Loading...
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
60static 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 (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
146static int
147ip6_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
185static int ip6_finish_output_gso(struct net *net, struct sock *sk,
186 struct sk_buff *skb, unsigned int mtu)
187{
188 if (!(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
189 !skb_gso_validate_network_len(skb, mtu))
190 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
191
192 return ip6_finish_output2(net, sk, skb);
193}
194
195static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
196{
197 unsigned int mtu;
198
199#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
200 /* Policy lookup after SNAT yielded a new policy */
201 if (skb_dst(skb)->xfrm) {
202 IP6CB(skb)->flags |= IP6SKB_REROUTED;
203 return dst_output(net, sk, skb);
204 }
205#endif
206
207 mtu = ip6_skb_dst_mtu(skb);
208 if (skb_is_gso(skb))
209 return ip6_finish_output_gso(net, sk, skb, mtu);
210
211 if (skb->len > mtu ||
212 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
213 return ip6_fragment(net, sk, skb, ip6_finish_output2);
214
215 return ip6_finish_output2(net, sk, skb);
216}
217
218static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
219{
220 int ret;
221
222 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
223 switch (ret) {
224 case NET_XMIT_SUCCESS:
225 case NET_XMIT_CN:
226 return __ip6_finish_output(net, sk, skb) ? : ret;
227 default:
228 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
229 return ret;
230 }
231}
232
233int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
234{
235 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
236 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
237
238 skb->protocol = htons(ETH_P_IPV6);
239 skb->dev = dev;
240
241 if (unlikely(!idev || READ_ONCE(idev->cnf.disable_ipv6))) {
242 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
243 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
244 return 0;
245 }
246
247 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
248 net, sk, skb, indev, dev,
249 ip6_finish_output,
250 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
251}
252EXPORT_SYMBOL(ip6_output);
253
254bool ip6_autoflowlabel(struct net *net, const struct sock *sk)
255{
256 if (!inet6_test_bit(AUTOFLOWLABEL_SET, sk))
257 return ip6_default_np_autolabel(net);
258 return inet6_test_bit(AUTOFLOWLABEL, sk);
259}
260
261/*
262 * xmit an sk_buff (used by TCP, SCTP and DCCP)
263 * Note : socket lock is not held for SYNACK packets, but might be modified
264 * by calls to skb_set_owner_w() and ipv6_local_error(),
265 * which are using proper atomic operations or spinlocks.
266 */
267int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
268 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
269{
270 struct net *net = sock_net(sk);
271 const struct ipv6_pinfo *np = inet6_sk(sk);
272 struct in6_addr *first_hop = &fl6->daddr;
273 struct dst_entry *dst = skb_dst(skb);
274 struct net_device *dev = dst->dev;
275 struct inet6_dev *idev = ip6_dst_idev(dst);
276 struct hop_jumbo_hdr *hop_jumbo;
277 int hoplen = sizeof(*hop_jumbo);
278 unsigned int head_room;
279 struct ipv6hdr *hdr;
280 u8 proto = fl6->flowi6_proto;
281 int seg_len = skb->len;
282 int hlimit = -1;
283 u32 mtu;
284
285 head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
286 if (opt)
287 head_room += opt->opt_nflen + opt->opt_flen;
288
289 if (unlikely(head_room > skb_headroom(skb))) {
290 /* Make sure idev stays alive */
291 rcu_read_lock();
292 skb = skb_expand_head(skb, head_room);
293 if (!skb) {
294 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
295 rcu_read_unlock();
296 return -ENOBUFS;
297 }
298 rcu_read_unlock();
299 }
300
301 if (opt) {
302 seg_len += opt->opt_nflen + opt->opt_flen;
