xref: /openbmc/linux/net/ipv6/ip6_output.c (revision e20d5a22)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	IPv6 output functions
4  *	Linux INET6 implementation
5  *
6  *	Authors:
7  *	Pedro Roque		<roque@di.fc.ul.pt>
8  *
9  *	Based on linux/net/ipv4/ip_output.c
10  *
11  *	Changes:
12  *	A.N.Kuznetsov	:	airthmetics in fragmentation.
13  *				extension headers are implemented.
14  *				route changes now work.
15  *				ip6_forward does not confuse sniffers.
16  *				etc.
17  *
18  *      H. von Brand    :       Added missing #include <linux/string.h>
19  *	Imran Patel	:	frag id should be in NBO
20  *      Kazunori MIYAZAWA @USAGI
21  *			:       add ip6_append_data and related functions
22  *				for datagram xmit
23  */
24 
25 #include <linux/errno.h>
26 #include <linux/kernel.h>
27 #include <linux/string.h>
28 #include <linux/socket.h>
29 #include <linux/net.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/in6.h>
33 #include <linux/tcp.h>
34 #include <linux/route.h>
35 #include <linux/module.h>
36 #include <linux/slab.h>
37 
38 #include <linux/bpf-cgroup.h>
39 #include <linux/netfilter.h>
40 #include <linux/netfilter_ipv6.h>
41 
42 #include <net/sock.h>
43 #include <net/snmp.h>
44 
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 #include <net/protocol.h>
48 #include <net/ip6_route.h>
49 #include <net/addrconf.h>
50 #include <net/rawv6.h>
51 #include <net/icmp.h>
52 #include <net/xfrm.h>
53 #include <net/checksum.h>
54 #include <linux/mroute6.h>
55 #include <net/l3mdev.h>
56 #include <net/lwtunnel.h>
57 #include <net/ip_tunnels.h>
58 
59 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
60 {
61 	struct dst_entry *dst = skb_dst(skb);
62 	struct net_device *dev = dst->dev;
63 	struct inet6_dev *idev = ip6_dst_idev(dst);
64 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
65 	const struct in6_addr *daddr, *nexthop;
66 	struct ipv6hdr *hdr;
67 	struct neighbour *neigh;
68 	int ret;
69 
70 	/* Be paranoid, rather than too clever. */
71 	if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
72 		skb = skb_expand_head(skb, hh_len);
73 		if (!skb) {
74 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
75 			return -ENOMEM;
76 		}
77 	}
78 
79 	hdr = ipv6_hdr(skb);
80 	daddr = &hdr->daddr;
81 	if (ipv6_addr_is_multicast(daddr)) {
82 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
83 		    ((mroute6_is_socket(net, skb) &&
84 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
85 		     ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
86 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
87 
88 			/* Do not check for IFF_ALLMULTI; multicast routing
89 			   is not supported in any case.
90 			 */
91 			if (newskb)
92 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
93 					net, sk, newskb, NULL, newskb->dev,
94 					dev_loopback_xmit);
95 
96 			if (hdr->hop_limit == 0) {
97 				IP6_INC_STATS(net, idev,
98 					      IPSTATS_MIB_OUTDISCARDS);
99 				kfree_skb(skb);
100 				return 0;
101 			}
102 		}
103 
104 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
105 		if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
106 		    !(dev->flags & IFF_LOOPBACK)) {
107 			kfree_skb(skb);
108 			return 0;
109 		}
110 	}
111 
112 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
113 		int res = lwtunnel_xmit(skb);
114 
115 		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
116 			return res;
117 	}
118 
119 	rcu_read_lock_bh();
120 	nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
121 	neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
122 
123 	if (unlikely(IS_ERR_OR_NULL(neigh))) {
124 		if (unlikely(!neigh))
125 			neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
126 		if (IS_ERR(neigh)) {
127 			rcu_read_unlock_bh();
128 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
129 			kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
130 			return -EINVAL;
131 		}
132 	}
133 	sock_confirm_neigh(skb, neigh);
134 	ret = neigh_output(neigh, skb, false);
135 	rcu_read_unlock_bh();
136 	return ret;
137 }
138 
139 static int
140 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
141 				    struct sk_buff *skb, unsigned int mtu)
142 {
143 	struct sk_buff *segs, *nskb;
144 	netdev_features_t features;
145 	int ret = 0;
146 
147 	/* Please see corresponding comment in ip_finish_output_gso
148 	 * describing the cases where GSO segment length exceeds the
149 	 * egress MTU.
150 	 */
151 	features = netif_skb_features(skb);
152 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
153 	if (IS_ERR_OR_NULL(segs)) {
154 		kfree_skb(skb);
155 		return -ENOMEM;
156 	}
157 
158 	consume_skb(skb);
159 
160 	skb_list_walk_safe(segs, segs, nskb) {
161 		int err;
162 
163 		skb_mark_not_on_list(segs);
164 		err = ip6_fragment(net, sk, segs, ip6_finish_output2);
165 		if (err && ret == 0)
166 			ret = err;
167 	}
168 
169 	return ret;
170 }
171 
172 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
173 {
174 	unsigned int mtu;
175 
176 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
177 	/* Policy lookup after SNAT yielded a new policy */
178 	if (skb_dst(skb)->xfrm) {
179 		IP6CB(skb)->flags |= IP6SKB_REROUTED;
180 		return dst_output(net, sk, skb);
181 	}
182 #endif
183 
184 	mtu = ip6_skb_dst_mtu(skb);
185 	if (skb_is_gso(skb) &&
186 	    !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
187 	    !skb_gso_validate_network_len(skb, mtu))
188 		return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
189 
190 	if ((skb->len > mtu && !skb_is_gso(skb)) ||
191 	    dst_allfrag(skb_dst(skb)) ||
192 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
193 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
194 	else
195 		return ip6_finish_output2(net, sk, skb);
196 }
197 
198 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
199 {
200 	int ret;
201 
202 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
203 	switch (ret) {
204 	case NET_XMIT_SUCCESS:
205 	case NET_XMIT_CN:
206 		return __ip6_finish_output(net, sk, skb) ? : ret;
207 	default:
208 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
209 		return ret;
210 	}
211 }
212 
213 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
214 {
215 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
216 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
217 
218 	skb->protocol = htons(ETH_P_IPV6);
219 	skb->dev = dev;
220 
221 	if (unlikely(idev->cnf.disable_ipv6)) {
222 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
223 		kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
224 		return 0;
225 	}
226 
227 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
228 			    net, sk, skb, indev, dev,
229 			    ip6_finish_output,
230 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
231 }
232 EXPORT_SYMBOL(ip6_output);
233 
234 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
235 {
236 	if (!np->autoflowlabel_set)
237 		return ip6_default_np_autolabel(net);
238 	else
239 		return np->autoflowlabel;
240 }
241 
242 /*
243  * xmit an sk_buff (used by TCP, SCTP and DCCP)
244  * Note : socket lock is not held for SYNACK packets, but might be modified
245  * by calls to skb_set_owner_w() and ipv6_local_error(),
246  * which are using proper atomic operations or spinlocks.
