xref: /openbmc/linux/net/ipv6/ip6_output.c (revision 278002edb19bce2c628fafb0af936e77000f3a5b)
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/gso.h>
46 #include <net/ipv6.h>
47 #include <net/ndisc.h>
48 #include <net/protocol.h>
49 #include <net/ip6_route.h>
50 #include <net/addrconf.h>
51 #include <net/rawv6.h>
52 #include <net/icmp.h>
53 #include <net/xfrm.h>
54 #include <net/checksum.h>
55 #include <linux/mroute6.h>
56 #include <net/l3mdev.h>
57 #include <net/lwtunnel.h>
58 #include <net/ip_tunnels.h>
59 
ip6_finish_output2(struct net * net,struct sock * sk,struct sk_buff * skb)60 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
61 {
62 	struct dst_entry *dst = skb_dst(skb);
63 	struct net_device *dev = dst->dev;
64 	struct inet6_dev *idev = ip6_dst_idev(dst);
65 	unsigned int hh_len = LL_RESERVED_SPACE(dev);
66 	const struct in6_addr *daddr, *nexthop;
67 	struct ipv6hdr *hdr;
68 	struct neighbour *neigh;
69 	int ret;
70 
71 	/* Be paranoid, rather than too clever. */
72 	if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
73 		/* Make sure idev stays alive */
74 		rcu_read_lock();
75 		skb = skb_expand_head(skb, hh_len);
76 		if (!skb) {
77 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
78 			rcu_read_unlock();
79 			return -ENOMEM;
80 		}
81 		rcu_read_unlock();
82 	}
83 
84 	hdr = ipv6_hdr(skb);
85 	daddr = &hdr->daddr;
86 	if (ipv6_addr_is_multicast(daddr)) {
87 		if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
88 		    ((mroute6_is_socket(net, skb) &&
89 		     !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
90 		     ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
91 			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
92 
93 			/* Do not check for IFF_ALLMULTI; multicast routing
94 			   is not supported in any case.
95 			 */
96 			if (newskb)
97 				NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
98 					net, sk, newskb, NULL, newskb->dev,
99 					dev_loopback_xmit);
100 
101 			if (hdr->hop_limit == 0) {
102 				IP6_INC_STATS(net, idev,
103 					      IPSTATS_MIB_OUTDISCARDS);
104 				kfree_skb(skb);
105 				return 0;
106 			}
107 		}
108 
109 		IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
110 		if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
111 		    !(dev->flags & IFF_LOOPBACK)) {
112 			kfree_skb(skb);
113 			return 0;
114 		}
115 	}
116 
117 	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
118 		int res = lwtunnel_xmit(skb);
119 
120 		if (res != LWTUNNEL_XMIT_CONTINUE)
121 			return res;
122 	}
123 
124 	IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
125 
126 	rcu_read_lock();
127 	nexthop = rt6_nexthop(dst_rt6_info(dst), daddr);
128 	neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
129 
130 	if (unlikely(IS_ERR_OR_NULL(neigh))) {
131 		if (unlikely(!neigh))
132 			neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
133 		if (IS_ERR(neigh)) {
134 			rcu_read_unlock();
135 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
136 			kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
137 			return -EINVAL;
138 		}
139 	}
140 	sock_confirm_neigh(skb, neigh);
141 	ret = neigh_output(neigh, skb, false);
142 	rcu_read_unlock();
143 	return ret;
144 }
145 
146 static int
ip6_finish_output_gso_slowpath_drop(struct net * net,struct sock * sk,struct sk_buff * skb,unsigned int mtu)147 ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
148 				    struct sk_buff *skb, unsigned int mtu)
149 {
150 	struct sk_buff *segs, *nskb;
151 	netdev_features_t features;
152 	int ret = 0;
153 
154 	/* Please see corresponding comment in ip_finish_output_gso
155 	 * describing the cases where GSO segment length exceeds the
156 	 * egress MTU.
157 	 */
158 	features = netif_skb_features(skb);
159 	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
160 	if (IS_ERR_OR_NULL(segs)) {
161 		kfree_skb(skb);
162 		return -ENOMEM;
163 	}
164 
165 	consume_skb(skb);
166 
167 	skb_list_walk_safe(segs, segs, nskb) {
168 		int err;
169 
170 		skb_mark_not_on_list(segs);
171 		/* Last GSO segment can be smaller than gso_size (and MTU).
172 		 * Adding a fragment header would produce an "atomic fragment",
173 		 * which is considered harmful (RFC-8021). Avoid that.
174 		 */
175 		err = segs->len > mtu ?
176 			ip6_fragment(net, sk, segs, ip6_finish_output2) :
177 			ip6_finish_output2(net, sk, segs);
178 		if (err && ret == 0)
179 			ret = err;
180 	}
181 
182 	return ret;
183 }
184 
__ip6_finish_output(struct net * net,struct sock * sk,struct sk_buff * skb)185 static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
186 {
187 	unsigned int mtu;
188 
189 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
190 	/* Policy lookup after SNAT yielded a new policy */
191 	if (skb_dst(skb)->xfrm) {
192 		IP6CB(skb)->flags |= IP6SKB_REROUTED;
193 		return dst_output(net, sk, skb);
194 	}
195 #endif
196 
197 	mtu = ip6_skb_dst_mtu(skb);
198 	if (skb_is_gso(skb) &&
199 	    !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
200 	    !skb_gso_validate_network_len(skb, mtu))
201 		return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
202 
203 	if ((skb->len > mtu && !skb_is_gso(skb)) ||
204 	    dst_allfrag(skb_dst(skb)) ||
205 	    (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
206 		return ip6_fragment(net, sk, skb, ip6_finish_output2);
207 	else
208 		return ip6_finish_output2(net, sk, skb);
209 }
210 
ip6_finish_output(struct net * net,struct sock * sk,struct sk_buff * skb)211 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
212 {
213 	int ret;
214 
215 	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
216 	switch (ret) {
217 	case NET_XMIT_SUCCESS:
218 	case NET_XMIT_CN:
219 		return __ip6_finish_output(net, sk, skb) ? : ret;
220 	default:
221 		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
222 		return ret;
223 	}
224 }
225 
ip6_output(struct net * net,struct sock * sk,struct sk_buff * skb)226 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
227 {
228 	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
229 	struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
230 
231 	skb->protocol = htons(ETH_P_IPV6);
232 	skb->dev = dev;
233 
234 	if (unlikely(!idev || READ_ONCE(idev->cnf.disable_ipv6))) {
235 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
236 		kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
237 		return 0;
238 	}
239 
240 	return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
241 			    net, sk, skb, indev, dev,
242 			    ip6_finish_output,
243 			    !(IP6CB(skb)->flags & IP6SKB_REROUTED));
244 }
245 EXPORT_SYMBOL(ip6_output);
246 
ip6_autoflowlabel(struct net * net,const struct ipv6_pinfo * np)247 bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
248 {
249 	if (!np->autoflowlabel_set)
250 		return ip6_default_np_autolabel(net);
251 	else
252 		return np->autoflowlabel;
253 }
254 
255 /*
256  * xmit an sk_buff (used by TCP, SCTP and DCCP)
257  * Note : socket lock is not held for SYNACK packets, but might be modified
258  * by calls to skb_set_owner_w() and ipv6_local_error(),
259  * which are using proper atomic operations or spinlocks.
260  */
ip6_xmit(const struct sock * sk,struct sk_buff * skb,struct flowi6 * fl6,__u32 mark,struct ipv6_txoptions * opt,int tclass,u32 priority)261 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
262 	     __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
263 {
264 	struct net *net = sock_net(sk);
265 	const struct ipv6_pinfo *np = inet6_sk(sk);
266 	struct in6_addr *first_hop = &fl6->daddr;
267 	struct dst_entry *dst = skb_dst(skb);
268 	struct net_device *dev = dst->dev;
269 	struct inet6_dev *idev = ip6_dst_idev(dst);
270 	struct hop_jumbo_hdr *hop_jumbo;
271 	int hoplen = sizeof(*hop_jumbo);
272 	unsigned int head_room;
273 	struct ipv6hdr *hdr;
274 	u8  proto = fl6->flowi6_proto;
275 	int seg_len = skb->len;
276 	int hlimit = -1;
277 	u32 mtu;
278 
279 	head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
280 	if (opt)
281 		head_room += opt->opt_nflen + opt->opt_flen;
282 
283 	if (unlikely(head_room > skb_headroom(skb))) {
284 		/* Make sure idev stays alive */
285 		rcu_read_lock();
286 		skb = skb_expand_head(skb, head_room);
287 		if (!skb) {
288 			IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
289 			rcu_read_unlock();
290 			return -ENOBUFS;
291 		}
292 		rcu_read_unlock();
293 	}
294 
295 	if (opt) {
296 		seg_len += opt->opt_nflen + opt->opt_flen;
297 
298 		if (opt->opt_flen)
299 			ipv6_push_frag_opts(skb, opt, &proto);
300 
301 		if (opt->opt_nflen)
302 			ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
303 					     &fl6->saddr);
304 	}
305 
306 	if (unlikely(seg_len > IPV6_MAXPLEN)) {
307 		hop_jumbo = skb_push(skb, hoplen);
308 
309 		hop_jumbo->nexthdr = proto;
310 		hop_jumbo->hdrlen = 0;
311 		hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
312 		hop_jumbo->tlv_len = 4;
313 		hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
314 
315 		proto = IPPROTO_HOPOPTS;
316 		seg_len = 0;
317 		IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
318 	}
319 
320 	skb_push(skb, sizeof(struct ipv6hdr));
321 	skb_reset_network_header(skb);
322 	hdr = ipv6_hdr(skb);
323 
324 	/*
325 	 *	Fill in the IPv6 header
326 	 */
327 	if (np)
328 		hlimit = np->hop_limit;
329 	if (hlimit < 0)
330 		hlimit = ip6_dst_hoplimit(dst);
331 
332 	ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
333 				ip6_autoflowlabel(net, np), fl6));
334 
335 	hdr->payload_len = htons(seg_len);
336 	hdr->nexthdr = proto;
337 	hdr->hop_limit = hlimit;
338 
339 	hdr->saddr = fl6->saddr;
340 	hdr->daddr = *first_hop;
341 
342 	skb->protocol = htons(ETH_P_IPV6);
343 	skb->priority = priority;
344 	skb->mark = mark;
345 
346 	mtu = dst_mtu(dst);
347 	if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
348 		IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTREQUESTS);
349 
350 		/* if egress device is enslaved to an L3 master device pass the
351 		 * skb to its handler for processing
352 		 */
353 		skb = l3mdev_ip6_out((struct sock *)sk, skb);
354 		if (unlikely(!skb))
355 			return 0;
356 
357 		/* hooks should never assume socket lock is held.