303
304 if (opt->opt_flen)
305 ipv6_push_frag_opts(skb, opt, &proto);
306
307 if (opt->opt_nflen)
308 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
309 &fl6->saddr);
310 }
311
312 if (unlikely(seg_len > IPV6_MAXPLEN)) {
313 hop_jumbo = skb_push(skb, hoplen);
314
315 hop_jumbo->nexthdr = proto;
316 hop_jumbo->hdrlen = 0;
317 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
318 hop_jumbo->tlv_len = 4;
319 hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
320
321 proto = IPPROTO_HOPOPTS;
322 seg_len = 0;
323 IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
324 }
325
326 skb_push(skb, sizeof(struct ipv6hdr));
327 skb_reset_network_header(skb);
328 hdr = ipv6_hdr(skb);
329
330 /*
331 * Fill in the IPv6 header
332 */
333 if (np)
334 hlimit = READ_ONCE(np->hop_limit);
335 if (hlimit < 0)
336 hlimit = ip6_dst_hoplimit(dst);
337
338 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
339 ip6_autoflowlabel(net, sk), fl6));
340
341 hdr->payload_len = htons(seg_len);
342 hdr->nexthdr = proto;
343 hdr->hop_limit = hlimit;
344
345 hdr->saddr = fl6->saddr;
346 hdr->daddr = *first_hop;
347
348 skb->protocol = htons(ETH_P_IPV6);
349 skb->priority = priority;
350 skb->mark = mark;
351
352 mtu = dst_mtu(dst);
353 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
354 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTREQUESTS);
355
356 /* if egress device is enslaved to an L3 master device pass the
357 * skb to its handler for processing
358 */
359 skb = l3mdev_ip6_out((struct sock *)sk, skb);
360 if (unlikely(!skb))
361 return 0;
362
363 /* hooks should never assume socket lock is held.
364 * we promote our socket to non const
365 */
366 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
367 net, (struct sock *)sk, skb, NULL, dev,
368 dst_output);
369 }
370
371 skb->dev = dev;
372 /* ipv6_local_error() does not require socket lock,
373 * we promote our socket to non const
374 */
375 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
376
377 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
378 kfree_skb(skb);
379 return -EMSGSIZE;
380}
381EXPORT_SYMBOL(ip6_xmit);
382
383static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
384{
385 struct ip6_ra_chain *ra;
386 struct sock *last = NULL;
387
388 read_lock(&ip6_ra_lock);
389 for (ra = ip6_ra_chain; ra; ra = ra->next) {
390 struct sock *sk = ra->sk;
391 if (sk && ra->sel == sel &&
392 (!sk->sk_bound_dev_if ||
393 sk->sk_bound_dev_if == skb->dev->ifindex)) {
394
395 if (inet6_test_bit(RTALERT_ISOLATE, sk) &&
396 !net_eq(sock_net(sk), dev_net(skb->dev))) {
397 continue;
398 }
399 if (last) {
400 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
401 if (skb2)
402 rawv6_rcv(last, skb2);
403 }
404 last = sk;
405 }
406 }
407
408 if (last) {
409 rawv6_rcv(last, skb);
410 read_unlock(&ip6_ra_lock);
411 return 1;
412 }
413 read_unlock(&ip6_ra_lock);
414 return 0;
415}
416
417static int ip6_forward_proxy_check(struct sk_buff *skb)
418{
419 struct ipv6hdr *hdr = ipv6_hdr(skb);
420 u8 nexthdr = hdr->nexthdr;
421 __be16 frag_off;
422 int offset;
423
424 if (ipv6_ext_hdr(nexthdr)) {
425 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
426 if (offset < 0)
427 return 0;
428 } else
429 offset = sizeof(struct ipv6hdr);
430
431 if (nexthdr == IPPROTO_ICMPV6) {
432 struct icmp6hdr *icmp6;
433
434 if (!pskb_may_pull(skb, (skb_network_header(skb) +
435 offset + 1 - skb->data)))
436 return 0;
437
438 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
439
440 switch (icmp6->icmp6_type) {
441 case NDISC_ROUTER_SOLICITATION:
442 case NDISC_ROUTER_ADVERTISEMENT:
443 case NDISC_NEIGHBOUR_SOLICITATION:
444 case NDISC_NEIGHBOUR_ADVERTISEMENT:
445 case NDISC_REDIRECT:
446 /* For reaction involving unicast neighbor discovery
447 * message destined to the proxied address, pass it to
448 * input function.