247  */
248 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
249 	     __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
250 {
251 	struct net *net = sock_net(sk);
252 	const struct ipv6_pinfo *np = inet6_sk(sk);
253 	struct in6_addr *first_hop = &fl6->daddr;
254 	struct dst_entry *dst = skb_dst(skb);
255 	struct net_device *dev = dst->dev;
256 	struct inet6_dev *idev = ip6_dst_idev(dst);
257 	struct hop_jumbo_hdr *hop_jumbo;
258 	int hoplen = sizeof(*hop_jumbo);
259 	unsigned int head_room;
260 	struct ipv6hdr *hdr;
261 	u8  proto = fl6->flowi6_proto;
262 	int seg_len = skb->len;
263 	int hlimit = -1;
264 	u32 mtu;
265 
266 	head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
267 	if (opt)
268 		head_room += opt->opt_nflen + opt->opt_flen;
269 
270 	if (unlikely(head_room > skb_headroom(skb))) {
271 		skb = skb_expand_head(skb, head_room);
272 		if (!skb) {
273 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
274 			return -ENOBUFS;
275 		}
276 	}
277 
278 	if (opt) {
279 		seg_len += opt->opt_nflen + opt->opt_flen;
280 
281 		if (opt->opt_flen)
282 			ipv6_push_frag_opts(skb, opt, &proto);
283 
284 		if (opt->opt_nflen)
285 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
286 					     &fl6->saddr);
287 	}
288 
289 	if (unlikely(seg_len > IPV6_MAXPLEN)) {
290 		hop_jumbo = skb_push(skb, hoplen);
291 
292 		hop_jumbo->nexthdr = proto;
293 		hop_jumbo->hdrlen = 0;
294 		hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
295 		hop_jumbo->tlv_len = 4;
296 		hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
297 
298 		proto = IPPROTO_HOPOPTS;
299 		seg_len = 0;
300 		IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
301 	}
302 
303 	skb_push(skb, sizeof(struct ipv6hdr));
304 	skb_reset_network_header(skb);
305 	hdr = ipv6_hdr(skb);
306 
307 	/*
308 	 *	Fill in the IPv6 header
309 	 */
310 	if (np)
311 		hlimit = np->hop_limit;
312 	if (hlimit < 0)
313 		hlimit = ip6_dst_hoplimit(dst);
314 
315 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
316 				ip6_autoflowlabel(net, np), fl6));
317 
318 	hdr->payload_len = htons(seg_len);
319 	hdr->nexthdr = proto;
320 	hdr->hop_limit = hlimit;
321 
322 	hdr->saddr = fl6->saddr;
323 	hdr->daddr = *first_hop;
324 
325 	skb->protocol = htons(ETH_P_IPV6);
326 	skb->priority = priority;
327 	skb->mark = mark;
328 
329 	mtu = dst_mtu(dst);
330 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
331 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
332 
333 		/* if egress device is enslaved to an L3 master device pass the
334 		 * skb to its handler for processing
335 		 */
336 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
337 		if (unlikely(!skb))
338 			return 0;
339 
340 		/* hooks should never assume socket lock is held.
341 		 * we promote our socket to non const
342 		 */
343 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
344 			       net, (struct sock *)sk, skb, NULL, dev,
345 			       dst_output);
346 	}
347 
348 	skb->dev = dev;
349 	/* ipv6_local_error() does not require socket lock,
350 	 * we promote our socket to non const
351 	 */
352 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
353 
354 	IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
355 	kfree_skb(skb);
356 	return -EMSGSIZE;
357 }
358 EXPORT_SYMBOL(ip6_xmit);
359 
360 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
361 {
362 	struct ip6_ra_chain *ra;
363 	struct sock *last = NULL;
364 
365 	read_lock(&ip6_ra_lock);
366 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
367 		struct sock *sk = ra->sk;
368 		if (sk && ra->sel == sel &&
369 		    (!sk->sk_bound_dev_if ||
370 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
371 			struct ipv6_pinfo *np = inet6_sk(sk);
372 
373 			if (np && np->rtalert_isolate &&
374 			    !net_eq(sock_net(sk), dev_net(skb->dev))) {
375 				continue;
376 			}
377 			if (last) {
378 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
379 				if (skb2)
380 					rawv6_rcv(last, skb2);
381 			}
382 			last = sk;
383 		}
384 	}
385 
386 	if (last) {
387 		rawv6_rcv(last, skb);
388 		read_unlock(&ip6_ra_lock);
389 		return 1;
390 	}
391 	read_unlock(&ip6_ra_lock);
392 	return 0;
393 }
394 
395 static int ip6_forward_proxy_check(struct sk_buff *skb)
396 {
397 	struct ipv6hdr *hdr = ipv6_hdr(skb);
398 	u8 nexthdr = hdr->nexthdr;
399 	__be16 frag_off;
400 	int offset;
401 
402 	if (ipv6_ext_hdr(nexthdr)) {
403 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
404 		if (offset < 0)
405 			return 0;
406 	} else
407 		offset = sizeof(struct ipv6hdr);
408 
409 	if (nexthdr == IPPROTO_ICMPV6) {
410 		struct icmp6hdr *icmp6;
411 
412 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
413 					 offset + 1 - skb->data)))
414 			return 0;
415 
416 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
417 
418 		switch (icmp6->icmp6_type) {
419 		case NDISC_ROUTER_SOLICITATION:
420 		case NDISC_ROUTER_ADVERTISEMENT:
421 		case NDISC_NEIGHBOUR_SOLICITATION:
422 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
423 		case NDISC_REDIRECT:
424 			/* For reaction involving unicast neighbor discovery
425 			 * message destined to the proxied address, pass it to
426 			 * input function.
427 			 */
428 			return 1;
429 		default:
430 			break;
431 		}
432 	}
433 
434 	/*
435 	 * The proxying router can't forward traffic sent to a link-local
436 	 * address, so signal the sender and discard the packet. This
437 	 * behavior is clarified by the MIPv6 specification.
438 	 */
439 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
440 		dst_link_failure(skb);
441 		return -1;
442 	}
443 
444 	return 0;
445 }
446 
447 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
448 				     struct sk_buff *skb)
449 {
450 	struct dst_entry *dst = skb_dst(skb);
451 
452 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
453 	__IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
454 
455 #ifdef CONFIG_NET_SWITCHDEV
456 	if (skb->offload_l3_fwd_mark) {
457 		consume_skb(skb);
458 		return 0;
459 	}
460 #endif
461 
462 	skb_clear_tstamp(skb);
463 	return dst_output(net, sk, skb);
464 }
465 
466 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
467 {
468 	if (skb->len <= mtu)
469 		return false;
470 
471 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
472 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
473 		return true;
474 
475 	if (skb->ignore_df)
476 		return false;
477 
478 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
479 		return false;
480 
481 	return true;
482 }
483 
484 int ip6_forward(struct sk_buff *skb)
485 {
486 	struct dst_entry *dst = skb_dst(skb);
487 	struct ipv6hdr *hdr = ipv6_hdr(skb);
488 	struct inet6_skb_parm *opt = IP6CB(skb);
489 	struct net *net = dev_net(dst->dev);
490 	struct inet6_dev *idev;
491 	SKB_DR(reason);
492 	u32 mtu;
493 
494 	idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
495 	if (net->ipv6.devconf_all->forwarding == 0)
496 		goto error;
497 
498 	if (skb->pkt_type != PACKET_HOST)
499 		goto drop;
500 
501 	if (unlikely(skb->sk))
502 		goto drop;
503 
504 	if (skb_warn_if_lro(skb))
505 		goto drop;
506 
507 	if (!net->ipv6.devconf_all->disable_policy &&
508 	    (!idev || !idev->cnf.disable_policy) &&
509 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
510 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
511 		goto drop;
512 	}
513 
514 	skb_forward_csum(skb);
515 
516 	/*
517 	 *	We DO NOT make any processing on
518 	 *	RA packets, pushing them to user level AS IS
519 	 *	without ane WARRANTY that application will be able
520 	 *	to interpret them. The reason is that we
521 	 *	cannot make anything clever here.
522 	 *
523 	 *	We are not end-node, so that if packet contains
524 	 *	AH/ESP, we cannot make anything.
525 	 *	Defragmentation also would be mistake, RA packets
526 	 *	cannot be fragmented, because there is no warranty
527 	 *	that different fragments will go along one path. --ANK
528 	 */
529 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
530 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
531 			return 0;
532 	}
533 
534 	/*
535 	 *	check and decrement ttl
536 	 */
537 	if (hdr->hop_limit <= 1) {
538 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
539 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
540 
541 		kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
542 		return -ETIMEDOUT;
543 	}
544 
545 	/* XXX: idev->cnf.proxy_ndp? */
546 	if (net->ipv6.devconf_all->proxy_ndp &&
547 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
548 		int proxied = ip6_forward_proxy_check(skb);
549 		if (proxied > 0) {
550 			hdr->hop_limit--;
551 			return ip6_input(skb);
552 		} else if (proxied < 0) {
553 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
554 			goto drop;
555 		}
556 	}
557 
558 	if (!xfrm6_route_forward(skb)) {
559 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
560 		SKB_DR_SET(reason, XFRM_POLICY);
561 		goto drop;
562 	}
563 	dst = skb_dst(skb);
564 
565 	/* IPv6 specs say nothing about it, but it is clear that we cannot
566 	   send redirects to source routed frames.