358 		 * we promote our socket to non const
359 		 */
360 		return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
361 			       net, (struct sock *)sk, skb, NULL, dev,
362 			       dst_output);
363 	}
364 
365 	skb->dev = dev;
366 	/* ipv6_local_error() does not require socket lock,
367 	 * we promote our socket to non const
368 	 */
369 	ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
370 
371 	IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
372 	kfree_skb(skb);
373 	return -EMSGSIZE;
374 }
375 EXPORT_SYMBOL(ip6_xmit);
376 
ip6_call_ra_chain(struct sk_buff * skb,int sel)377 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
378 {
379 	struct ip6_ra_chain *ra;
380 	struct sock *last = NULL;
381 
382 	read_lock(&ip6_ra_lock);
383 	for (ra = ip6_ra_chain; ra; ra = ra->next) {
384 		struct sock *sk = ra->sk;
385 		if (sk && ra->sel == sel &&
386 		    (!sk->sk_bound_dev_if ||
387 		     sk->sk_bound_dev_if == skb->dev->ifindex)) {
388 			struct ipv6_pinfo *np = inet6_sk(sk);
389 
390 			if (np && np->rtalert_isolate &&
391 			    !net_eq(sock_net(sk), dev_net(skb->dev))) {
392 				continue;
393 			}
394 			if (last) {
395 				struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
396 				if (skb2)
397 					rawv6_rcv(last, skb2);
398 			}
399 			last = sk;
400 		}
401 	}
402 
403 	if (last) {
404 		rawv6_rcv(last, skb);
405 		read_unlock(&ip6_ra_lock);
406 		return 1;
407 	}
408 	read_unlock(&ip6_ra_lock);
409 	return 0;
410 }
411 
ip6_forward_proxy_check(struct sk_buff * skb)412 static int ip6_forward_proxy_check(struct sk_buff *skb)
413 {
414 	struct ipv6hdr *hdr = ipv6_hdr(skb);
415 	u8 nexthdr = hdr->nexthdr;
416 	__be16 frag_off;
417 	int offset;
418 
419 	if (ipv6_ext_hdr(nexthdr)) {
420 		offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
421 		if (offset < 0)
422 			return 0;
423 	} else
424 		offset = sizeof(struct ipv6hdr);
425 
426 	if (nexthdr == IPPROTO_ICMPV6) {
427 		struct icmp6hdr *icmp6;
428 
429 		if (!pskb_may_pull(skb, (skb_network_header(skb) +
430 					 offset + 1 - skb->data)))
431 			return 0;
432 
433 		icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
434 
435 		switch (icmp6->icmp6_type) {
436 		case NDISC_ROUTER_SOLICITATION:
437 		case NDISC_ROUTER_ADVERTISEMENT:
438 		case NDISC_NEIGHBOUR_SOLICITATION:
439 		case NDISC_NEIGHBOUR_ADVERTISEMENT:
440 		case NDISC_REDIRECT:
441 			/* For reaction involving unicast neighbor discovery
442 			 * message destined to the proxied address, pass it to
443 			 * input function.
444 			 */
445 			return 1;
446 		default:
447 			break;
448 		}
449 	}
450 
451 	/*
452 	 * The proxying router can't forward traffic sent to a link-local
453 	 * address, so signal the sender and discard the packet. This
454 	 * behavior is clarified by the MIPv6 specification.
455 	 */
456 	if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
457 		dst_link_failure(skb);
458 		return -1;
459 	}
460 
461 	return 0;
462 }
463 
ip6_forward_finish(struct net * net,struct sock * sk,struct sk_buff * skb)464 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
465 				     struct sk_buff *skb)
466 {
467 	struct dst_entry *dst = skb_dst(skb);
468 
469 	__IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
470 
471 #ifdef CONFIG_NET_SWITCHDEV
472 	if (skb->offload_l3_fwd_mark) {
473 		consume_skb(skb);
474 		return 0;
475 	}
476 #endif
477 
478 	skb_clear_tstamp(skb);
479 	return dst_output(net, sk, skb);
480 }
481 
ip6_pkt_too_big(const struct sk_buff * skb,unsigned int mtu)482 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
483 {
484 	if (skb->len <= mtu)
485 		return false;
486 
487 	/* ipv6 conntrack defrag sets max_frag_size + ignore_df */
488 	if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
489 		return true;
490 
491 	if (skb->ignore_df)
492 		return false;
493 
494 	if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
495 		return false;
496 
497 	return true;
498 }
499 
ip6_forward(struct sk_buff * skb)500 int ip6_forward(struct sk_buff *skb)
501 {
502 	struct dst_entry *dst = skb_dst(skb);
503 	struct ipv6hdr *hdr = ipv6_hdr(skb);
504 	struct inet6_skb_parm *opt = IP6CB(skb);
505 	struct net *net = dev_net(dst->dev);
506 	struct inet6_dev *idev;
507 	SKB_DR(reason);
508 	u32 mtu;
509 
510 	idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
511 	if (net->ipv6.devconf_all->forwarding == 0)
512 		goto error;
513 
514 	if (skb->pkt_type != PACKET_HOST)
515 		goto drop;
516 
517 	if (unlikely(skb->sk))
518 		goto drop;
519 
520 	if (skb_warn_if_lro(skb))
521 		goto drop;
522 
523 	if (!net->ipv6.devconf_all->disable_policy &&
524 	    (!idev || !idev->cnf.disable_policy) &&
525 	    !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
526 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
527 		goto drop;
528 	}
529 
530 	skb_forward_csum(skb);
531 
532 	/*
533 	 *	We DO NOT make any processing on
534 	 *	RA packets, pushing them to user level AS IS
535 	 *	without ane WARRANTY that application will be able
536 	 *	to interpret them. The reason is that we
537 	 *	cannot make anything clever here.
538 	 *
539 	 *	We are not end-node, so that if packet contains
540 	 *	AH/ESP, we cannot make anything.
541 	 *	Defragmentation also would be mistake, RA packets
542 	 *	cannot be fragmented, because there is no warranty
543 	 *	that different fragments will go along one path. --ANK
544 	 */
545 	if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
546 		if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
547 			return 0;
548 	}
549 
550 	/*
551 	 *	check and decrement ttl
552 	 */
553 	if (hdr->hop_limit <= 1) {
554 		icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
555 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
556 
557 		kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
558 		return -ETIMEDOUT;
559 	}
560 
561 	/* XXX: idev->cnf.proxy_ndp? */
562 	if (net->ipv6.devconf_all->proxy_ndp &&
563 	    pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
564 		int proxied = ip6_forward_proxy_check(skb);
565 		if (proxied > 0) {
566 			/* It's tempting to decrease the hop limit
567 			 * here by 1, as we do at the end of the
568 			 * function too.
569 			 *
570 			 * But that would be incorrect, as proxying is
571 			 * not forwarding.  The ip6_input function
572 			 * will handle this packet locally, and it
573 			 * depends on the hop limit being unchanged.
574 			 *
575 			 * One example is the NDP hop limit, that
576 			 * always has to stay 255, but other would be
577 			 * similar checks around RA packets, where the
578 			 * user can even change the desired limit.
579 			 */
580 			return ip6_input(skb);
581 		} else if (proxied < 0) {
582 			__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
583 			goto drop;
584 		}
585 	}
586 
587 	if (!xfrm6_route_forward(skb)) {
588 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
589 		SKB_DR_SET(reason, XFRM_POLICY);
590 		goto drop;
591 	}
592 	dst = skb_dst(skb);
593 
594 	/* IPv6 specs say nothing about it, but it is clear that we cannot
595 	   send redirects to source routed frames.
596 	   We don't send redirects to frames decapsulated from IPsec.
597 	 */
598 	if (IP6CB(skb)->iif == dst->dev->ifindex &&
599 	    opt->srcrt == 0 && !skb_sec_path(skb)) {
600 		struct in6_addr *target = NULL;
601 		struct inet_peer *peer;
602 		struct rt6_info *rt;
603 
604 		/*
605 		 *	incoming and outgoing devices are the same
606 		 *	send a redirect.