449 */
450 return 1;
451 default:
452 break;
453 }
454 }
455
456 /*
457 * The proxying router can't forward traffic sent to a link-local
458 * address, so signal the sender and discard the packet. This
459 * behavior is clarified by the MIPv6 specification.
460 */
461 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
462 dst_link_failure(skb);
463 return -1;
464 }
465
466 return 0;
467}
468
469static inline int ip6_forward_finish(struct net *net, struct sock *sk,
470 struct sk_buff *skb)
471{
472#ifdef CONFIG_NET_SWITCHDEV
473 if (skb->offload_l3_fwd_mark) {
474 consume_skb(skb);
475 return 0;
476 }
477#endif
478
479 skb_clear_tstamp(skb);
480 return dst_output(net, sk, skb);
481}
482
483static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
484{
485 if (skb->len <= mtu)
486 return false;
487
488 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
489 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
490 return true;
491
492 if (skb->ignore_df)
493 return false;
494
495 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
496 return false;
497
498 return true;
499}
500
501int ip6_forward(struct sk_buff *skb)
502{
503 struct dst_entry *dst = skb_dst(skb);
504 struct ipv6hdr *hdr = ipv6_hdr(skb);
505 struct inet6_skb_parm *opt = IP6CB(skb);
506 struct net *net = dev_net(dst->dev);
507 struct inet6_dev *idev;
508 SKB_DR(reason);
509 u32 mtu;
510
511 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
512 if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
513 goto error;
514
515 if (skb->pkt_type != PACKET_HOST)
516 goto drop;
517
518 if (unlikely(skb->sk))
519 goto drop;
520
521 if (skb_warn_if_lro(skb))
522 goto drop;
523
524 if (!READ_ONCE(net->ipv6.devconf_all->disable_policy) &&
525 (!idev || !READ_ONCE(idev->cnf.disable_policy)) &&
526 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
527 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
528 goto drop;
529 }
530
531 skb_forward_csum(skb);
532
533 /*
534 * We DO NOT make any processing on
535 * RA packets, pushing them to user level AS IS
536 * without ane WARRANTY that application will be able
537 * to interpret them. The reason is that we
538 * cannot make anything clever here.
539 *
540 * We are not end-node, so that if packet contains
541 * AH/ESP, we cannot make anything.
542 * Defragmentation also would be mistake, RA packets
543 * cannot be fragmented, because there is no warranty
544 * that different fragments will go along one path. --ANK
545 */
546 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
547 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
548 return 0;
549 }
550
551 /*
552 * check and decrement ttl
553 */
554 if (hdr->hop_limit <= 1) {
555 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
556 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
557
558 kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
559 return -ETIMEDOUT;
560 }
561
562 /* XXX: idev->cnf.proxy_ndp? */
563 if (READ_ONCE(net->ipv6.devconf_all->proxy_ndp) &&
564 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
565 int proxied = ip6_forward_proxy_check(skb);
566 if (proxied > 0) {
567 /* It's tempting to decrease the hop limit
568 * here by 1, as we do at the end of the
569 * function too.
570 *
571 * But that would be incorrect, as proxying is
572 * not forwarding. The ip6_input function
573 * will handle this packet locally, and it
574 * depends on the hop limit being unchanged.
575 *
576 * One example is the NDP hop limit, that
577 * always has to stay 255, but other would be
578 * similar checks around RA packets, where the
579 * user can even change the desired limit.
580 */
581 return ip6_input(skb);
582 } else if (proxied < 0) {
583 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
584 goto drop;
585 }
586 }
587
588 if (!xfrm6_route_forward(skb)) {
589 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
590 SKB_DR_SET(reason, XFRM_POLICY);
591 goto drop;
592 }
593 dst = skb_dst(skb);
594
595 /* IPv6 specs say nothing about it, but it is clear that we cannot
596 send redirects to source routed frames.