567 	   We don't send redirects to frames decapsulated from IPsec.
568 	 */
569 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
570 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
571 		struct in6_addr *target = NULL;
572 		struct inet_peer *peer;
573 		struct rt6_info *rt;
574 
575 		/*
576 		 *	incoming and outgoing devices are the same
577 		 *	send a redirect.
578 		 */
579 
580 		rt = (struct rt6_info *) dst;
581 		if (rt->rt6i_flags & RTF_GATEWAY)
582 			target = &rt->rt6i_gateway;
583 		else
584 			target = &hdr->daddr;
585 
586 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
587 
588 		/* Limit redirects both by destination (here)
589 		   and by source (inside ndisc_send_redirect)
590 		 */
591 		if (inet_peer_xrlim_allow(peer, 1*HZ))
592 			ndisc_send_redirect(skb, target);
593 		if (peer)
594 			inet_putpeer(peer);
595 	} else {
596 		int addrtype = ipv6_addr_type(&hdr->saddr);
597 
598 		/* This check is security critical. */
599 		if (addrtype == IPV6_ADDR_ANY ||
600 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
601 			goto error;
602 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
603 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
604 				    ICMPV6_NOT_NEIGHBOUR, 0);
605 			goto error;
606 		}
607 	}
608 
609 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
610 	if (mtu < IPV6_MIN_MTU)
611 		mtu = IPV6_MIN_MTU;
612 
613 	if (ip6_pkt_too_big(skb, mtu)) {
614 		/* Again, force OUTPUT device used as source address */
615 		skb->dev = dst->dev;
616 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
617 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
618 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
619 				IPSTATS_MIB_FRAGFAILS);
620 		kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
621 		return -EMSGSIZE;
622 	}
623 
624 	if (skb_cow(skb, dst->dev->hard_header_len)) {
625 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
626 				IPSTATS_MIB_OUTDISCARDS);
627 		goto drop;
628 	}
629 
630 	hdr = ipv6_hdr(skb);
631 
632 	/* Mangling hops number delayed to point after skb COW */
633 
634 	hdr->hop_limit--;
635 
636 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
637 		       net, NULL, skb, skb->dev, dst->dev,
638 		       ip6_forward_finish);
639 
640 error:
641 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
642 	SKB_DR_SET(reason, IP_INADDRERRORS);
643 drop:
644 	kfree_skb_reason(skb, reason);
645 	return -EINVAL;
646 }
647 
648 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
649 {
650 	to->pkt_type = from->pkt_type;
651 	to->priority = from->priority;
652 	to->protocol = from->protocol;
653 	skb_dst_drop(to);
654 	skb_dst_set(to, dst_clone(skb_dst(from)));
655 	to->dev = from->dev;
656 	to->mark = from->mark;
657 
658 	skb_copy_hash(to, from);
659 
660 #ifdef CONFIG_NET_SCHED
661 	to->tc_index = from->tc_index;
662 #endif
663 	nf_copy(to, from);
664 	skb_ext_copy(to, from);
665 	skb_copy_secmark(to, from);
666 }
667 
668 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
669 		      u8 nexthdr, __be32 frag_id,
670 		      struct ip6_fraglist_iter *iter)
671 {
672 	unsigned int first_len;
673 	struct frag_hdr *fh;
674 
675 	/* BUILD HEADER */
676 	*prevhdr = NEXTHDR_FRAGMENT;
677 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
678 	if (!iter->tmp_hdr)
679 		return -ENOMEM;
680 
681 	iter->frag = skb_shinfo(skb)->frag_list;
682 	skb_frag_list_init(skb);
683 
684 	iter->offset = 0;
685 	iter->hlen = hlen;
686 	iter->frag_id = frag_id;
687 	iter->nexthdr = nexthdr;
688 
689 	__skb_pull(skb, hlen);
690 	fh = __skb_push(skb, sizeof(struct frag_hdr));
691 	__skb_push(skb, hlen);
692 	skb_reset_network_header(skb);
693 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
694 
695 	fh->nexthdr = nexthdr;
696 	fh->reserved = 0;
697 	fh->frag_off = htons(IP6_MF);
698 	fh->identification = frag_id;
699 
700 	first_len = skb_pagelen(skb);
701 	skb->data_len = first_len - skb_headlen(skb);
702 	skb->len = first_len;
703 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
704 
705 	return 0;
706 }
707 EXPORT_SYMBOL(ip6_fraglist_init);
708 
709 void ip6_fraglist_prepare(struct sk_buff *skb,
710 			  struct ip6_fraglist_iter *iter)
711 {
712 	struct sk_buff *frag = iter->frag;
713 	unsigned int hlen = iter->hlen;
714 	struct frag_hdr *fh;
715 
716 	frag->ip_summed = CHECKSUM_NONE;
717 	skb_reset_transport_header(frag);
718 	fh = __skb_push(frag, sizeof(struct frag_hdr));
719 	__skb_push(frag, hlen);
720 	skb_reset_network_header(frag);
721 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
722 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
723 	fh->nexthdr = iter->nexthdr;
724 	fh->reserved = 0;
725 	fh->frag_off = htons(iter->offset);
726 	if (frag->next)
727 		fh->frag_off |= htons(IP6_MF);
728 	fh->identification = iter->frag_id;
729 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
730 	ip6_copy_metadata(frag, skb);
731 }
732 EXPORT_SYMBOL(ip6_fraglist_prepare);
733 
734 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
735 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
736 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
737 {
738 	state->prevhdr = prevhdr;
739 	state->nexthdr = nexthdr;
740 	state->frag_id = frag_id;
741 
742 	state->hlen = hlen;
743 	state->mtu = mtu;
744 
745 	state->left = skb->len - hlen;	/* Space per frame */
746 	state->ptr = hlen;		/* Where to start from */
747 
748 	state->hroom = hdr_room;
749 	state->troom = needed_tailroom;
750 
751 	state->offset = 0;
752 }
753 EXPORT_SYMBOL(ip6_frag_init);
754 
755 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
756 {
757 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
758 	struct sk_buff *frag;
759 	struct frag_hdr *fh;
760 	unsigned int len;
761 
762 	len = state->left;
763 	/* IF: it doesn't fit, use 'mtu' - the data space left */
764 	if (len > state->mtu)
765 		len = state->mtu;
766 	/* IF: we are not sending up to and including the packet end
767 	   then align the next start on an eight byte boundary */
768 	if (len < state->left)
769 		len &= ~7;
770 
771 	/* Allocate buffer */
772 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
773 			 state->hroom + state->troom, GFP_ATOMIC);
774 	if (!frag)
775 		return ERR_PTR(-ENOMEM);
776 
777 	/*
778 	 *	Set up data on packet
779 	 */
780 
781 	ip6_copy_metadata(frag, skb);
782 	skb_reserve(frag, state->hroom);
783 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
784 	skb_reset_network_header(frag);
785 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
786 	frag->transport_header = (frag->network_header + state->hlen +
787 				  sizeof(struct frag_hdr));
788 
789 	/*
790 	 *	Charge the memory for the fragment to any owner
791 	 *	it might possess
792 	 */
793 	if (skb->sk)
794 		skb_set_owner_w(frag, skb->sk);
795 
796 	/*
797 	 *	Copy the packet header into the new buffer.
798 	 */
799 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
800 
801 	fragnexthdr_offset = skb_network_header(frag);
802 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
803 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
804 
805 	/*
806 	 *	Build fragment header.
807 	 */
808 	fh->nexthdr = state->nexthdr;
809 	fh->reserved = 0;
810 	fh->identification = state->frag_id;
811 
812 	/*
813 	 *	Copy a block of the IP datagram.