607 		 */
608 
609 		rt = dst_rt6_info(dst);
610 		if (rt->rt6i_flags & RTF_GATEWAY)
611 			target = &rt->rt6i_gateway;
612 		else
613 			target = &hdr->daddr;
614 
615 		peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
616 
617 		/* Limit redirects both by destination (here)
618 		   and by source (inside ndisc_send_redirect)
619 		 */
620 		if (inet_peer_xrlim_allow(peer, 1*HZ))
621 			ndisc_send_redirect(skb, target);
622 		if (peer)
623 			inet_putpeer(peer);
624 	} else {
625 		int addrtype = ipv6_addr_type(&hdr->saddr);
626 
627 		/* This check is security critical. */
628 		if (addrtype == IPV6_ADDR_ANY ||
629 		    addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
630 			goto error;
631 		if (addrtype & IPV6_ADDR_LINKLOCAL) {
632 			icmpv6_send(skb, ICMPV6_DEST_UNREACH,
633 				    ICMPV6_NOT_NEIGHBOUR, 0);
634 			goto error;
635 		}
636 	}
637 
638 	mtu = ip6_dst_mtu_maybe_forward(dst, true);
639 	if (mtu < IPV6_MIN_MTU)
640 		mtu = IPV6_MIN_MTU;
641 
642 	if (ip6_pkt_too_big(skb, mtu)) {
643 		/* Again, force OUTPUT device used as source address */
644 		skb->dev = dst->dev;
645 		icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
646 		__IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
647 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
648 				IPSTATS_MIB_FRAGFAILS);
649 		kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
650 		return -EMSGSIZE;
651 	}
652 
653 	if (skb_cow(skb, dst->dev->hard_header_len)) {
654 		__IP6_INC_STATS(net, ip6_dst_idev(dst),
655 				IPSTATS_MIB_OUTDISCARDS);
656 		goto drop;
657 	}
658 
659 	hdr = ipv6_hdr(skb);
660 
661 	/* Mangling hops number delayed to point after skb COW */
662 
663 	hdr->hop_limit--;
664 
665 	return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
666 		       net, NULL, skb, skb->dev, dst->dev,
667 		       ip6_forward_finish);
668 
669 error:
670 	__IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
671 	SKB_DR_SET(reason, IP_INADDRERRORS);
672 drop:
673 	kfree_skb_reason(skb, reason);
674 	return -EINVAL;
675 }
676 
ip6_copy_metadata(struct sk_buff * to,struct sk_buff * from)677 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
678 {
679 	to->pkt_type = from->pkt_type;
680 	to->priority = from->priority;
681 	to->protocol = from->protocol;
682 	skb_dst_drop(to);
683 	skb_dst_set(to, dst_clone(skb_dst(from)));
684 	to->dev = from->dev;
685 	to->mark = from->mark;
686 
687 	skb_copy_hash(to, from);
688 
689 #ifdef CONFIG_NET_SCHED
690 	to->tc_index = from->tc_index;
691 #endif
692 	nf_copy(to, from);
693 	skb_ext_copy(to, from);
694 	skb_copy_secmark(to, from);
695 }
696 
ip6_fraglist_init(struct sk_buff * skb,unsigned int hlen,u8 * prevhdr,u8 nexthdr,__be32 frag_id,struct ip6_fraglist_iter * iter)697 int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
698 		      u8 nexthdr, __be32 frag_id,
699 		      struct ip6_fraglist_iter *iter)
700 {
701 	unsigned int first_len;
702 	struct frag_hdr *fh;
703 
704 	/* BUILD HEADER */
705 	*prevhdr = NEXTHDR_FRAGMENT;
706 	iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
707 	if (!iter->tmp_hdr)
708 		return -ENOMEM;
709 
710 	iter->frag = skb_shinfo(skb)->frag_list;
711 	skb_frag_list_init(skb);
712 
713 	iter->offset = 0;
714 	iter->hlen = hlen;
715 	iter->frag_id = frag_id;
716 	iter->nexthdr = nexthdr;
717 
718 	__skb_pull(skb, hlen);
719 	fh = __skb_push(skb, sizeof(struct frag_hdr));
720 	__skb_push(skb, hlen);
721 	skb_reset_network_header(skb);
722 	memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
723 
724 	fh->nexthdr = nexthdr;
725 	fh->reserved = 0;
726 	fh->frag_off = htons(IP6_MF);
727 	fh->identification = frag_id;
728 
729 	first_len = skb_pagelen(skb);
730 	skb->data_len = first_len - skb_headlen(skb);
731 	skb->len = first_len;
732 	ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
733 
734 	return 0;
735 }
736 EXPORT_SYMBOL(ip6_fraglist_init);
737 
ip6_fraglist_prepare(struct sk_buff * skb,struct ip6_fraglist_iter * iter)738 void ip6_fraglist_prepare(struct sk_buff *skb,
739 			  struct ip6_fraglist_iter *iter)
740 {
741 	struct sk_buff *frag = iter->frag;
742 	unsigned int hlen = iter->hlen;
743 	struct frag_hdr *fh;
744 
745 	frag->ip_summed = CHECKSUM_NONE;
746 	skb_reset_transport_header(frag);
747 	fh = __skb_push(frag, sizeof(struct frag_hdr));
748 	__skb_push(frag, hlen);
749 	skb_reset_network_header(frag);
750 	memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
751 	iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
752 	fh->nexthdr = iter->nexthdr;
753 	fh->reserved = 0;
754 	fh->frag_off = htons(iter->offset);
755 	if (frag->next)
756 		fh->frag_off |= htons(IP6_MF);
757 	fh->identification = iter->frag_id;
758 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
759 	ip6_copy_metadata(frag, skb);
760 }
761 EXPORT_SYMBOL(ip6_fraglist_prepare);
762 
ip6_frag_init(struct sk_buff * skb,unsigned int hlen,unsigned int mtu,unsigned short needed_tailroom,int hdr_room,u8 * prevhdr,u8 nexthdr,__be32 frag_id,struct ip6_frag_state * state)763 void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
764 		   unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
765 		   u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
766 {
767 	state->prevhdr = prevhdr;
768 	state->nexthdr = nexthdr;
769 	state->frag_id = frag_id;
770 
771 	state->hlen = hlen;
772 	state->mtu = mtu;
773 
774 	state->left = skb->len - hlen;	/* Space per frame */
775 	state->ptr = hlen;		/* Where to start from */
776 
777 	state->hroom = hdr_room;
778 	state->troom = needed_tailroom;
779 
780 	state->offset = 0;
781 }
782 EXPORT_SYMBOL(ip6_frag_init);
783 
ip6_frag_next(struct sk_buff * skb,struct ip6_frag_state * state)784 struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
785 {
786 	u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
787 	struct sk_buff *frag;
788 	struct frag_hdr *fh;
789 	unsigned int len;
790 
791 	len = state->left;
792 	/* IF: it doesn't fit, use 'mtu' - the data space left */
793 	if (len > state->mtu)
794 		len = state->mtu;
795 	/* IF: we are not sending up to and including the packet end
796 	   then align the next start on an eight byte boundary */
797 	if (len < state->left)
798 		len &= ~7;
799 
800 	/* Allocate buffer */
801 	frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
802 			 state->hroom + state->troom, GFP_ATOMIC);
803 	if (!frag)
804 		return ERR_PTR(-ENOMEM);
805 
806 	/*
807 	 *	Set up data on packet
808 	 */
809 
810 	ip6_copy_metadata(frag, skb);
811 	skb_reserve(frag, state->hroom);
812 	skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
813 	skb_reset_network_header(frag);
814 	fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
815 	frag->transport_header = (frag->network_header + state->hlen +
816 				  sizeof(struct frag_hdr));
817 
818 	/*
819 	 *	Charge the memory for the fragment to any owner
820 	 *	it might possess
821 	 */
822 	if (skb->sk)
823 		skb_set_owner_w(frag, skb->sk);
824 
825 	/*
826 	 *	Copy the packet header into the new buffer.
827 	 */
828 	skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
829 
830 	fragnexthdr_offset = skb_network_header(frag);
831 	fragnexthdr_offset += prevhdr - skb_network_header(skb);
832 	*fragnexthdr_offset = NEXTHDR_FRAGMENT;
833 
834 	/*
835 	 *	Build fragment header.
836 	 */
837 	fh->nexthdr = state->nexthdr;
838 	fh->reserved = 0;
839 	fh->identification = state->frag_id;
840 
841 	/*
842 	 *	Copy a block of the IP datagram.