597 We don't send redirects to frames decapsulated from IPsec.
598 */
599 if (IP6CB(skb)->iif == dst->dev->ifindex &&
600 opt->srcrt == 0 && !skb_sec_path(skb)) {
601 struct in6_addr *target = NULL;
602 struct inet_peer *peer;
603 struct rt6_info *rt;
604
605 /*
606 * incoming and outgoing devices are the same
607 * send a redirect.
608 */
609
610 rt = dst_rt6_info(dst);
611 if (rt->rt6i_flags & RTF_GATEWAY)
612 target = &rt->rt6i_gateway;
613 else
614 target = &hdr->daddr;
615
616 rcu_read_lock();
617 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr);
618
619 /* Limit redirects both by destination (here)
620 and by source (inside ndisc_send_redirect)
621 */
622 if (inet_peer_xrlim_allow(peer, 1*HZ))
623 ndisc_send_redirect(skb, target);
624 rcu_read_unlock();
625 } else {
626 int addrtype = ipv6_addr_type(&hdr->saddr);
627
628 /* This check is security critical. */
629 if (addrtype == IPV6_ADDR_ANY ||
630 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
631 goto error;
632 if (addrtype & IPV6_ADDR_LINKLOCAL) {
633 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
634 ICMPV6_NOT_NEIGHBOUR, 0);
635 goto error;
636 }
637 }
638
639 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
640
641 mtu = ip6_dst_mtu_maybe_forward(dst, true);
642 if (mtu < IPV6_MIN_MTU)
643 mtu = IPV6_MIN_MTU;
644
645 if (ip6_pkt_too_big(skb, mtu)) {
646 /* Again, force OUTPUT device used as source address */
647 skb->dev = dst->dev;
648 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
649 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
650 __IP6_INC_STATS(net, ip6_dst_idev(dst),
651 IPSTATS_MIB_FRAGFAILS);
652 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
653 return -EMSGSIZE;
654 }
655
656 if (skb_cow(skb, dst->dev->hard_header_len)) {
657 __IP6_INC_STATS(net, ip6_dst_idev(dst),
658 IPSTATS_MIB_OUTDISCARDS);
659 goto drop;
660 }
661
662 hdr = ipv6_hdr(skb);
663
664 /* Mangling hops number delayed to point after skb COW */
665
666 hdr->hop_limit--;
667
668 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
669 net, NULL, skb, skb->dev, dst->dev,
670 ip6_forward_finish);
671
672error:
673 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
674 SKB_DR_SET(reason, IP_INADDRERRORS);
675drop:
676 kfree_skb_reason(skb, reason);
677 return -EINVAL;
678}
679
680static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
681{
682 to->pkt_type = from->pkt_type;
683 to->priority = from->priority;
684 to->protocol = from->protocol;
685 skb_dst_drop(to);
686 skb_dst_set(to, dst_clone(skb_dst(from)));
687 to->dev = from->dev;
688 to->mark = from->mark;
689
690 skb_copy_hash(to, from);
691
692#ifdef CONFIG_NET_SCHED
693 to->tc_index = from->tc_index;
694#endif
695 nf_copy(to, from);
696 skb_ext_copy(to, from);
697 skb_copy_secmark(to, from);
698}
699
700int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
701 u8 nexthdr, __be32 frag_id,
702 struct ip6_fraglist_iter *iter)
703{
704 unsigned int first_len;
705 struct frag_hdr *fh;
706
707 /* BUILD HEADER */
708 *prevhdr = NEXTHDR_FRAGMENT;
709 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
710 if (!iter->tmp_hdr)
711 return -ENOMEM;
712
713 iter->frag = skb_shinfo(skb)->frag_list;
714 skb_frag_list_init(skb);
715
716 iter->offset = 0;
717 iter->hlen = hlen;
718 iter->frag_id = frag_id;
719 iter->nexthdr = nexthdr;
720
721 __skb_pull(skb, hlen);
722 fh = __skb_push(skb, sizeof(struct frag_hdr));
723 __skb_push(skb, hlen);
724 skb_reset_network_header(skb);