814 	 */
815 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
816 			     len));
817 	state->left -= len;
818 
819 	fh->frag_off = htons(state->offset);
820 	if (state->left > 0)
821 		fh->frag_off |= htons(IP6_MF);
822 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
823 
824 	state->ptr += len;
825 	state->offset += len;
826 
827 	return frag;
828 }
829 EXPORT_SYMBOL(ip6_frag_next);
830 
831 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
832 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
833 {
834 	struct sk_buff *frag;
835 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
836 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
837 				inet6_sk(skb->sk) : NULL;
838 	bool mono_delivery_time = skb->mono_delivery_time;
839 	struct ip6_frag_state state;
840 	unsigned int mtu, hlen, nexthdr_offset;
841 	ktime_t tstamp = skb->tstamp;
842 	int hroom, err = 0;
843 	__be32 frag_id;
844 	u8 *prevhdr, nexthdr = 0;
845 
846 	err = ip6_find_1stfragopt(skb, &prevhdr);
847 	if (err < 0)
848 		goto fail;
849 	hlen = err;
850 	nexthdr = *prevhdr;
851 	nexthdr_offset = prevhdr - skb_network_header(skb);
852 
853 	mtu = ip6_skb_dst_mtu(skb);
854 
855 	/* We must not fragment if the socket is set to force MTU discovery
856 	 * or if the skb it not generated by a local socket.
857 	 */
858 	if (unlikely(!skb->ignore_df && skb->len > mtu))
859 		goto fail_toobig;
860 
861 	if (IP6CB(skb)->frag_max_size) {
862 		if (IP6CB(skb)->frag_max_size > mtu)
863 			goto fail_toobig;
864 
865 		/* don't send fragments larger than what we received */
866 		mtu = IP6CB(skb)->frag_max_size;
867 		if (mtu < IPV6_MIN_MTU)
868 			mtu = IPV6_MIN_MTU;
869 	}
870 
871 	if (np && np->frag_size < mtu) {
872 		if (np->frag_size)
873 			mtu = np->frag_size;
874 	}
875 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
876 		goto fail_toobig;
877 	mtu -= hlen + sizeof(struct frag_hdr);
878 
879 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
880 				    &ipv6_hdr(skb)->saddr);
881 
882 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
883 	    (err = skb_checksum_help(skb)))
884 		goto fail;
885 
886 	prevhdr = skb_network_header(skb) + nexthdr_offset;
887 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
888 	if (skb_has_frag_list(skb)) {
889 		unsigned int first_len = skb_pagelen(skb);
890 		struct ip6_fraglist_iter iter;
891 		struct sk_buff *frag2;
892 
893 		if (first_len - hlen > mtu ||
894 		    ((first_len - hlen) & 7) ||
895 		    skb_cloned(skb) ||
896 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
897 			goto slow_path;
898 
899 		skb_walk_frags(skb, frag) {
900 			/* Correct geometry. */
901 			if (frag->len > mtu ||
902 			    ((frag->len & 7) && frag->next) ||
903 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
904 				goto slow_path_clean;
905 
906 			/* Partially cloned skb? */
907 			if (skb_shared(frag))
908 				goto slow_path_clean;
909 
910 			BUG_ON(frag->sk);
911 			if (skb->sk) {
912 				frag->sk = skb->sk;
913 				frag->destructor = sock_wfree;
914 			}
915 			skb->truesize -= frag->truesize;
916 		}
917 
918 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
919 					&iter);
920 		if (err < 0)
921 			goto fail;
922 
923 		/* We prevent @rt from being freed. */
924 		rcu_read_lock();
925 
926 		for (;;) {
927 			/* Prepare header of the next frame,
928 			 * before previous one went down. */
929 			if (iter.frag)
930 				ip6_fraglist_prepare(skb, &iter);
931 
932 			skb_set_delivery_time(skb, tstamp, mono_delivery_time);
933 			err = output(net, sk, skb);
934 			if (!err)
935 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
936 					      IPSTATS_MIB_FRAGCREATES);
937 
938 			if (err || !iter.frag)
939 				break;
940 
941 			skb = ip6_fraglist_next(&iter);
942 		}
943 
944 		kfree(iter.tmp_hdr);
945 
946 		if (err == 0) {
947 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
948 				      IPSTATS_MIB_FRAGOKS);
949 			rcu_read_unlock();
950 			return 0;
951 		}
952 
953 		kfree_skb_list(iter.frag);
954 
955 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
956 			      IPSTATS_MIB_FRAGFAILS);
957 		rcu_read_unlock();
958 		return err;
959 
960 slow_path_clean:
961 		skb_walk_frags(skb, frag2) {
962 			if (frag2 == frag)
963 				break;
964 			frag2->sk = NULL;
965 			frag2->destructor = NULL;
966 			skb->truesize += frag2->truesize;
967 		}
968 	}
969 
970 slow_path:
971 	/*
972 	 *	Fragment the datagram.
973 	 */
974 
975 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
976 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
977 		      &state);
978 
979 	/*
980 	 *	Keep copying data until we run out.
981 	 */
982 
983 	while (state.left > 0) {
984 		frag = ip6_frag_next(skb, &state);
985 		if (IS_ERR(frag)) {
986 			err = PTR_ERR(frag);
987 			goto fail;
988 		}
989 
990 		/*
991 		 *	Put this fragment into the sending queue.
992 		 */
993 		skb_set_delivery_time(frag, tstamp, mono_delivery_time);
994 		err = output(net, sk, frag);
995 		if (err)
996 			goto fail;
997 
998 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
999 			      IPSTATS_MIB_FRAGCREATES);
1000 	}
1001 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1002 		      IPSTATS_MIB_FRAGOKS);
1003 	consume_skb(skb);
1004 	return err;
1005 
1006 fail_toobig:
1007 	if (skb->sk && dst_allfrag(skb_dst(skb)))
1008 		sk_gso_disable(skb->sk);
1009 
1010 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1011 	err = -EMSGSIZE;
1012 
1013 fail:
1014 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1015 		      IPSTATS_MIB_FRAGFAILS);
1016 	kfree_skb(skb);
1017 	return err;
1018 }
1019 
1020 static inline int ip6_rt_check(const struct rt6key *rt_key,
1021 			       const struct in6_addr *fl_addr,
1022 			       const struct in6_addr *addr_cache)
1023 {
1024 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1025 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1026 }
1027 
1028 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1029 					  struct dst_entry *dst,
1030 					  const struct flowi6 *fl6)
1031 {
1032 	struct ipv6_pinfo *np = inet6_sk(sk);
1033 	struct rt6_info *rt;
1034 
1035 	if (!dst)
1036 		goto out;
1037 
1038 	if (dst->ops->family != AF_INET6) {
1039 		dst_release(dst);
1040 		return NULL;
1041 	}
1042 
1043 	rt = (struct rt6_info *)dst;
1044 	/* Yes, checking route validity in not connected
1045 	 * case is not very simple. Take into account,
1046 	 * that we do not support routing by source, TOS,
1047 	 * and MSG_DONTROUTE		--ANK (980726)
1048 	 *
1049 	 * 1. ip6_rt_check(): If route was host route,
1050 	 *    check that cached destination is current.
1051 	 *    If it is network route, we still may
1052 	 *    check its validity using saved pointer
1053 	 *    to the last used address: daddr_cache.
1054 	 *    We do not want to save whole address now,
1055 	 *    (because main consumer of this service
1056 	 *    is tcp, which has not this problem),
1057 	 *    so that the last trick works only on connected
1058 	 *    sockets.
1059 	 * 2. oif also should be the same.
1060 	 */
1061 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1062 #ifdef CONFIG_IPV6_SUBTREES
1063 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1064 #endif
1065 	   (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1066 		dst_release(dst);
1067 		dst = NULL;
1068 	}
1069 
1070 out:
1071 	return dst;
1072 }
1073 
1074 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1075 			       struct dst_entry **dst, struct flowi6 *fl6)
1076 {
1077 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1078 	struct neighbour *n;
1079 	struct rt6_info *rt;
1080 #endif
1081 	int err;
1082 	int flags = 0;
1083 
1084 	/* The correct way to handle this would be to do
1085 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1086 	 * the route-specific preferred source forces the
1087 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1088 	 *
1089 	 * In source specific routing (no src=any default route),
1090 	 * ip6_route_output will fail given src=any saddr, though, so
1091 	 * that's why we try it again later.