843 	 */
844 	BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
845 			     len));
846 	state->left -= len;
847 
848 	fh->frag_off = htons(state->offset);
849 	if (state->left > 0)
850 		fh->frag_off |= htons(IP6_MF);
851 	ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
852 
853 	state->ptr += len;
854 	state->offset += len;
855 
856 	return frag;
857 }
858 EXPORT_SYMBOL(ip6_frag_next);
859 
ip6_fragment(struct net * net,struct sock * sk,struct sk_buff * skb,int (* output)(struct net *,struct sock *,struct sk_buff *))860 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
861 		 int (*output)(struct net *, struct sock *, struct sk_buff *))
862 {
863 	struct sk_buff *frag;
864 	struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
865 	struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
866 				inet6_sk(skb->sk) : NULL;
867 	bool mono_delivery_time = skb->mono_delivery_time;
868 	struct ip6_frag_state state;
869 	unsigned int mtu, hlen, nexthdr_offset;
870 	ktime_t tstamp = skb->tstamp;
871 	int hroom, err = 0;
872 	__be32 frag_id;
873 	u8 *prevhdr, nexthdr = 0;
874 
875 	err = ip6_find_1stfragopt(skb, &prevhdr);
876 	if (err < 0)
877 		goto fail;
878 	hlen = err;
879 	nexthdr = *prevhdr;
880 	nexthdr_offset = prevhdr - skb_network_header(skb);
881 
882 	mtu = ip6_skb_dst_mtu(skb);
883 
884 	/* We must not fragment if the socket is set to force MTU discovery
885 	 * or if the skb it not generated by a local socket.
886 	 */
887 	if (unlikely(!skb->ignore_df && skb->len > mtu))
888 		goto fail_toobig;
889 
890 	if (IP6CB(skb)->frag_max_size) {
891 		if (IP6CB(skb)->frag_max_size > mtu)
892 			goto fail_toobig;
893 
894 		/* don't send fragments larger than what we received */
895 		mtu = IP6CB(skb)->frag_max_size;
896 		if (mtu < IPV6_MIN_MTU)
897 			mtu = IPV6_MIN_MTU;
898 	}
899 
900 	if (np && np->frag_size < mtu) {
901 		if (np->frag_size)
902 			mtu = np->frag_size;
903 	}
904 	if (mtu < hlen + sizeof(struct frag_hdr) + 8)
905 		goto fail_toobig;
906 	mtu -= hlen + sizeof(struct frag_hdr);
907 
908 	frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
909 				    &ipv6_hdr(skb)->saddr);
910 
911 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
912 	    (err = skb_checksum_help(skb)))
913 		goto fail;
914 
915 	prevhdr = skb_network_header(skb) + nexthdr_offset;
916 	hroom = LL_RESERVED_SPACE(rt->dst.dev);
917 	if (skb_has_frag_list(skb)) {
918 		unsigned int first_len = skb_pagelen(skb);
919 		struct ip6_fraglist_iter iter;
920 		struct sk_buff *frag2;
921 
922 		if (first_len - hlen > mtu ||
923 		    ((first_len - hlen) & 7) ||
924 		    skb_cloned(skb) ||
925 		    skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
926 			goto slow_path;
927 
928 		skb_walk_frags(skb, frag) {
929 			/* Correct geometry. */
930 			if (frag->len > mtu ||
931 			    ((frag->len & 7) && frag->next) ||
932 			    skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
933 				goto slow_path_clean;
934 
935 			/* Partially cloned skb? */
936 			if (skb_shared(frag))
937 				goto slow_path_clean;
938 
939 			BUG_ON(frag->sk);
940 			if (skb->sk) {
941 				frag->sk = skb->sk;
942 				frag->destructor = sock_wfree;
943 			}
944 			skb->truesize -= frag->truesize;
945 		}
946 
947 		err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
948 					&iter);
949 		if (err < 0)
950 			goto fail;
951 
952 		/* We prevent @rt from being freed. */
953 		rcu_read_lock();
954 
955 		for (;;) {
956 			/* Prepare header of the next frame,
957 			 * before previous one went down. */
958 			if (iter.frag)
959 				ip6_fraglist_prepare(skb, &iter);
960 
961 			skb_set_delivery_time(skb, tstamp, mono_delivery_time);
962 			err = output(net, sk, skb);
963 			if (!err)
964 				IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
965 					      IPSTATS_MIB_FRAGCREATES);
966 
967 			if (err || !iter.frag)
968 				break;
969 
970 			skb = ip6_fraglist_next(&iter);
971 		}
972 
973 		kfree(iter.tmp_hdr);
974 
975 		if (err == 0) {
976 			IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
977 				      IPSTATS_MIB_FRAGOKS);
978 			rcu_read_unlock();
979 			return 0;
980 		}
981 
982 		kfree_skb_list(iter.frag);
983 
984 		IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
985 			      IPSTATS_MIB_FRAGFAILS);
986 		rcu_read_unlock();
987 		return err;
988 
989 slow_path_clean:
990 		skb_walk_frags(skb, frag2) {
991 			if (frag2 == frag)
992 				break;
993 			frag2->sk = NULL;
994 			frag2->destructor = NULL;
995 			skb->truesize += frag2->truesize;
996 		}
997 	}
998 
999 slow_path:
1000 	/*
1001 	 *	Fragment the datagram.
1002 	 */
1003 
1004 	ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
1005 		      LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
1006 		      &state);
1007 
1008 	/*
1009 	 *	Keep copying data until we run out.
1010 	 */
1011 
1012 	while (state.left > 0) {
1013 		frag = ip6_frag_next(skb, &state);
1014 		if (IS_ERR(frag)) {
1015 			err = PTR_ERR(frag);
1016 			goto fail;
1017 		}
1018 
1019 		/*
1020 		 *	Put this fragment into the sending queue.
1021 		 */
1022 		skb_set_delivery_time(frag, tstamp, mono_delivery_time);
1023 		err = output(net, sk, frag);
1024 		if (err)
1025 			goto fail;
1026 
1027 		IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1028 			      IPSTATS_MIB_FRAGCREATES);
1029 	}
1030 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1031 		      IPSTATS_MIB_FRAGOKS);
1032 	consume_skb(skb);
1033 	return err;
1034 
1035 fail_toobig:
1036 	if (skb->sk && dst_allfrag(skb_dst(skb)))
1037 		sk_gso_disable(skb->sk);
1038 
1039 	icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1040 	err = -EMSGSIZE;
1041 
1042 fail:
1043 	IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1044 		      IPSTATS_MIB_FRAGFAILS);
1045 	kfree_skb(skb);
1046 	return err;
1047 }
1048 
ip6_rt_check(const struct rt6key * rt_key,const struct in6_addr * fl_addr,const struct in6_addr * addr_cache)1049 static inline int ip6_rt_check(const struct rt6key *rt_key,
1050 			       const struct in6_addr *fl_addr,
1051 			       const struct in6_addr *addr_cache)
1052 {
1053 	return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1054 		(!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1055 }
1056 
ip6_sk_dst_check(struct sock * sk,struct dst_entry * dst,const struct flowi6 * fl6)1057 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1058 					  struct dst_entry *dst,
1059 					  const struct flowi6 *fl6)
1060 {
1061 	struct ipv6_pinfo *np = inet6_sk(sk);
1062 	struct rt6_info *rt;
1063 
1064 	if (!dst)
1065 		goto out;
1066 
1067 	if (dst->ops->family != AF_INET6) {
1068 		dst_release(dst);
1069 		return NULL;
1070 	}
1071 
1072 	rt = dst_rt6_info(dst);
1073 	/* Yes, checking route validity in not connected
1074 	 * case is not very simple. Take into account,
1075 	 * that we do not support routing by source, TOS,
1076 	 * and MSG_DONTROUTE		--ANK (980726)
1077 	 *
1078 	 * 1. ip6_rt_check(): If route was host route,
1079 	 *    check that cached destination is current.
1080 	 *    If it is network route, we still may
1081 	 *    check its validity using saved pointer
1082 	 *    to the last used address: daddr_cache.
1083 	 *    We do not want to save whole address now,
1084 	 *    (because main consumer of this service
1085 	 *    is tcp, which has not this problem),
1086 	 *    so that the last trick works only on connected
1087 	 *    sockets.
1088 	 * 2. oif also should be the same.
1089 	 */
1090 	if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1091 #ifdef CONFIG_IPV6_SUBTREES
1092 	    ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1093 #endif
1094 	   (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1095 		dst_release(dst);
1096 		dst = NULL;
1097 	}
1098 
1099 out:
1100 	return dst;
1101 }
1102 
ip6_dst_lookup_tail(struct net * net,const struct sock * sk,struct dst_entry ** dst,struct flowi6 * fl6)1103 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1104 			       struct dst_entry **dst, struct flowi6 *fl6)
1105 {
1106 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1107 	struct neighbour *n;
1108 	struct rt6_info *rt;
1109 #endif
1110 	int err;
1111 	int flags = 0;
1112 
1113 	/* The correct way to handle this would be to do
1114 	 * ip6_route_get_saddr, and then ip6_route_output; however,
1115 	 * the route-specific preferred source forces the
1116 	 * ip6_route_output call _before_ ip6_route_get_saddr.
1117 	 *
1118 	 * In source specific routing (no src=any default route),
1119 	 * ip6_route_output will fail given src=any saddr, though, so
1120 	 * that's why we try it again later.
1121 	 */
1122 	if (ipv6_addr_any(&fl6->saddr)) {
1123 		struct fib6_info *from;
1124 		struct rt6_info *rt;
1125 
1126 		*dst = ip6_route_output(net, sk, fl6);
1127 		rt = (*dst)->error ? NULL : dst_rt6_info(*dst);
1128 
1129 		rcu_read_lock();
1130 		from = rt ? rcu_dereference(rt->from) : NULL;
1131 		err = ip6_route_get_saddr(net, from, &fl6->daddr,
1132 					  sk ? inet6_sk(sk)->srcprefs : 0,
1133 					  fl6->flowi6_l3mdev,
1134 					  &fl6->saddr);
1135 		rcu_read_unlock();
1136 
1137 		if (err)
1138 			goto out_err_release;
1139 
1140 		/* If we had an erroneous initial result, pretend it
1141 		 * never existed and let the SA-enabled version take
1142 		 * over.
