 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1484 	 * the first fragment
1485 	 */
1486 	if (headersize + transhdrlen > mtu)
1487 		goto emsgsize;
1488 
1489 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1490 	    (sk->sk_protocol == IPPROTO_UDP ||
1491 	     sk->sk_protocol == IPPROTO_RAW)) {
1492 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1493 				sizeof(struct ipv6hdr));
1494 		goto emsgsize;
1495 	}
1496 
1497 	if (ip6_sk_ignore_df(sk))
1498 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1499 	else
1500 		maxnonfragsize = mtu;
1501 
1502 	if (cork->length + length > maxnonfragsize - headersize) {
1503 emsgsize:
1504 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1505 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1506 		return -EMSGSIZE;
1507 	}
1508 
1509 	/* CHECKSUM_PARTIAL only with no extension headers and when
1510 	 * we are not going to fragment
1511 	 */
1512 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1513 	    headersize == sizeof(struct ipv6hdr) &&
1514 	    length <= mtu - headersize &&
1515 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1516 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1517 		csummode = CHECKSUM_PARTIAL;
1518 
1519 	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1520 		uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1521 		if (!uarg)
1522 			return -ENOBUFS;
1523 		extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1524 		if (rt->dst.dev->features & NETIF_F_SG &&
1525 		    csummode == CHECKSUM_PARTIAL) {
1526 			paged = true;
1527 		} else {
1528 			uarg->zerocopy = 0;
1529 			skb_zcopy_set(skb, uarg, &extra_uref);
1530 		}
1531 	}
1532 
1533 	/*
1534 	 * Let's try using as much space as possible.
1535 	 * Use MTU if total length of the message fits into the MTU.
1536 	 * Otherwise, we need to reserve fragment header and
1537 	 * fragment alignment (= 8-15 octects, in total).
1538 	 *
1539 	 * Note that we may need to "move" the data from the tail
1540 	 * of the buffer to the new fragment when we split
1541 	 * the message.
1542 	 *
1543 	 * FIXME: It may be fragmented into multiple chunks
1544 	 *        at once if non-fragmentable extension headers
1545 	 *        are too large.
1546 	 * --yoshfuji
1547 	 */
1548 
1549 	cork->length += length;
1550 	if (!skb)
1551 		goto alloc_new_skb;
1552 
1553 	while (length > 0) {
1554 		/* Check if the remaining data fits into current packet. */
1555 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1556 		if (copy < length)
1557 			copy = maxfraglen - skb->len;
1558 
1559 		if (copy <= 0) {
1560 			char *data;
1561 			unsigned int datalen;
1562 			unsigned int fraglen;
1563 			unsigned int fraggap;
1564 			unsigned int alloclen, alloc_extra;
1565 			unsigned int pagedlen;
1566 alloc_new_skb:
1567 			/* There's no room in the current skb */
1568 			if (skb)
1569 				fraggap = skb->len - maxfraglen;
1570 			else
1571 				fraggap = 0;
1572 			/* update mtu and maxfraglen if necessary */
1573 			if (!skb || !skb_prev)
1574 				ip6_append_data_mtu(&mtu, &maxfraglen,
1575 						    fragheaderlen, skb, rt,
1576 						    orig_mtu);
1577 
1578 			skb_prev = skb;
1579 
1580 			/*
1581 			 * If remaining data exceeds the mtu,
1582 			 * we know we need more fragment(s).
1583 			 */
1584 			datalen = length + fraggap;
1585 
1586 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1587 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1588 			fraglen = datalen + fragheaderlen;
1589 			pagedlen = 0;
1590 
1591 			alloc_extra = hh_len;
1592 			alloc_extra += dst_exthdrlen;
1593 			alloc_extra += rt->dst.trailer_len;
1594 
1595 			/* We just reserve space for fragment header.
1596 			 * Note: this may be overallocation if the message
1597 			 * (without MSG_MORE) fits into the MTU.
1598 			 */
1599 			alloc_extra += sizeof(struct frag_hdr);
1600 
1601 			if ((flags & MSG_MORE) &&
1602 			    !(rt->dst.dev->features&NETIF_F_SG))
1603 				alloclen = mtu;
1604 			else if (!paged &&
1605 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1606 				  !(rt->dst.dev->features & NETIF_F_SG)))
1607 				alloclen = fraglen;
1608 			else {
1609 				alloclen = min_t(int, fraglen, MAX_HEADER);
1610 				pagedlen = fraglen - alloclen;
1611 			}
1612 			alloclen += alloc_extra;
1613 
1614 			if (datalen != length + fraggap) {
1615 				/*
1616 				 * this is not the last fragment, the trailer
1617 				 * space is regarded as data space.