1092 	 */
1093 	if (ipv6_addr_any(&fl6->saddr)) {
1094 		struct fib6_info *from;
1095 		struct rt6_info *rt;
1096 
1097 		*dst = ip6_route_output(net, sk, fl6);
1098 		rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1099 
1100 		rcu_read_lock();
1101 		from = rt ? rcu_dereference(rt->from) : NULL;
1102 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1103 					  sk ? inet6_sk(sk)->srcprefs : 0,
1104 					  &fl6->saddr);
1105 		rcu_read_unlock();
1106 
1107 		if (err)
1108 			goto out_err_release;
1109 
1110 		/* If we had an erroneous initial result, pretend it
1111 		 * never existed and let the SA-enabled version take
1112 		 * over.
1113 		 */
1114 		if ((*dst)->error) {
1115 			dst_release(*dst);
1116 			*dst = NULL;
1117 		}
1118 
1119 		if (fl6->flowi6_oif)
1120 			flags |= RT6_LOOKUP_F_IFACE;
1121 	}
1122 
1123 	if (!*dst)
1124 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1125 
1126 	err = (*dst)->error;
1127 	if (err)
1128 		goto out_err_release;
1129 
1130 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1131 	/*
1132 	 * Here if the dst entry we've looked up
1133 	 * has a neighbour entry that is in the INCOMPLETE
1134 	 * state and the src address from the flow is
1135 	 * marked as OPTIMISTIC, we release the found
1136 	 * dst entry and replace it instead with the
1137 	 * dst entry of the nexthop router
1138 	 */
1139 	rt = (struct rt6_info *) *dst;
1140 	rcu_read_lock_bh();
1141 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1142 				      rt6_nexthop(rt, &fl6->daddr));
1143 	err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1144 	rcu_read_unlock_bh();
1145 
1146 	if (err) {
1147 		struct inet6_ifaddr *ifp;
1148 		struct flowi6 fl_gw6;
1149 		int redirect;
1150 
1151 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1152 				      (*dst)->dev, 1);
1153 
1154 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1155 		if (ifp)
1156 			in6_ifa_put(ifp);
1157 
1158 		if (redirect) {
1159 			/*
1160 			 * We need to get the dst entry for the
1161 			 * default router instead
1162 			 */
1163 			dst_release(*dst);
1164 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1165 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1166 			*dst = ip6_route_output(net, sk, &fl_gw6);
1167 			err = (*dst)->error;
1168 			if (err)
1169 				goto out_err_release;
1170 		}
1171 	}
1172 #endif
1173 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1174 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1175 		err = -EAFNOSUPPORT;
1176 		goto out_err_release;
1177 	}
1178 
1179 	return 0;
1180 
1181 out_err_release:
1182 	dst_release(*dst);
1183 	*dst = NULL;
1184 
1185 	if (err == -ENETUNREACH)
1186 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1187 	return err;
1188 }
1189 
1190 /**
1191  *	ip6_dst_lookup - perform route lookup on flow
1192  *	@net: Network namespace to perform lookup in
1193  *	@sk: socket which provides route info
1194  *	@dst: pointer to dst_entry * for result
1195  *	@fl6: flow to lookup
1196  *
1197  *	This function performs a route lookup on the given flow.
1198  *
1199  *	It returns zero on success, or a standard errno code on error.
1200  */
1201 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1202 		   struct flowi6 *fl6)
1203 {
1204 	*dst = NULL;
1205 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1206 }
1207 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1208 
1209 /**
1210  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1211  *	@net: Network namespace to perform lookup in
1212  *	@sk: socket which provides route info
1213  *	@fl6: flow to lookup
1214  *	@final_dst: final destination address for ipsec lookup
1215  *
1216  *	This function performs a route lookup on the given flow.
1217  *
1218  *	It returns a valid dst pointer on success, or a pointer encoded
1219  *	error code.
1220  */
1221 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1222 				      const struct in6_addr *final_dst)
1223 {
1224 	struct dst_entry *dst = NULL;
1225 	int err;
1226 
1227 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1228 	if (err)
1229 		return ERR_PTR(err);
1230 	if (final_dst)
1231 		fl6->daddr = *final_dst;
1232 
1233 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1234 }
1235 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1236 
1237 /**
1238  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1239  *	@sk: socket which provides the dst cache and route info
1240  *	@fl6: flow to lookup
1241  *	@final_dst: final destination address for ipsec lookup
1242  *	@connected: whether @sk is connected or not
1243  *
1244  *	This function performs a route lookup on the given flow with the
1245  *	possibility of using the cached route in the socket if it is valid.
1246  *	It will take the socket dst lock when operating on the dst cache.
1247  *	As a result, this function can only be used in process context.
1248  *
1249  *	In addition, for a connected socket, cache the dst in the socket
1250  *	if the current cache is not valid.
1251  *
1252  *	It returns a valid dst pointer on success, or a pointer encoded
1253  *	error code.
1254  */
1255 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1256 					 const struct in6_addr *final_dst,
1257 					 bool connected)
1258 {
1259 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1260 
1261 	dst = ip6_sk_dst_check(sk, dst, fl6);
1262 	if (dst)
1263 		return dst;
1264 
1265 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1266 	if (connected && !IS_ERR(dst))
1267 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1268 
1269 	return dst;
1270 }
1271 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1272 
1273 /**
1274  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1275  *      @skb: Packet for which lookup is done
1276  *      @dev: Tunnel device
1277  *      @net: Network namespace of tunnel device
1278  *      @sock: Socket which provides route info
1279  *      @saddr: Memory to store the src ip address
1280  *      @info: Tunnel information
1281  *      @protocol: IP protocol
1282  *      @use_cache: Flag to enable cache usage
1283  *      This function performs a route lookup on a tunnel
1284  *
1285  *      It returns a valid dst pointer and stores src address to be used in
1286  *      tunnel in param saddr on success, else a pointer encoded error code.
1287  */
1288 
1289 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1290 					struct net_device *dev,
1291 					struct net *net,
1292 					struct socket *sock,
1293 					struct in6_addr *saddr,
1294 					const struct ip_tunnel_info *info,
1295 					u8 protocol,
1296 					bool use_cache)
1297 {
1298 	struct dst_entry *dst = NULL;
1299 #ifdef CONFIG_DST_CACHE
1300 	struct dst_cache *dst_cache;
1301 #endif
1302 	struct flowi6 fl6;
1303 	__u8 prio;
1304 
1305 #ifdef CONFIG_DST_CACHE
1306 	dst_cache = (struct dst_cache *)&info->dst_cache;
1307 	if (use_cache) {
1308 		dst = dst_cache_get_ip6(dst_cache, saddr);
1309 		if (dst)
1310 			return dst;
1311 	}
1312 #endif
1313 	memset(&fl6, 0, sizeof(fl6));
1314 	fl6.flowi6_mark = skb->mark;
1315 	fl6.flowi6_proto = protocol;
1316 	fl6.daddr = info->key.u.ipv6.dst;
1317 	fl6.saddr = info->key.u.ipv6.src;
1318 	prio = info->key.tos;
1319 	fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1320 
1321 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1322 					      NULL);
1323 	if (IS_ERR(dst)) {
1324 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1325 		return ERR_PTR(-ENETUNREACH);
1326 	}
1327 	if (dst->dev == dev) { /* is this necessary? */
1328 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1329 		dst_release(dst);
1330 		return ERR_PTR(-ELOOP);
1331 	}
1332 #ifdef CONFIG_DST_CACHE
1333 	if (use_cache)
1334 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1335 #endif
1336 	*saddr = fl6.saddr;
1337 	return dst;
1338 }
1339 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1340 
1341 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1342 					       gfp_t gfp)
1343 {
1344 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1345 }
1346 
1347 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1348 						gfp_t gfp)
1349 {
1350 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1351 }
1352 
1353 static void ip6_append_data_mtu(unsigned int *mtu,
1354 				int *maxfraglen,
1355 				unsigned int fragheaderlen,
1356 				struct sk_buff *skb,
1357 				struct rt6_info *rt,
1358 				unsigned int orig_mtu)
1359 {
1360 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1361 		if (!skb) {
1362 			/* first fragment, reserve header_len */
1363 			*mtu = orig_mtu - rt->dst.header_len;
1364 
1365 		} else {
1366 			/*
1367 			 * this fragment is not first, the headers
1368 			 * space is regarded as data space.