1143 		 */
1144 		if ((*dst)->error) {
1145 			dst_release(*dst);
1146 			*dst = NULL;
1147 		}
1148 
1149 		if (fl6->flowi6_oif)
1150 			flags |= RT6_LOOKUP_F_IFACE;
1151 	}
1152 
1153 	if (!*dst)
1154 		*dst = ip6_route_output_flags(net, sk, fl6, flags);
1155 
1156 	err = (*dst)->error;
1157 	if (err)
1158 		goto out_err_release;
1159 
1160 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1161 	/*
1162 	 * Here if the dst entry we've looked up
1163 	 * has a neighbour entry that is in the INCOMPLETE
1164 	 * state and the src address from the flow is
1165 	 * marked as OPTIMISTIC, we release the found
1166 	 * dst entry and replace it instead with the
1167 	 * dst entry of the nexthop router
1168 	 */
1169 	rt = dst_rt6_info(*dst);
1170 	rcu_read_lock();
1171 	n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1172 				      rt6_nexthop(rt, &fl6->daddr));
1173 	err = n && !(READ_ONCE(n->nud_state) & NUD_VALID) ? -EINVAL : 0;
1174 	rcu_read_unlock();
1175 
1176 	if (err) {
1177 		struct inet6_ifaddr *ifp;
1178 		struct flowi6 fl_gw6;
1179 		int redirect;
1180 
1181 		ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1182 				      (*dst)->dev, 1);
1183 
1184 		redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1185 		if (ifp)
1186 			in6_ifa_put(ifp);
1187 
1188 		if (redirect) {
1189 			/*
1190 			 * We need to get the dst entry for the
1191 			 * default router instead
1192 			 */
1193 			dst_release(*dst);
1194 			memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1195 			memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1196 			*dst = ip6_route_output(net, sk, &fl_gw6);
1197 			err = (*dst)->error;
1198 			if (err)
1199 				goto out_err_release;
1200 		}
1201 	}
1202 #endif
1203 	if (ipv6_addr_v4mapped(&fl6->saddr) &&
1204 	    !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1205 		err = -EAFNOSUPPORT;
1206 		goto out_err_release;
1207 	}
1208 
1209 	return 0;
1210 
1211 out_err_release:
1212 	dst_release(*dst);
1213 	*dst = NULL;
1214 
1215 	if (err == -ENETUNREACH)
1216 		IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1217 	return err;
1218 }
1219 
1220 /**
1221  *	ip6_dst_lookup - perform route lookup on flow
1222  *	@net: Network namespace to perform lookup in
1223  *	@sk: socket which provides route info
1224  *	@dst: pointer to dst_entry * for result
1225  *	@fl6: flow to lookup
1226  *
1227  *	This function performs a route lookup on the given flow.
1228  *
1229  *	It returns zero on success, or a standard errno code on error.
1230  */
ip6_dst_lookup(struct net * net,struct sock * sk,struct dst_entry ** dst,struct flowi6 * fl6)1231 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1232 		   struct flowi6 *fl6)
1233 {
1234 	*dst = NULL;
1235 	return ip6_dst_lookup_tail(net, sk, dst, fl6);
1236 }
1237 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1238 
1239 /**
1240  *	ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1241  *	@net: Network namespace to perform lookup in
1242  *	@sk: socket which provides route info
1243  *	@fl6: flow to lookup
1244  *	@final_dst: final destination address for ipsec lookup
1245  *
1246  *	This function performs a route lookup on the given flow.
1247  *
1248  *	It returns a valid dst pointer on success, or a pointer encoded
1249  *	error code.
1250  */
ip6_dst_lookup_flow(struct net * net,const struct sock * sk,struct flowi6 * fl6,const struct in6_addr * final_dst)1251 struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1252 				      const struct in6_addr *final_dst)
1253 {
1254 	struct dst_entry *dst = NULL;
1255 	int err;
1256 
1257 	err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1258 	if (err)
1259 		return ERR_PTR(err);
1260 	if (final_dst)
1261 		fl6->daddr = *final_dst;
1262 
1263 	return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1264 }
1265 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1266 
1267 /**
1268  *	ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1269  *	@sk: socket which provides the dst cache and route info
1270  *	@fl6: flow to lookup
1271  *	@final_dst: final destination address for ipsec lookup
1272  *	@connected: whether @sk is connected or not
1273  *
1274  *	This function performs a route lookup on the given flow with the
1275  *	possibility of using the cached route in the socket if it is valid.
1276  *	It will take the socket dst lock when operating on the dst cache.
1277  *	As a result, this function can only be used in process context.
1278  *
1279  *	In addition, for a connected socket, cache the dst in the socket
1280  *	if the current cache is not valid.
1281  *
1282  *	It returns a valid dst pointer on success, or a pointer encoded
1283  *	error code.
1284  */
ip6_sk_dst_lookup_flow(struct sock * sk,struct flowi6 * fl6,const struct in6_addr * final_dst,bool connected)1285 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1286 					 const struct in6_addr *final_dst,
1287 					 bool connected)
1288 {
1289 	struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1290 
1291 	dst = ip6_sk_dst_check(sk, dst, fl6);
1292 	if (dst)
1293 		return dst;
1294 
1295 	dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1296 	if (connected && !IS_ERR(dst))
1297 		ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1298 
1299 	return dst;
1300 }
1301 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1302 
1303 /**
1304  *      ip6_dst_lookup_tunnel - perform route lookup on tunnel
1305  *      @skb: Packet for which lookup is done
1306  *      @dev: Tunnel device
1307  *      @net: Network namespace of tunnel device
1308  *      @sock: Socket which provides route info
1309  *      @saddr: Memory to store the src ip address
1310  *      @info: Tunnel information
1311  *      @protocol: IP protocol
1312  *      @use_cache: Flag to enable cache usage
1313  *      This function performs a route lookup on a tunnel
1314  *
1315  *      It returns a valid dst pointer and stores src address to be used in
1316  *      tunnel in param saddr on success, else a pointer encoded error code.
1317  */
1318 
ip6_dst_lookup_tunnel(struct sk_buff * skb,struct net_device * dev,struct net * net,struct socket * sock,struct in6_addr * saddr,const struct ip_tunnel_info * info,u8 protocol,bool use_cache)1319 struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1320 					struct net_device *dev,
1321 					struct net *net,
1322 					struct socket *sock,
1323 					struct in6_addr *saddr,
1324 					const struct ip_tunnel_info *info,
1325 					u8 protocol,
1326 					bool use_cache)
1327 {
1328 	struct dst_entry *dst = NULL;
1329 #ifdef CONFIG_DST_CACHE
1330 	struct dst_cache *dst_cache;
1331 #endif
1332 	struct flowi6 fl6;
1333 	__u8 prio;
1334 
1335 #ifdef CONFIG_DST_CACHE
1336 	dst_cache = (struct dst_cache *)&info->dst_cache;
1337 	if (use_cache) {
1338 		dst = dst_cache_get_ip6(dst_cache, saddr);
1339 		if (dst)
1340 			return dst;
1341 	}
1342 #endif
1343 	memset(&fl6, 0, sizeof(fl6));
1344 	fl6.flowi6_mark = skb->mark;
1345 	fl6.flowi6_proto = protocol;
1346 	fl6.daddr = info->key.u.ipv6.dst;
1347 	fl6.saddr = info->key.u.ipv6.src;
1348 	prio = info->key.tos;
1349 	fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1350 
1351 	dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1352 					      NULL);
1353 	if (IS_ERR(dst)) {
1354 		netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1355 		return ERR_PTR(-ENETUNREACH);
1356 	}
1357 	if (dst->dev == dev) { /* is this necessary? */
1358 		netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1359 		dst_release(dst);
1360 		return ERR_PTR(-ELOOP);
1361 	}
1362 #ifdef CONFIG_DST_CACHE
1363 	if (use_cache)
1364 		dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1365 #endif
1366 	*saddr = fl6.saddr;
1367 	return dst;
1368 }
1369 EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1370 
ip6_opt_dup(struct ipv6_opt_hdr * src,gfp_t gfp)1371 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1372 					       gfp_t gfp)
1373 {
1374 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1375 }
1376 
ip6_rthdr_dup(struct ipv6_rt_hdr * src,gfp_t gfp)1377 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1378 						gfp_t gfp)
1379 {
1380 	return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1381 }
1382 
ip6_append_data_mtu(unsigned int * mtu,int * maxfraglen,unsigned int fragheaderlen,struct sk_buff * skb,struct rt6_info * rt,unsigned int orig_mtu)1383 static void ip6_append_data_mtu(unsigned int *mtu,
1384 				int *maxfraglen,
1385 				unsigned int fragheaderlen,
1386 				struct sk_buff *skb,
1387 				struct rt6_info *rt,
1388 				unsigned int orig_mtu)
1389 {
1390 	if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1391 		if (!skb) {
1392 			/* first fragment, reserve header_len */
1393 			*mtu = orig_mtu - rt->dst.header_len;
1394 
1395 		} else {
1396 			/*
1397 			 * this fragment is not first, the headers
1398 			 * space is regarded as data space.