1618 				 */
1619 				datalen += rt->dst.trailer_len;
1620 			}
1621 
1622 			fraglen = datalen + fragheaderlen;
1623 
1624 			copy = datalen - transhdrlen - fraggap - pagedlen;
1625 			if (copy < 0) {
1626 				err = -EINVAL;
1627 				goto error;
1628 			}
1629 			if (transhdrlen) {
1630 				skb = sock_alloc_send_skb(sk, alloclen,
1631 						(flags & MSG_DONTWAIT), &err);
1632 			} else {
1633 				skb = NULL;
1634 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1635 				    2 * sk->sk_sndbuf)
1636 					skb = alloc_skb(alloclen,
1637 							sk->sk_allocation);
1638 				if (unlikely(!skb))
1639 					err = -ENOBUFS;
1640 			}
1641 			if (!skb)
1642 				goto error;
1643 			/*
1644 			 *	Fill in the control structures
1645 			 */
1646 			skb->protocol = htons(ETH_P_IPV6);
1647 			skb->ip_summed = csummode;
1648 			skb->csum = 0;
1649 			/* reserve for fragmentation and ipsec header */
1650 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1651 				    dst_exthdrlen);
1652 
1653 			/*
1654 			 *	Find where to start putting bytes
1655 			 */
1656 			data = skb_put(skb, fraglen - pagedlen);
1657 			skb_set_network_header(skb, exthdrlen);
1658 			data += fragheaderlen;
1659 			skb->transport_header = (skb->network_header +
1660 						 fragheaderlen);
1661 			if (fraggap) {
1662 				skb->csum = skb_copy_and_csum_bits(
1663 					skb_prev, maxfraglen,
1664 					data + transhdrlen, fraggap);
1665 				skb_prev->csum = csum_sub(skb_prev->csum,
1666 							  skb->csum);
1667 				data += fraggap;
1668 				pskb_trim_unique(skb_prev, maxfraglen);
1669 			}
1670 			if (copy > 0 &&
1671 			    getfrag(from, data + transhdrlen, offset,
1672 				    copy, fraggap, skb) < 0) {
1673 				err = -EFAULT;
1674 				kfree_skb(skb);
1675 				goto error;
1676 			}
1677 
1678 			offset += copy;
1679 			length -= copy + transhdrlen;
1680 			transhdrlen = 0;
1681 			exthdrlen = 0;
1682 			dst_exthdrlen = 0;
1683 
1684 			/* Only the initial fragment is time stamped */
1685 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1686 			cork->tx_flags = 0;
1687 			skb_shinfo(skb)->tskey = tskey;
1688 			tskey = 0;
1689 			skb_zcopy_set(skb, uarg, &extra_uref);
1690 
1691 			if ((flags & MSG_CONFIRM) && !skb_prev)
1692 				skb_set_dst_pending_confirm(skb, 1);
1693 
1694 			/*
1695 			 * Put the packet on the pending queue
1696 			 */
1697 			if (!skb->destructor) {
1698 				skb->destructor = sock_wfree;
1699 				skb->sk = sk;
1700 				wmem_alloc_delta += skb->truesize;
1701 			}
1702 			__skb_queue_tail(queue, skb);
1703 			continue;
1704 		}
1705 
1706 		if (copy > length)
1707 			copy = length;
1708 
1709 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1710 		    skb_tailroom(skb) >= copy) {
1711 			unsigned int off;
1712 
1713 			off = skb->len;
1714 			if (getfrag(from, skb_put(skb, copy),
1715 						offset, copy, off, skb) < 0) {
1716 				__skb_trim(skb, off);
1717 				err = -EFAULT;
1718 				goto error;
1719 			}
1720 		} else if (!uarg || !uarg->zerocopy) {
1721 			int i = skb_shinfo(skb)->nr_frags;
1722 
1723 			err = -ENOMEM;
1724 			if (!sk_page_frag_refill(sk, pfrag))
1725 				goto error;
1726 
1727 			if (!skb_can_coalesce(skb, i, pfrag->page,
1728 					      pfrag->offset)) {
1729 				err = -EMSGSIZE;
1730 				if (i == MAX_SKB_FRAGS)
1731 					goto error;
1732 
1733 				__skb_fill_page_desc(skb, i, pfrag->page,
1734 						     pfrag->offset, 0);
1735 				skb_shinfo(skb)->nr_frags = ++i;
1736 				get_page(pfrag->page);
1737 			}
1738 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1739 			if (getfrag(from,
1740 				    page_address(pfrag->page) + pfrag->offset,
1741 				    offset, copy, skb->len, skb) < 0)
1742 				goto error_efault;
1743 
1744 			pfrag->offset += copy;
1745 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1746 			skb->len += copy;
1747 			skb->data_len += copy;
1748 			skb->truesize += copy;