725 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
726
727 fh->nexthdr = nexthdr;
728 fh->reserved = 0;
729 fh->frag_off = htons(IP6_MF);
730 fh->identification = frag_id;
731
732 first_len = skb_pagelen(skb);
733 skb->data_len = first_len - skb_headlen(skb);
734 skb->len = first_len;
735 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
736
737 return 0;
738}
739EXPORT_SYMBOL(ip6_fraglist_init);
740
741void ip6_fraglist_prepare(struct sk_buff *skb,
742 struct ip6_fraglist_iter *iter)
743{
744 struct sk_buff *frag = iter->frag;
745 unsigned int hlen = iter->hlen;
746 struct frag_hdr *fh;
747
748 frag->ip_summed = CHECKSUM_NONE;
749 skb_reset_transport_header(frag);
750 fh = __skb_push(frag, sizeof(struct frag_hdr));
751 __skb_push(frag, hlen);
752 skb_reset_network_header(frag);
753 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
754 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
755 fh->nexthdr = iter->nexthdr;
756 fh->reserved = 0;
757 fh->frag_off = htons(iter->offset);
758 if (frag->next)
759 fh->frag_off |= htons(IP6_MF);
760 fh->identification = iter->frag_id;
761 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
762 ip6_copy_metadata(frag, skb);
763}
764EXPORT_SYMBOL(ip6_fraglist_prepare);
765
766void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
767 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
768 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
769{
770 state->prevhdr = prevhdr;
771 state->nexthdr = nexthdr;
772 state->frag_id = frag_id;
773
774 state->hlen = hlen;
775 state->mtu = mtu;
776
777 state->left = skb->len - hlen; /* Space per frame */
778 state->ptr = hlen; /* Where to start from */
779
780 state->hroom = hdr_room;
781 state->troom = needed_tailroom;
782
783 state->offset = 0;
784}
785EXPORT_SYMBOL(ip6_frag_init);
786
787struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
788{
789 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
790 struct sk_buff *frag;
791 struct frag_hdr *fh;
792 unsigned int len;
793
794 len = state->left;
795 /* IF: it doesn't fit, use 'mtu' - the data space left */
796 if (len > state->mtu)
797 len = state->mtu;
798 /* IF: we are not sending up to and including the packet end
799 then align the next start on an eight byte boundary */
800 if (len < state->left)
801 len &= ~7;
802
803 /* Allocate buffer */
804 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
805 state->hroom + state->troom, GFP_ATOMIC);
806 if (!frag)
807 return ERR_PTR(-ENOMEM);
808
809 /*
810 * Set up data on packet
811 */
812
813 ip6_copy_metadata(frag, skb);
814 skb_reserve(frag, state->hroom);
815 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
816 skb_reset_network_header(frag);
817 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
818 frag->transport_header = (frag->network_header + state->hlen +
819 sizeof(struct frag_hdr));
820
821 /*
822 * Charge the memory for the fragment to any owner
823 * it might possess
824 */
825 if (skb->sk)
826 skb_set_owner_w(frag, skb->sk);
827
828 /*
829 * Copy the packet header into the new buffer.
830 */
831 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
832
833 fragnexthdr_offset = skb_network_header(frag);
834 fragnexthdr_offset += prevhdr - skb_network_header(skb);
835 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
836
837 /*
838 * Build fragment header.
839 */
840 fh->nexthdr = state->nexthdr;
841 fh->reserved = 0;
842 fh->identification = state->frag_id;
843
844 /*
845 * Copy a block of the IP datagram.