1399 			 */
1400 			*mtu = orig_mtu;
1401 		}
1402 		*maxfraglen = ((*mtu - fragheaderlen) & ~7)
1403 			      + fragheaderlen - sizeof(struct frag_hdr);
1404 	}
1405 }
1406 
ip6_setup_cork(struct sock * sk,struct inet_cork_full * cork,struct inet6_cork * v6_cork,struct ipcm6_cookie * ipc6,struct rt6_info * rt)1407 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1408 			  struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1409 			  struct rt6_info *rt)
1410 {
1411 	struct ipv6_pinfo *np = inet6_sk(sk);
1412 	unsigned int mtu;
1413 	struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1414 
1415 	/* callers pass dst together with a reference, set it first so
1416 	 * ip6_cork_release() can put it down even in case of an error.
1417 	 */
1418 	cork->base.dst = &rt->dst;
1419 
1420 	/*
1421 	 * setup for corking
1422 	 */
1423 	if (opt) {
1424 		if (WARN_ON(v6_cork->opt))
1425 			return -EINVAL;
1426 
1427 		nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1428 		if (unlikely(!nopt))
1429 			return -ENOBUFS;
1430 
1431 		nopt->tot_len = sizeof(*opt);
1432 		nopt->opt_flen = opt->opt_flen;
1433 		nopt->opt_nflen = opt->opt_nflen;
1434 
1435 		nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1436 		if (opt->dst0opt && !nopt->dst0opt)
1437 			return -ENOBUFS;
1438 
1439 		nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1440 		if (opt->dst1opt && !nopt->dst1opt)
1441 			return -ENOBUFS;
1442 
1443 		nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1444 		if (opt->hopopt && !nopt->hopopt)
1445 			return -ENOBUFS;
1446 
1447 		nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1448 		if (opt->srcrt && !nopt->srcrt)
1449 			return -ENOBUFS;
1450 
1451 		/* need source address above miyazawa*/
1452 	}
1453 	v6_cork->hop_limit = ipc6->hlimit;
1454 	v6_cork->tclass = ipc6->tclass;
1455 	if (rt->dst.flags & DST_XFRM_TUNNEL)
1456 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1457 		      READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1458 	else
1459 		mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1460 			READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1461 	if (np->frag_size < mtu) {
1462 		if (np->frag_size)
1463 			mtu = np->frag_size;
1464 	}
1465 	cork->base.fragsize = mtu;
1466 	cork->base.gso_size = ipc6->gso_size;
1467 	cork->base.tx_flags = 0;
1468 	cork->base.mark = ipc6->sockc.mark;
1469 	sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1470 
1471 	if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1472 		cork->base.flags |= IPCORK_ALLFRAG;
1473 	cork->base.length = 0;
1474 
1475 	cork->base.transmit_time = ipc6->sockc.transmit_time;
1476 
1477 	return 0;
1478 }
1479 
__ip6_append_data(struct sock * sk,struct sk_buff_head * queue,struct inet_cork_full * cork_full,struct inet6_cork * v6_cork,struct page_frag * pfrag,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,size_t length,int transhdrlen,unsigned int flags,struct ipcm6_cookie * ipc6)1480 static int __ip6_append_data(struct sock *sk,
1481 			     struct sk_buff_head *queue,
1482 			     struct inet_cork_full *cork_full,
1483 			     struct inet6_cork *v6_cork,
1484 			     struct page_frag *pfrag,
1485 			     int getfrag(void *from, char *to, int offset,
1486 					 int len, int odd, struct sk_buff *skb),
1487 			     void *from, size_t length, int transhdrlen,
1488 			     unsigned int flags, struct ipcm6_cookie *ipc6)
1489 {
1490 	struct sk_buff *skb, *skb_prev = NULL;
1491 	struct inet_cork *cork = &cork_full->base;
1492 	struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1493 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1494 	struct ubuf_info *uarg = NULL;
1495 	int exthdrlen = 0;
1496 	int dst_exthdrlen = 0;
1497 	int hh_len;
1498 	int copy;
1499 	int err;
1500 	int offset = 0;
1501 	bool zc = false;
1502 	u32 tskey = 0;
1503 	struct rt6_info *rt = dst_rt6_info(cork->dst);
1504 	bool paged, hold_tskey, extra_uref = false;
1505 	struct ipv6_txoptions *opt = v6_cork->opt;
1506 	int csummode = CHECKSUM_NONE;
1507 	unsigned int maxnonfragsize, headersize;
1508 	unsigned int wmem_alloc_delta = 0;
1509 
1510 	skb = skb_peek_tail(queue);
1511 	if (!skb) {
1512 		exthdrlen = opt ? opt->opt_flen : 0;
1513 		dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1514 	}
1515 
1516 	paged = !!cork->gso_size;
1517 	mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1518 	orig_mtu = mtu;
1519 
1520 	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1521 
1522 	fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1523 			(opt ? opt->opt_nflen : 0);
1524 
1525 	headersize = sizeof(struct ipv6hdr) +
1526 		     (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1527 		     (dst_allfrag(&rt->dst) ?
1528 		      sizeof(struct frag_hdr) : 0) +
1529 		     rt->rt6i_nfheader_len;
1530 
1531 	if (mtu <= fragheaderlen ||
1532 	    ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1533 		goto emsgsize;
1534 
1535 	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1536 		     sizeof(struct frag_hdr);
1537 
1538 	/* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1539 	 * the first fragment
1540 	 */
1541 	if (headersize + transhdrlen > mtu)
1542 		goto emsgsize;
1543 
1544 	if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1545 	    (sk->sk_protocol == IPPROTO_UDP ||
1546 	     sk->sk_protocol == IPPROTO_ICMPV6 ||
1547 	     sk->sk_protocol == IPPROTO_RAW)) {
1548 		ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1549 				sizeof(struct ipv6hdr));
1550 		goto emsgsize;
1551 	}
1552 
1553 	if (ip6_sk_ignore_df(sk))
1554 		maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1555 	else
1556 		maxnonfragsize = mtu;
1557 
1558 	if (cork->length + length > maxnonfragsize - headersize) {
1559 emsgsize:
1560 		pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1561 		ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1562 		return -EMSGSIZE;
1563 	}
1564 
1565 	/* CHECKSUM_PARTIAL only with no extension headers and when
1566 	 * we are not going to fragment
1567 	 */
1568 	if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1569 	    headersize == sizeof(struct ipv6hdr) &&
1570 	    length <= mtu - headersize &&
1571 	    (!(flags & MSG_MORE) || cork->gso_size) &&
1572 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1573 		csummode = CHECKSUM_PARTIAL;
1574 
1575 	if ((flags & MSG_ZEROCOPY) && length) {
1576 		struct msghdr *msg = from;
1577 
1578 		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1579 			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1580 				return -EINVAL;
1581 
1582 			/* Leave uarg NULL if can't zerocopy, callers should
1583 			 * be able to handle it.
1584 			 */
1585 			if ((rt->dst.dev->features & NETIF_F_SG) &&
1586 			    csummode == CHECKSUM_PARTIAL) {
1587 				paged = true;
1588 				zc = true;
1589 				uarg = msg->msg_ubuf;
1590 			}
1591 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1592 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1593 			if (!uarg)
1594 				return -ENOBUFS;
1595 			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1596 			if (rt->dst.dev->features & NETIF_F_SG &&
1597 			    csummode == CHECKSUM_PARTIAL) {
1598 				paged = true;
1599 				zc = true;
1600 			} else {
1601 				uarg_to_msgzc(uarg)->zerocopy = 0;
1602 				skb_zcopy_set(skb, uarg, &extra_uref);
1603 			}
1604 		}
1605 	} else if ((flags & MSG_SPLICE_PAGES) && length) {
1606 		if (inet_test_bit(HDRINCL, sk))
1607 			return -EPERM;
1608 		if (rt->dst.dev->features & NETIF_F_SG &&
1609 		    getfrag == ip_generic_getfrag)
1610 			/* We need an empty buffer to attach stuff to */
1611 			paged = true;
1612 		else
1613 			flags &= ~MSG_SPLICE_PAGES;
1614 	}
1615 
1616 	hold_tskey = cork->tx_flags & SKBTX_ANY_TSTAMP &&
1617 		     READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID;
1618 	if (hold_tskey)
1619 		tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1620 
1621 	/*
1622 	 * Let's try using as much space as possible.
1623 	 * Use MTU if total length of the message fits into the MTU.
1624 	 * Otherwise, we need to reserve fragment header and
1625 	 * fragment alignment (= 8-15 octects, in total).
1626 	 *
1627 	 * Note that we may need to "move" the data from the tail
1628 	 * of the buffer to the new fragment when we split
1629 	 * the message.
1630 	 *
1631 	 * FIXME: It may be fragmented into multiple chunks
1632 	 *        at once if non-fragmentable extension headers
1633 	 *        are too large.