1369 			 */
1370 			*mtu = orig_mtu;
1371 		}
1372 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1373 			      + fragheaderlen - sizeof(struct frag_hdr);
1374 	}
1375 }
1376 
1377 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1378 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1379 			  struct rt6_info *rt)
1380 {
1381 	struct ipv6_pinfo *np = inet6_sk(sk);
1382 	unsigned int mtu;
1383 	struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1384 
1385 	/* callers pass dst together with a reference, set it first so
1386 	 * ip6_cork_release() can put it down even in case of an error.
1387 	 */
1388 	cork->base.dst = &rt->dst;
1389 
1390 	/*
1391 	 * setup for corking
1392 	 */
1393 	if (opt) {
1394 		if (WARN_ON(v6_cork->opt))
1395 			return -EINVAL;
1396 
1397 		nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1398 		if (unlikely(!nopt))
1399 			return -ENOBUFS;
1400 
1401 		nopt->tot_len = sizeof(*opt);
1402 		nopt->opt_flen = opt->opt_flen;
1403 		nopt->opt_nflen = opt->opt_nflen;
1404 
1405 		nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1406 		if (opt->dst0opt && !nopt->dst0opt)
1407 			return -ENOBUFS;
1408 
1409 		nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1410 		if (opt->dst1opt && !nopt->dst1opt)
1411 			return -ENOBUFS;
1412 
1413 		nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1414 		if (opt->hopopt && !nopt->hopopt)
1415 			return -ENOBUFS;
1416 
1417 		nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1418 		if (opt->srcrt && !nopt->srcrt)
1419 			return -ENOBUFS;
1420 
1421 		/* need source address above miyazawa*/
1422 	}
1423 	v6_cork->hop_limit = ipc6->hlimit;
1424 	v6_cork->tclass = ipc6->tclass;
1425 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1426 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1427 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1428 	else
1429 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1430 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1431 	if (np->frag_size < mtu) {
1432 		if (np->frag_size)
1433 			mtu = np->frag_size;
1434 	}
1435 	cork->base.fragsize = mtu;
1436 	cork->base.gso_size = ipc6->gso_size;
1437 	cork->base.tx_flags = 0;
1438 	cork->base.mark = ipc6->sockc.mark;
1439 	sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1440 
1441 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1442 		cork->base.flags |= IPCORK_ALLFRAG;
1443 	cork->base.length = 0;
1444 
1445 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1446 
1447 	return 0;
1448 }
1449 
1450 static int __ip6_append_data(struct sock *sk,
1451 			     struct sk_buff_head *queue,
1452 			     struct inet_cork_full *cork_full,
1453 			     struct inet6_cork *v6_cork,
1454 			     struct page_frag *pfrag,
1455 			     int getfrag(void *from, char *to, int offset,
1456 					 int len, int odd, struct sk_buff *skb),
1457 			     void *from, size_t length, int transhdrlen,
1458 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1459 {
1460 	struct sk_buff *skb, *skb_prev = NULL;
1461 	struct inet_cork *cork = &cork_full->base;
1462 	struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1463 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1464 	struct ubuf_info *uarg = NULL;
1465 	int exthdrlen = 0;
1466 	int dst_exthdrlen = 0;
1467 	int hh_len;
1468 	int copy;
1469 	int err;
1470 	int offset = 0;
1471 	bool zc = false;
1472 	u32 tskey = 0;
1473 	struct rt6_info *rt = (struct rt6_info *)cork->dst;
1474 	struct ipv6_txoptions *opt = v6_cork->opt;
1475 	int csummode = CHECKSUM_NONE;
1476 	unsigned int maxnonfragsize, headersize;
1477 	unsigned int wmem_alloc_delta = 0;
1478 	bool paged, extra_uref = false;
1479 
1480 	skb = skb_peek_tail(queue);
1481 	if (!skb) {
1482 		exthdrlen = opt ? opt->opt_flen : 0;
1483 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1484 	}
1485 
1486 	paged = !!cork->gso_size;
1487 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1488 	orig_mtu = mtu;
1489 
1490 	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1491 	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1492 		tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1493 
1494 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1495 
1496 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1497 			(opt ? opt->opt_nflen : 0);
1498 
1499 	headersize = sizeof(struct ipv6hdr) +
1500 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1501 		     (dst_allfrag(&rt->dst) ?
1502 		      sizeof(struct frag_hdr) : 0) +
1503 		     rt->rt6i_nfheader_len;
1504 
1505 	if (mtu <= fragheaderlen ||
1506 	    ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1507 		goto emsgsize;
1508 
1509 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1510 		     sizeof(struct frag_hdr);
1511 
1512 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1513 	 * the first fragment
1514 	 */
1515 	if (headersize + transhdrlen > mtu)
1516 		goto emsgsize;
1517 
1518 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1519 	    (sk->sk_protocol == IPPROTO_UDP ||
1520 	     sk->sk_protocol == IPPROTO_ICMPV6 ||
1521 	     sk->sk_protocol == IPPROTO_RAW)) {
1522 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1523 				sizeof(struct ipv6hdr));
1524 		goto emsgsize;
1525 	}
1526 
1527 	if (ip6_sk_ignore_df(sk))
1528 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1529 	else
1530 		maxnonfragsize = mtu;
1531 
1532 	if (cork->length + length > maxnonfragsize - headersize) {
1533 emsgsize:
1534 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1535 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1536 		return -EMSGSIZE;
1537 	}
1538 
1539 	/* CHECKSUM_PARTIAL only with no extension headers and when
1540 	 * we are not going to fragment
1541 	 */
1542 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1543 	    headersize == sizeof(struct ipv6hdr) &&
1544 	    length <= mtu - headersize &&
1545 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1546 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1547 		csummode = CHECKSUM_PARTIAL;
1548 
1549 	if ((flags & MSG_ZEROCOPY) && length) {
1550 		struct msghdr *msg = from;
1551 
1552 		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1553 			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1554 				return -EINVAL;
1555 
1556 			/* Leave uarg NULL if can't zerocopy, callers should
1557 			 * be able to handle it.
1558 			 */
1559 			if ((rt->dst.dev->features & NETIF_F_SG) &&
1560 			    csummode == CHECKSUM_PARTIAL) {
1561 				paged = true;
1562 				zc = true;
1563 				uarg = msg->msg_ubuf;
1564 			}
1565 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1566 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1567 			if (!uarg)
1568 				return -ENOBUFS;
1569 			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1570 			if (rt->dst.dev->features & NETIF_F_SG &&
1571 			    csummode == CHECKSUM_PARTIAL) {
1572 				paged = true;
1573 				zc = true;
1574 			} else {
1575 				uarg_to_msgzc(uarg)->zerocopy = 0;
1576 				skb_zcopy_set(skb, uarg, &extra_uref);
1577 			}
1578 		}
1579 	}
1580 
1581 	/*
1582 	 * Let's try using as much space as possible.
1583 	 * Use MTU if total length of the message fits into the MTU.
1584 	 * Otherwise, we need to reserve fragment header and
1585 	 * fragment alignment (= 8-15 octects, in total).
1586 	 *
1587 	 * Note that we may need to "move" the data from the tail
1588 	 * of the buffer to the new fragment when we split
1589 	 * the message.
1590 	 *
1591 	 * FIXME: It may be fragmented into multiple chunks
1592 	 *        at once if non-fragmentable extension headers
1593 	 *        are too large.