1749 			wmem_alloc_delta += copy;
1750 		} else {
1751 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1752 			if (err < 0)
1753 				goto error;
1754 		}
1755 		offset += copy;
1756 		length -= copy;
1757 	}
1758 
1759 	if (wmem_alloc_delta)
1760 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1761 	return 0;
1762 
1763 error_efault:
1764 	err = -EFAULT;
1765 error:
1766 	net_zcopy_put_abort(uarg, extra_uref);
1767 	cork->length -= length;
1768 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1769 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1770 	return err;
1771 }
1772 
1773 int ip6_append_data(struct sock *sk,
1774 		    int getfrag(void *from, char *to, int offset, int len,
1775 				int odd, struct sk_buff *skb),
1776 		    void *from, int length, int transhdrlen,
1777 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1778 		    struct rt6_info *rt, unsigned int flags)
1779 {
1780 	struct inet_sock *inet = inet_sk(sk);
1781 	struct ipv6_pinfo *np = inet6_sk(sk);
1782 	int exthdrlen;
1783 	int err;
1784 
1785 	if (flags&MSG_PROBE)
1786 		return 0;
1787 	if (skb_queue_empty(&sk->sk_write_queue)) {
1788 		/*
1789 		 * setup for corking
1790 		 */
1791 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1792 				     ipc6, rt, fl6);
1793 		if (err)
1794 			return err;
1795 
1796 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1797 		length += exthdrlen;
1798 		transhdrlen += exthdrlen;
1799 	} else {
1800 		fl6 = &inet->cork.fl.u.ip6;
1801 		transhdrlen = 0;
1802 	}
1803 
1804 	return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1805 				 &np->cork, sk_page_frag(sk), getfrag,
1806 				 from, length, transhdrlen, flags, ipc6);
1807 }
1808 EXPORT_SYMBOL_GPL(ip6_append_data);
1809 
1810 static void ip6_cork_release(struct inet_cork_full *cork,
1811 			     struct inet6_cork *v6_cork)
1812 {
1813 	if (v6_cork->opt) {
1814 		kfree(v6_cork->opt->dst0opt);
1815 		kfree(v6_cork->opt->dst1opt);
1816 		kfree(v6_cork->opt->hopopt);
1817 		kfree(v6_cork->opt->srcrt);
1818 		kfree(v6_cork->opt);
1819 		v6_cork->opt = NULL;
1820 	}
1821 
1822 	if (cork->base.dst) {
1823 		dst_release(cork->base.dst);
1824 		cork->base.dst = NULL;
1825 		cork->base.flags &= ~IPCORK_ALLFRAG;
1826 	}
1827 	memset(&cork->fl, 0, sizeof(cork->fl));
1828 }
1829 
1830 struct sk_buff *__ip6_make_skb(struct sock *sk,
1831 			       struct sk_buff_head *queue,
1832 			       struct inet_cork_full *cork,
1833 			       struct inet6_cork *v6_cork)
1834 {
1835 	struct sk_buff *skb, *tmp_skb;
1836 	struct sk_buff **tail_skb;
1837 	struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1838 	struct ipv6_pinfo *np = inet6_sk(sk);
1839 	struct net *net = sock_net(sk);
1840 	struct ipv6hdr *hdr;
1841 	struct ipv6_txoptions *opt = v6_cork->opt;
1842 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1843 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1844 	unsigned char proto = fl6->flowi6_proto;
1845 
1846 	skb = __skb_dequeue(queue);
1847 	if (!skb)
1848 		goto out;
1849 	tail_skb = &(skb_shinfo(skb)->frag_list);
1850 
1851 	/* move skb->data to ip header from ext header */
1852 	if (skb->data < skb_network_header(skb))
1853 		__skb_pull(skb, skb_network_offset(skb));
1854 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1855 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1856 		*tail_skb = tmp_skb;
1857 		tail_skb = &(tmp_skb->next);
1858 		skb->len += tmp_skb->len;
1859 		skb->data_len += tmp_skb->len;
1860 		skb->truesize += tmp_skb->truesize;
1861 		tmp_skb->destructor = NULL;
1862 		tmp_skb->sk = NULL;
1863 	}
1864 
1865 	/* Allow local fragmentation. */
1866 	skb->ignore_df = ip6_sk_ignore_df(sk);
1867 
1868 	*final_dst = fl6->daddr;
1869 	__skb_pull(skb, skb_network_header_len(skb));
1870 	if (opt && opt->opt_flen)
1871 		ipv6_push_frag_opts(skb, opt, &proto);