1634 	 * --yoshfuji
1635 	 */
1636 
1637 	cork->length += length;
1638 	if (!skb)
1639 		goto alloc_new_skb;
1640 
1641 	while (length > 0) {
1642 		/* Check if the remaining data fits into current packet. */
1643 		copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1644 		if (copy < length)
1645 			copy = maxfraglen - skb->len;
1646 
1647 		if (copy <= 0) {
1648 			char *data;
1649 			unsigned int datalen;
1650 			unsigned int fraglen;
1651 			unsigned int fraggap;
1652 			unsigned int alloclen, alloc_extra;
1653 			unsigned int pagedlen;
1654 alloc_new_skb:
1655 			/* There's no room in the current skb */
1656 			if (skb)
1657 				fraggap = skb->len - maxfraglen;
1658 			else
1659 				fraggap = 0;
1660 			/* update mtu and maxfraglen if necessary */
1661 			if (!skb || !skb_prev)
1662 				ip6_append_data_mtu(&mtu, &maxfraglen,
1663 						    fragheaderlen, skb, rt,
1664 						    orig_mtu);
1665 
1666 			skb_prev = skb;
1667 
1668 			/*
1669 			 * If remaining data exceeds the mtu,
1670 			 * we know we need more fragment(s).
1671 			 */
1672 			datalen = length + fraggap;
1673 
1674 			if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1675 				datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1676 			fraglen = datalen + fragheaderlen;
1677 			pagedlen = 0;
1678 
1679 			alloc_extra = hh_len;
1680 			alloc_extra += dst_exthdrlen;
1681 			alloc_extra += rt->dst.trailer_len;
1682 
1683 			/* We just reserve space for fragment header.
1684 			 * Note: this may be overallocation if the message
1685 			 * (without MSG_MORE) fits into the MTU.
1686 			 */
1687 			alloc_extra += sizeof(struct frag_hdr);
1688 
1689 			if ((flags & MSG_MORE) &&
1690 			    !(rt->dst.dev->features&NETIF_F_SG))
1691 				alloclen = mtu;
1692 			else if (!paged &&
1693 				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1694 				  !(rt->dst.dev->features & NETIF_F_SG)))
1695 				alloclen = fraglen;
1696 			else {
1697 				alloclen = fragheaderlen + transhdrlen;
1698 				pagedlen = datalen - transhdrlen;
1699 			}
1700 			alloclen += alloc_extra;
1701 
1702 			if (datalen != length + fraggap) {
1703 				/*
1704 				 * this is not the last fragment, the trailer
1705 				 * space is regarded as data space.
1706 				 */
1707 				datalen += rt->dst.trailer_len;
1708 			}
1709 
1710 			fraglen = datalen + fragheaderlen;
1711 
1712 			copy = datalen - transhdrlen - fraggap - pagedlen;
1713 			/* [!] NOTE: copy may be negative if pagedlen>0
1714 			 * because then the equation may reduces to -fraggap.
1715 			 */
1716 			if (copy < 0 && !(flags & MSG_SPLICE_PAGES)) {
1717 				err = -EINVAL;
1718 				goto error;
1719 			}
1720 			if (transhdrlen) {
1721 				skb = sock_alloc_send_skb(sk, alloclen,
1722 						(flags & MSG_DONTWAIT), &err);
1723 			} else {
1724 				skb = NULL;
1725 				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1726 				    2 * sk->sk_sndbuf)
1727 					skb = alloc_skb(alloclen,
1728 							sk->sk_allocation);
1729 				if (unlikely(!skb))
1730 					err = -ENOBUFS;
1731 			}
1732 			if (!skb)
1733 				goto error;
1734 			/*
1735 			 *	Fill in the control structures
1736 			 */
1737 			skb->protocol = htons(ETH_P_IPV6);
1738 			skb->ip_summed = csummode;
1739 			skb->csum = 0;
1740 			/* reserve for fragmentation and ipsec header */
1741 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1742 				    dst_exthdrlen);
1743 
1744 			/*
1745 			 *	Find where to start putting bytes
1746 			 */
1747 			data = skb_put(skb, fraglen - pagedlen);
1748 			skb_set_network_header(skb, exthdrlen);
1749 			data += fragheaderlen;
1750 			skb->transport_header = (skb->network_header +
1751 						 fragheaderlen);
1752 			if (fraggap) {
1753 				skb->csum = skb_copy_and_csum_bits(
1754 					skb_prev, maxfraglen,
1755 					data + transhdrlen, fraggap);
1756 				skb_prev->csum = csum_sub(skb_prev->csum,
1757 							  skb->csum);
1758 				data += fraggap;
1759 				pskb_trim_unique(skb_prev, maxfraglen);
1760 			}
1761 			if (copy > 0 &&
1762 			    getfrag(from, data + transhdrlen, offset,
1763 				    copy, fraggap, skb) < 0) {
1764 				err = -EFAULT;
1765 				kfree_skb(skb);
1766 				goto error;
1767 			} else if (flags & MSG_SPLICE_PAGES) {
1768 				copy = 0;
1769 			}
1770 
1771 			offset += copy;
1772 			length -= copy + transhdrlen;
1773 			transhdrlen = 0;
1774 			exthdrlen = 0;
1775 			dst_exthdrlen = 0;
1776 
1777 			/* Only the initial fragment is time stamped */
1778 			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1779 			cork->tx_flags = 0;
1780 			skb_shinfo(skb)->tskey = tskey;
1781 			tskey = 0;
1782 			skb_zcopy_set(skb, uarg, &extra_uref);
1783 
1784 			if ((flags & MSG_CONFIRM) && !skb_prev)
1785 				skb_set_dst_pending_confirm(skb, 1);
1786 
1787 			/*
1788 			 * Put the packet on the pending queue
1789 			 */
1790 			if (!skb->destructor) {
1791 				skb->destructor = sock_wfree;
1792 				skb->sk = sk;
1793 				wmem_alloc_delta += skb->truesize;
1794 			}
1795 			__skb_queue_tail(queue, skb);
1796 			continue;
1797 		}
1798 
1799 		if (copy > length)
1800 			copy = length;
1801 
1802 		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1803 		    skb_tailroom(skb) >= copy) {
1804 			unsigned int off;
1805 
1806 			off = skb->len;
1807 			if (getfrag(from, skb_put(skb, copy),
1808 						offset, copy, off, skb) < 0) {
1809 				__skb_trim(skb, off);
1810 				err = -EFAULT;
1811 				goto error;
1812 			}
1813 		} else if (flags & MSG_SPLICE_PAGES) {
1814 			struct msghdr *msg = from;
1815 
1816 			err = -EIO;
1817 			if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1818 				goto error;
1819 
1820 			err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1821 						   sk->sk_allocation);
1822 			if (err < 0)
1823 				goto error;
1824 			copy = err;
1825 			wmem_alloc_delta += copy;
1826 		} else if (!zc) {
1827 			int i = skb_shinfo(skb)->nr_frags;
1828 
1829 			err = -ENOMEM;
1830 			if (!sk_page_frag_refill(sk, pfrag))
1831 				goto error;
1832 
1833 			skb_zcopy_downgrade_managed(skb);
1834 			if (!skb_can_coalesce(skb, i, pfrag->page,
1835 					      pfrag->offset)) {
1836 				err = -EMSGSIZE;
1837 				if (i == MAX_SKB_FRAGS)
1838 					goto error;
1839 
1840 				__skb_fill_page_desc(skb, i, pfrag->page,
1841 						     pfrag->offset, 0);
1842 				skb_shinfo(skb)->nr_frags = ++i;
1843 				get_page(pfrag->page);
1844 			}
1845 			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1846 			if (getfrag(from,
1847 				    page_address(pfrag->page) + pfrag->offset,
1848 				    offset, copy, skb->len, skb) < 0)
1849 				goto error_efault;
1850 
1851 			pfrag->offset += copy;
1852 			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1853 			skb->len += copy;
1854 			skb->data_len += copy;
1855 			skb->truesize += copy;
1856 			wmem_alloc_delta += copy;
1857 		} else {
1858 			err = skb_zerocopy_iter_dgram(skb, from, copy);
1859 			if (err < 0)
1860 				goto error;
1861 		}
1862 		offset += copy;
1863 		length -= copy;
1864 	}
1865 
1866 	if (wmem_alloc_delta)
1867 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1868 	return 0;
1869 
1870 error_efault:
1871 	err = -EFAULT;
1872 error:
1873 	net_zcopy_put_abort(uarg, extra_uref);
1874 	cork->length -= length;
1875 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1876 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1877 	if (hold_tskey)
1878 		atomic_dec(&sk->sk_tskey);
1879 	return err;
1880 }
1881 
ip6_append_data(struct sock * sk,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,size_t length,int transhdrlen,struct ipcm6_cookie * ipc6,struct flowi6 * fl6,struct rt6_info * rt,unsigned int flags)1882 int ip6_append_data(struct sock *sk,
1883 		    int getfrag(void *from, char *to, int offset, int len,
1884 				int odd, struct sk_buff *skb),
1885 		    void *from, size_t length, int transhdrlen,
1886 		    struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1887 		    struct rt6_info *rt, unsigned int flags)
1888 {
1889 	struct inet_sock *inet = inet_sk(sk);