1594 	 * --yoshfuji
1595 	 */
1596 
1597 	cork->length += length;
1598 	if (!skb)
1599 		goto alloc_new_skb;
1600 
1601 	while (length > 0) {
1602 		/* Check if the remaining data fits into current packet. */
1603 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1604 		if (copy < length)
1605 			copy = maxfraglen - skb->len;
1606 
1607 		if (copy <= 0) {
1608 			char *data;
1609 			unsigned int datalen;
1610 			unsigned int fraglen;
1611 			unsigned int fraggap;
1612 			unsigned int alloclen, alloc_extra;
1613 			unsigned int pagedlen;
1614 alloc_new_skb:
1615 			/* There's no room in the current skb */
1616 			if (skb)
1617 				fraggap = skb->len - maxfraglen;
1618 			else
1619 				fraggap = 0;
1620 			/* update mtu and maxfraglen if necessary */
1621 			if (!skb || !skb_prev)
1622 				ip6_append_data_mtu(&mtu, &maxfraglen,
1623 						    fragheaderlen, skb, rt,
1624 						    orig_mtu);
1625 
1626 			skb_prev = skb;
1627 
1628 			/*
1629 			 * If remaining data exceeds the mtu,
1630 			 * we know we need more fragment(s).
1631 			 */
1632 			datalen = length + fraggap;
1633 
1634 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1635 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1636 			fraglen = datalen + fragheaderlen;
1637 			pagedlen = 0;
1638 
1639 			alloc_extra = hh_len;
1640 			alloc_extra += dst_exthdrlen;
1641 			alloc_extra += rt->dst.trailer_len;
1642 
1643 			/* We just reserve space for fragment header.
1644 			 * Note: this may be overallocation if the message
1645 			 * (without MSG_MORE) fits into the MTU.
1646 			 */
1647 			alloc_extra += sizeof(struct frag_hdr);
1648 
1649 			if ((flags & MSG_MORE) &&
1650 			    !(rt->dst.dev->features&NETIF_F_SG))
1651 				alloclen = mtu;
1652 			else if (!paged &&
1653 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1654 				  !(rt->dst.dev->features & NETIF_F_SG)))
1655 				alloclen = fraglen;
1656 			else {
1657 				alloclen = fragheaderlen + transhdrlen;
1658 				pagedlen = datalen - transhdrlen;
1659 			}
1660 			alloclen += alloc_extra;
1661 
1662 			if (datalen != length + fraggap) {
1663 				/*
1664 				 * this is not the last fragment, the trailer
1665 				 * space is regarded as data space.
1666 				 */
1667 				datalen += rt->dst.trailer_len;
1668 			}
1669 
1670 			fraglen = datalen + fragheaderlen;
1671 
1672 			copy = datalen - transhdrlen - fraggap - pagedlen;
1673 			if (copy < 0) {
1674 				err = -EINVAL;
1675 				goto error;
1676 			}
1677 			if (transhdrlen) {
1678 				skb = sock_alloc_send_skb(sk, alloclen,
1679 						(flags & MSG_DONTWAIT), &err);
1680 			} else {
1681 				skb = NULL;
1682 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1683 				    2 * sk->sk_sndbuf)
1684 					skb = alloc_skb(alloclen,
1685 							sk->sk_allocation);
1686 				if (unlikely(!skb))
1687 					err = -ENOBUFS;
1688 			}
1689 			if (!skb)
1690 				goto error;
1691 			/*
1692 			 *	Fill in the control structures
1693 			 */
1694 			skb->protocol = htons(ETH_P_IPV6);
1695 			skb->ip_summed = csummode;
1696 			skb->csum = 0;
1697 			/* reserve for fragmentation and ipsec header */
1698 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1699 				    dst_exthdrlen);
1700 
1701 			/*
1702 			 *	Find where to start putting bytes
1703 			 */
1704 			data = skb_put(skb, fraglen - pagedlen);
1705 			skb_set_network_header(skb, exthdrlen);
1706 			data += fragheaderlen;
1707 			skb->transport_header = (skb->network_header +
1708 						 fragheaderlen);
1709 			if (fraggap) {
1710 				skb->csum = skb_copy_and_csum_bits(
1711 					skb_prev, maxfraglen,
1712 					data + transhdrlen, fraggap);
1713 				skb_prev->csum = csum_sub(skb_prev->csum,
1714 							  skb->csum);
1715 				data += fraggap;
1716 				pskb_trim_unique(skb_prev, maxfraglen);
1717 			}
1718 			if (copy > 0 &&
1719 			    getfrag(from, data + transhdrlen, offset,
1720 				    copy, fraggap, skb) < 0) {
1721 				err = -EFAULT;
1722 				kfree_skb(skb);
1723 				goto error;
1724 			}
1725 
1726 			offset += copy;
1727 			length -= copy + transhdrlen;
1728 			transhdrlen = 0;
1729 			exthdrlen = 0;
1730 			dst_exthdrlen = 0;
1731 
1732 			/* Only the initial fragment is time stamped */
1733 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1734 			cork->tx_flags = 0;
1735 			skb_shinfo(skb)->tskey = tskey;
1736 			tskey = 0;
1737 			skb_zcopy_set(skb, uarg, &extra_uref);
1738 
1739 			if ((flags & MSG_CONFIRM) && !skb_prev)
1740 				skb_set_dst_pending_confirm(skb, 1);
1741 
1742 			/*
1743 			 * Put the packet on the pending queue
1744 			 */
1745 			if (!skb->destructor) {
1746 				skb->destructor = sock_wfree;
1747 				skb->sk = sk;
1748 				wmem_alloc_delta += skb->truesize;
1749 			}
1750 			__skb_queue_tail(queue, skb);
1751 			continue;
1752 		}
1753 
1754 		if (copy > length)
1755 			copy = length;
1756 
1757 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1758 		    skb_tailroom(skb) >= copy) {
1759 			unsigned int off;
1760 
1761 			off = skb->len;
1762 			if (getfrag(from, skb_put(skb, copy),
1763 						offset, copy, off, skb) < 0) {
1764 				__skb_trim(skb, off);
1765 				err = -EFAULT;
1766 				goto error;
1767 			}
1768 		} else if (!zc) {
1769 			int i = skb_shinfo(skb)->nr_frags;
1770 
1771 			err = -ENOMEM;
1772 			if (!sk_page_frag_refill(sk, pfrag))
1773 				goto error;
1774 
1775 			skb_zcopy_downgrade_managed(skb);
1776 			if (!skb_can_coalesce(skb, i, pfrag->page,
1777 					      pfrag->offset)) {
1778 				err = -EMSGSIZE;
1779 				if (i == MAX_SKB_FRAGS)
1780 					goto error;
1781 
1782 				__skb_fill_page_desc(skb, i, pfrag->page,
1783 						     pfrag->offset, 0);
1784 				skb_shinfo(skb)->nr_frags = ++i;
1785 				get_page(pfrag->page);
1786 			}
1787 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1788 			if (getfrag(from,
1789 				    page_address(pfrag->page) + pfrag->offset,
1790 				    offset, copy, skb->len, skb) < 0)
1791 				goto error_efault;
1792 
1793 			pfrag->offset += copy;
1794 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1795 			skb->len += copy;
1796 			skb->data_len += copy;
1797 			skb->truesize += copy;
1798 			wmem_alloc_delta += copy;
1799 		} else {
1800 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1801 			if (err < 0)
1802 				goto error;
1803 		}
1804 		offset += copy;
1805 		length -= copy;
1806 	}
1807 
1808 	if (wmem_alloc_delta)
1809 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1810 	return 0;
1811 
1812 error_efault:
1813 	err = -EFAULT;
1814 error:
1815 	net_zcopy_put_abort(uarg, extra_uref);
1816 	cork->length -= length;
1817 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1818 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1819 	return err;
1820 }
1821 
1822 int ip6_append_data(struct sock *sk,
1823 		    int getfrag(void *from, char *to, int offset, int len,
1824 				int odd, struct sk_buff *skb),
1825 		    void *from, size_t length, int transhdrlen,
1826 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1827 		    struct rt6_info *rt, unsigned int flags)