1872 	if (opt && opt->opt_nflen)
1873 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1874 
1875 	skb_push(skb, sizeof(struct ipv6hdr));
1876 	skb_reset_network_header(skb);
1877 	hdr = ipv6_hdr(skb);
1878 
1879 	ip6_flow_hdr(hdr, v6_cork->tclass,
1880 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1881 					ip6_autoflowlabel(net, np), fl6));
1882 	hdr->hop_limit = v6_cork->hop_limit;
1883 	hdr->nexthdr = proto;
1884 	hdr->saddr = fl6->saddr;
1885 	hdr->daddr = *final_dst;
1886 
1887 	skb->priority = sk->sk_priority;
1888 	skb->mark = cork->base.mark;
1889 
1890 	skb->tstamp = cork->base.transmit_time;
1891 
1892 	skb_dst_set(skb, dst_clone(&rt->dst));
1893 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1894 	if (proto == IPPROTO_ICMPV6) {
1895 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1896 
1897 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1898 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1899 	}
1900 
1901 	ip6_cork_release(cork, v6_cork);
1902 out:
1903 	return skb;
1904 }
1905 
1906 int ip6_send_skb(struct sk_buff *skb)
1907 {
1908 	struct net *net = sock_net(skb->sk);
1909 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1910 	int err;
1911 
1912 	err = ip6_local_out(net, skb->sk, skb);
1913 	if (err) {
1914 		if (err > 0)
1915 			err = net_xmit_errno(err);
1916 		if (err)
1917 			IP6_INC_STATS(net, rt->rt6i_idev,
1918 				      IPSTATS_MIB_OUTDISCARDS);
1919 	}
1920 
1921 	return err;
1922 }
1923 
1924 int ip6_push_pending_frames(struct sock *sk)
1925 {
1926 	struct sk_buff *skb;
1927 
1928 	skb = ip6_finish_skb(sk);
1929 	if (!skb)
1930 		return 0;
1931 
1932 	return ip6_send_skb(skb);
1933 }
1934 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1935 
1936 static void __ip6_flush_pending_frames(struct sock *sk,
1937 				       struct sk_buff_head *queue,
1938 				       struct inet_cork_full *cork,
1939 				       struct inet6_cork *v6_cork)
1940 {
1941 	struct sk_buff *skb;
1942 
1943 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1944 		if (skb_dst(skb))
1945 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1946 				      IPSTATS_MIB_OUTDISCARDS);
1947 		kfree_skb(skb);
1948 	}
1949 
1950 	ip6_cork_release(cork, v6_cork);
1951 }
1952 
1953 void ip6_flush_pending_frames(struct sock *sk)
1954 {
1955 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1956 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1957 }
1958 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1959 
1960 struct sk_buff *ip6_make_skb(struct sock *sk,
1961 			     int getfrag(void *from, char *to, int offset,
1962 					 int len, int odd, struct sk_buff *skb),
1963 			     void *from, int length, int transhdrlen,
1964 			     struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1965 			     struct rt6_info *rt, unsigned int flags,
1966 			     struct inet_cork_full *cork)
1967 {
1968 	struct inet6_cork v6_cork;
1969 	struct sk_buff_head queue;
1970 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1971 	int err;
1972 
1973 	if (flags & MSG_PROBE)
1974 		return NULL;
1975 
1976 	__skb_queue_head_init(&queue);
1977 
1978 	cork->base.flags = 0;
1979 	cork->base.addr = 0;
1980 	cork->base.opt = NULL;
1981 	cork->base.dst = NULL;
1982 	v6_cork.opt = NULL;
1983 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1984 	if (err) {
1985 		ip6_cork_release(cork, &v6_cork);
1986 		return ERR_PTR(err);
1987 	}
1988 	if (ipc6->dontfrag < 0)
1989 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1990 
1991 	err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1992 				&current->task_frag, getfrag, from,
1993 				length + exthdrlen, transhdrlen + exthdrlen,
1994 				flags, ipc6);
1995 	if (err) {
1996 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1997 		return ERR_PTR(err);
1998 	}
1999 
2000 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2001 }
2002