1890 	struct ipv6_pinfo *np = inet6_sk(sk);
1891 	int exthdrlen;
1892 	int err;
1893 
1894 	if (flags&MSG_PROBE)
1895 		return 0;
1896 	if (skb_queue_empty(&sk->sk_write_queue)) {
1897 		/*
1898 		 * setup for corking
1899 		 */
1900 		dst_hold(&rt->dst);
1901 		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1902 				     ipc6, rt);
1903 		if (err)
1904 			return err;
1905 
1906 		inet->cork.fl.u.ip6 = *fl6;
1907 		exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1908 		length += exthdrlen;
1909 		transhdrlen += exthdrlen;
1910 	} else {
1911 		transhdrlen = 0;
1912 	}
1913 
1914 	return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1915 				 &np->cork, sk_page_frag(sk), getfrag,
1916 				 from, length, transhdrlen, flags, ipc6);
1917 }
1918 EXPORT_SYMBOL_GPL(ip6_append_data);
1919 
ip6_cork_steal_dst(struct sk_buff * skb,struct inet_cork_full * cork)1920 static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1921 {
1922 	struct dst_entry *dst = cork->base.dst;
1923 
1924 	cork->base.dst = NULL;
1925 	cork->base.flags &= ~IPCORK_ALLFRAG;
1926 	skb_dst_set(skb, dst);
1927 }
1928 
ip6_cork_release(struct inet_cork_full * cork,struct inet6_cork * v6_cork)1929 static void ip6_cork_release(struct inet_cork_full *cork,
1930 			     struct inet6_cork *v6_cork)
1931 {
1932 	if (v6_cork->opt) {
1933 		struct ipv6_txoptions *opt = v6_cork->opt;
1934 
1935 		kfree(opt->dst0opt);
1936 		kfree(opt->dst1opt);
1937 		kfree(opt->hopopt);
1938 		kfree(opt->srcrt);
1939 		kfree(opt);
1940 		v6_cork->opt = NULL;
1941 	}
1942 
1943 	if (cork->base.dst) {
1944 		dst_release(cork->base.dst);
1945 		cork->base.dst = NULL;
1946 		cork->base.flags &= ~IPCORK_ALLFRAG;
1947 	}
1948 }
1949 
__ip6_make_skb(struct sock * sk,struct sk_buff_head * queue,struct inet_cork_full * cork,struct inet6_cork * v6_cork)1950 struct sk_buff *__ip6_make_skb(struct sock *sk,
1951 			       struct sk_buff_head *queue,
1952 			       struct inet_cork_full *cork,
1953 			       struct inet6_cork *v6_cork)
1954 {
1955 	struct sk_buff *skb, *tmp_skb;
1956 	struct sk_buff **tail_skb;
1957 	struct in6_addr *final_dst;
1958 	struct ipv6_pinfo *np = inet6_sk(sk);
1959 	struct net *net = sock_net(sk);
1960 	struct ipv6hdr *hdr;
1961 	struct ipv6_txoptions *opt = v6_cork->opt;
1962 	struct rt6_info *rt = dst_rt6_info(cork->base.dst);
1963 	struct flowi6 *fl6 = &cork->fl.u.ip6;
1964 	unsigned char proto = fl6->flowi6_proto;
1965 
1966 	skb = __skb_dequeue(queue);
1967 	if (!skb)
1968 		goto out;
1969 	tail_skb = &(skb_shinfo(skb)->frag_list);
1970 
1971 	/* move skb->data to ip header from ext header */
1972 	if (skb->data < skb_network_header(skb))
1973 		__skb_pull(skb, skb_network_offset(skb));
1974 	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1975 		__skb_pull(tmp_skb, skb_network_header_len(skb));
1976 		*tail_skb = tmp_skb;
1977 		tail_skb = &(tmp_skb->next);
1978 		skb->len += tmp_skb->len;
1979 		skb->data_len += tmp_skb->len;
1980 		skb->truesize += tmp_skb->truesize;
1981 		tmp_skb->destructor = NULL;
1982 		tmp_skb->sk = NULL;
1983 	}
1984 
1985 	/* Allow local fragmentation. */
1986 	skb->ignore_df = ip6_sk_ignore_df(sk);
1987 	__skb_pull(skb, skb_network_header_len(skb));
1988 
1989 	final_dst = &fl6->daddr;
1990 	if (opt && opt->opt_flen)
1991 		ipv6_push_frag_opts(skb, opt, &proto);
1992 	if (opt && opt->opt_nflen)
1993 		ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1994 
1995 	skb_push(skb, sizeof(struct ipv6hdr));
1996 	skb_reset_network_header(skb);
1997 	hdr = ipv6_hdr(skb);
1998 
1999 	ip6_flow_hdr(hdr, v6_cork->tclass,
2000 		     ip6_make_flowlabel(net, skb, fl6->flowlabel,
2001 					ip6_autoflowlabel(net, np), fl6));
2002 	hdr->hop_limit = v6_cork->hop_limit;
2003 	hdr->nexthdr = proto;
2004 	hdr->saddr = fl6->saddr;
2005 	hdr->daddr = *final_dst;
2006 
2007 	skb->priority = sk->sk_priority;
2008 	skb->mark = cork->base.mark;
2009 	skb->tstamp = cork->base.transmit_time;
2010 
2011 	ip6_cork_steal_dst(skb, cork);
2012 	IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
2013 	if (proto == IPPROTO_ICMPV6) {
2014 		struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
2015 		u8 icmp6_type;
2016 
2017 		if (sk->sk_socket->type == SOCK_RAW &&
2018 		   !(fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH))
2019 			icmp6_type = fl6->fl6_icmp_type;
2020 		else
2021 			icmp6_type = icmp6_hdr(skb)->icmp6_type;
2022 		ICMP6MSGOUT_INC_STATS(net, idev, icmp6_type);
2023 		ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
2024 	}
2025 
2026 	ip6_cork_release(cork, v6_cork);
2027 out:
2028 	return skb;
2029 }
2030 
ip6_send_skb(struct sk_buff * skb)2031 int ip6_send_skb(struct sk_buff *skb)
2032 {
2033 	struct net *net = sock_net(skb->sk);
2034 	struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
2035 	int err;
2036 
2037 	rcu_read_lock();
2038 	err = ip6_local_out(net, skb->sk, skb);
2039 	if (err) {
2040 		if (err > 0)
2041 			err = net_xmit_errno(err);
2042 		if (err)
2043 			IP6_INC_STATS(net, rt->rt6i_idev,
2044 				      IPSTATS_MIB_OUTDISCARDS);
2045 	}
2046 
2047 	rcu_read_unlock();
2048 	return err;
2049 }
2050 
ip6_push_pending_frames(struct sock * sk)2051 int ip6_push_pending_frames(struct sock *sk)
2052 {
2053 	struct sk_buff *skb;
2054 
2055 	skb = ip6_finish_skb(sk);
2056 	if (!skb)
2057 		return 0;
2058 
2059 	return ip6_send_skb(skb);
2060 }
2061 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2062 
__ip6_flush_pending_frames(struct sock * sk,struct sk_buff_head * queue,struct inet_cork_full * cork,struct inet6_cork * v6_cork)2063 static void __ip6_flush_pending_frames(struct sock *sk,
2064 				       struct sk_buff_head *queue,
2065 				       struct inet_cork_full *cork,
2066 				       struct inet6_cork *v6_cork)
2067 {
2068 	struct sk_buff *skb;
2069 
2070 	while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2071 		if (skb_dst(skb))
2072 			IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2073 				      IPSTATS_MIB_OUTDISCARDS);
2074 		kfree_skb(skb);
2075 	}
2076 
2077 	ip6_cork_release(cork, v6_cork);
2078 }
2079 
ip6_flush_pending_frames(struct sock * sk)2080 void ip6_flush_pending_frames(struct sock *sk)
2081 {
2082 	__ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2083 				   &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2084 }
2085 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2086 
ip6_make_skb(struct sock * sk,int getfrag (void * from,char * to,int offset,int len,int odd,struct sk_buff * skb),void * from,size_t length,int transhdrlen,struct ipcm6_cookie * ipc6,struct rt6_info * rt,unsigned int flags,struct inet_cork_full * cork)2087 struct sk_buff *ip6_make_skb(struct sock *sk,
2088 			     int getfrag(void *from, char *to, int offset,
2089 					 int len, int odd, struct sk_buff *skb),
2090 			     void *from, size_t length, int transhdrlen,
2091 			     struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2092 			     unsigned int flags, struct inet_cork_full *cork)
2093 {
2094 	struct inet6_cork v6_cork;
2095 	struct sk_buff_head queue;
2096 	int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2097 	int err;
2098 
2099 	if (flags & MSG_PROBE) {
2100 		dst_release(&rt->dst);
2101 		return NULL;
2102 	}
2103 
2104 	__skb_queue_head_init(&queue);
2105 
2106 	cork->base.flags = 0;
2107 	cork->base.addr = 0;
2108 	cork->base.opt = NULL;
2109 	v6_cork.opt = NULL;
2110 	err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2111 	if (err) {
2112 		ip6_cork_release(cork, &v6_cork);
2113 		return ERR_PTR(err);
2114 	}
2115 	if (ipc6->dontfrag < 0)
2116 		ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2117 
2118 	err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2119 				&current->task_frag, getfrag, from,
2120 				length + exthdrlen, transhdrlen + exthdrlen,
2121 				flags, ipc6);
2122 	if (err) {
2123 		__ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2124 		return ERR_PTR(err);
2125 	}
2126 
2127 	return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2128 }
2129