1828 {
1829 	struct inet_sock *inet = inet_sk(sk);
1830 	struct ipv6_pinfo *np = inet6_sk(sk);
1831 	int exthdrlen;
1832 	int err;
1833 
1834 	if (flags&MSG_PROBE)
1835 		return 0;
1836 	if (skb_queue_empty(&sk->sk_write_queue)) {
1837 		/*
1838 		 * setup for corking
1839 		 */
1840 		dst_hold(&rt->dst);
1841 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1842 				     ipc6, rt);
1843 		if (err)
1844 			return err;
1845 
1846 		inet->cork.fl.u.ip6 = *fl6;
1847 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1848 		length += exthdrlen;
1849 		transhdrlen += exthdrlen;
1850 	} else {
1851 		transhdrlen = 0;
1852 	}
1853 
1854 	return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1855 				 &np->cork, sk_page_frag(sk), getfrag,
1856 				 from, length, transhdrlen, flags, ipc6);
1857 }
1858 EXPORT_SYMBOL_GPL(ip6_append_data);
1859 
1860 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1861 {
1862 	struct dst_entry *dst = cork->base.dst;
1863 
1864 	cork->base.dst = NULL;
1865 	cork->base.flags &= ~IPCORK_ALLFRAG;
1866 	skb_dst_set(skb, dst);
1867 }
1868 
1869 static void ip6_cork_release(struct inet_cork_full *cork,
1870 			     struct inet6_cork *v6_cork)
1871 {
1872 	if (v6_cork->opt) {
1873 		struct ipv6_txoptions *opt = v6_cork->opt;
1874 
1875 		kfree(opt->dst0opt);
1876 		kfree(opt->dst1opt);
1877 		kfree(opt->hopopt);
1878 		kfree(opt->srcrt);
1879 		kfree(opt);
1880 		v6_cork->opt = NULL;
1881 	}
1882 
1883 	if (cork->base.dst) {
1884 		dst_release(cork->base.dst);
1885 		cork->base.dst = NULL;
1886 		cork->base.flags &= ~IPCORK_ALLFRAG;
1887 	}
1888 }
1889 
1890 struct sk_buff *__ip6_make_skb(struct sock *sk,
1891 			       struct sk_buff_head *queue,
1892 			       struct inet_cork_full *cork,
1893 			       struct inet6_cork *v6_cork)
1894 {
1895 	struct sk_buff *skb, *tmp_skb;
1896 	struct sk_buff **tail_skb;
1897 	struct in6_addr *final_dst;
1898 	struct ipv6_pinfo *np = inet6_sk(sk);
1899 	struct net *net = sock_net(sk);
1900 	struct ipv6hdr *hdr;
1901 	struct ipv6_txoptions *opt = v6_cork->opt;
1902 	struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1903 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1904 	unsigned char proto = fl6->flowi6_proto;
1905 
1906 	skb = __skb_dequeue(queue);
1907 	if (!skb)
1908 		goto out;
1909 	tail_skb = &(skb_shinfo(skb)->frag_list);
1910 
1911 	/* move skb->data to ip header from ext header */
1912 	if (skb->data < skb_network_header(skb))
1913 		__skb_pull(skb, skb_network_offset(skb));
1914 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1915 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1916 		*tail_skb = tmp_skb;
1917 		tail_skb = &(tmp_skb->next);
1918 		skb->len += tmp_skb->len;
1919 		skb->data_len += tmp_skb->len;
1920 		skb->truesize += tmp_skb->truesize;
1921 		tmp_skb->destructor = NULL;
1922 		tmp_skb->sk = NULL;
1923 	}
1924 
1925 	/* Allow local fragmentation. */
1926 	skb->ignore_df = ip6_sk_ignore_df(sk);
1927 	__skb_pull(skb, skb_network_header_len(skb));
1928 
1929 	final_dst = &fl6->daddr;
1930 	if (opt && opt->opt_flen)
1931 		ipv6_push_frag_opts(skb, opt, &proto);
1932 	if (opt && opt->opt_nflen)
1933 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1934 
1935 	skb_push(skb, sizeof(struct ipv6hdr));
1936 	skb_reset_network_header(skb);
1937 	hdr = ipv6_hdr(skb);
1938 
1939 	ip6_flow_hdr(hdr, v6_cork->tclass,
1940 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
1941 					ip6_autoflowlabel(net, np), fl6));
1942 	hdr->hop_limit = v6_cork->hop_limit;
1943 	hdr->nexthdr = proto;
1944 	hdr->saddr = fl6->saddr;
1945 	hdr->daddr = *final_dst;
1946 
1947 	skb->priority = sk->sk_priority;
1948 	skb->mark = cork->base.mark;
1949 	skb->tstamp = cork->base.transmit_time;
1950 
1951 	ip6_cork_steal_dst(skb, cork);
1952 	IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1953 	if (proto == IPPROTO_ICMPV6) {
1954 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1955 
1956 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1957 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1958 	}
1959 
1960 	ip6_cork_release(cork, v6_cork);
1961 out:
1962 	return skb;
1963 }
1964 
1965 int ip6_send_skb(struct sk_buff *skb)
1966 {
1967 	struct net *net = sock_net(skb->sk);
1968 	struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1969 	int err;
1970 
1971 	err = ip6_local_out(net, skb->sk, skb);
1972 	if (err) {
1973 		if (err > 0)
1974 			err = net_xmit_errno(err);
1975 		if (err)
1976 			IP6_INC_STATS(net, rt->rt6i_idev,
1977 				      IPSTATS_MIB_OUTDISCARDS);
1978 	}
1979 
1980 	return err;
1981 }
1982 
1983 int ip6_push_pending_frames(struct sock *sk)
1984 {
1985 	struct sk_buff *skb;
1986 
1987 	skb = ip6_finish_skb(sk);
1988 	if (!skb)
1989 		return 0;
1990 
1991 	return ip6_send_skb(skb);
1992 }
1993 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1994 
1995 static void __ip6_flush_pending_frames(struct sock *sk,
1996 				       struct sk_buff_head *queue,
1997 				       struct inet_cork_full *cork,
1998 				       struct inet6_cork *v6_cork)
1999 {
2000 	struct sk_buff *skb;
2001 
2002 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2003 		if (skb_dst(skb))
2004 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2005 				      IPSTATS_MIB_OUTDISCARDS);
2006 		kfree_skb(skb);
2007 	}
2008 
2009 	ip6_cork_release(cork, v6_cork);
2010 }
2011 
2012 void ip6_flush_pending_frames(struct sock *sk)
2013 {
2014 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2015 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2016 }
2017 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2018 
2019 struct sk_buff *ip6_make_skb(struct sock *sk,
2020 			     int getfrag(void *from, char *to, int offset,
2021 					 int len, int odd, struct sk_buff *skb),
2022 			     void *from, size_t length, int transhdrlen,
2023 			     struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2024 			     unsigned int flags, struct inet_cork_full *cork)
2025 {
2026 	struct inet6_cork v6_cork;
2027 	struct sk_buff_head queue;
2028 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2029 	int err;
2030 
2031 	if (flags & MSG_PROBE) {
2032 		dst_release(&rt->dst);
2033 		return NULL;
2034 	}
2035 
2036 	__skb_queue_head_init(&queue);
2037 
2038 	cork->base.flags = 0;
2039 	cork->base.addr = 0;
2040 	cork->base.opt = NULL;
2041 	v6_cork.opt = NULL;
2042 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2043 	if (err) {
2044 		ip6_cork_release(cork, &v6_cork);
2045 		return ERR_PTR(err);
2046 	}
2047 	if (ipc6->dontfrag < 0)
2048 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2049 
2050 	err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2051 				&current->task_frag, getfrag, from,
2052 				length + exthdrlen, transhdrlen + exthdrlen,
2053 				flags, ipc6);
2054 	if (err) {
2055 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2056 		return ERR_PTR(err);
2057 	}
2058 
2059 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2060 }
2061