846 */
847 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
848 len));
849 state->left -= len;
850
851 fh->frag_off = htons(state->offset);
852 if (state->left > 0)
853 fh->frag_off |= htons(IP6_MF);
854 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
855
856 state->ptr += len;
857 state->offset += len;
858
859 return frag;
860}
861EXPORT_SYMBOL(ip6_frag_next);
862
863int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
864 int (*output)(struct net *, struct sock *, struct sk_buff *))
865{
866 struct sk_buff *frag;
867 struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
868 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
869 inet6_sk(skb->sk) : NULL;
870 u8 tstamp_type = skb->tstamp_type;
871 struct ip6_frag_state state;
872 unsigned int mtu, hlen, nexthdr_offset;
873 ktime_t tstamp = skb->tstamp;
874 int hroom, err = 0;
875 __be32 frag_id;
876 u8 *prevhdr, nexthdr = 0;
877
878 err = ip6_find_1stfragopt(skb, &prevhdr);
879 if (err < 0)
880 goto fail;
881 hlen = err;
882 nexthdr = *prevhdr;
883 nexthdr_offset = prevhdr - skb_network_header(skb);
884
885 mtu = ip6_skb_dst_mtu(skb);
886
887 /* We must not fragment if the socket is set to force MTU discovery
888 * or if the skb it not generated by a local socket.
889 */
890 if (unlikely(!skb->ignore_df && skb->len > mtu))
891 goto fail_toobig;
892
893 if (IP6CB(skb)->frag_max_size) {
894 if (IP6CB(skb)->frag_max_size > mtu)
895 goto fail_toobig;
896
897 /* don't send fragments larger than what we received */
898 mtu = IP6CB(skb)->frag_max_size;
899 if (mtu < IPV6_MIN_MTU)
900 mtu = IPV6_MIN_MTU;
901 }
902
903 if (np) {
904 u32 frag_size = READ_ONCE(np->frag_size);
905
906 if (frag_size && frag_size < mtu)
907 mtu = frag_size;
908 }
909 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
910 goto fail_toobig;
911 mtu -= hlen + sizeof(struct frag_hdr);
912
913 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
914 &ipv6_hdr(skb)->saddr);
915
916 if (skb->ip_summed == CHECKSUM_PARTIAL &&
917 (err = skb_checksum_help(skb)))
918 goto fail;
919
920 prevhdr = skb_network_header(skb) + nexthdr_offset;
921 hroom = LL_RESERVED_SPACE(rt->dst.dev);
922 if (skb_has_frag_list(skb)) {
923 unsigned int first_len = skb_pagelen(skb);
924 struct ip6_fraglist_iter iter;
925 struct sk_buff *frag2;
926
927 if (first_len - hlen > mtu ||
928 ((first_len - hlen) & 7) ||
929 skb_cloned(skb) ||
930 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
931 goto slow_path;
932
933 skb_walk_frags(skb, frag) {
934 /* Correct geometry. */
935 if (frag->len > mtu ||
936 ((frag->len & 7) && frag->next) ||
937 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
938 goto slow_path_clean;
939
940 /* Partially cloned skb? */
941 if (skb_shared(frag))
942 goto slow_path_clean;
943
944 BUG_ON(frag->sk);
945 if (skb->sk) {
946 frag->sk = skb->sk;
947 frag->destructor = sock_wfree;
948 }
949 skb->truesize -= frag->truesize;
950 }
951
952 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
953 &iter);
954 if (err < 0)
955 goto fail;
956
957 /* We prevent @rt from being freed. */
958 rcu_read_lock();
959
960 for (;;) {
961 /* Prepare header of the next frame,
962 * before previous one went down. */
963 if (iter.frag)
964 ip6_fraglist_prepare(skb, &iter);
965
966 skb_set_delivery_time(skb, tstamp, tstamp_type);
967 err = output(net, sk, skb);
968 if (!err)
969 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
970 IPSTATS_MIB_FRAGCREATES);
971
972 if (err || !iter.frag)
973 break;
974
975 skb = ip6_fraglist_next(&iter);
976 }
977
978 kfree(iter.tmp_hdr);
979
980 if (err == 0) {
981 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
982 IPSTATS_MIB_FRAGOKS);
983 rcu_read_unlock();
984 return 0;
985 }
986
987 kfree_skb_list(iter.frag);
988
989 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
990 IPSTATS_MIB_FRAGFAILS);
991 rcu_read_unlock();
992 return err;
993
994slow_path_clean:
995 skb_walk_frags(skb, frag2) {
996 if (frag2 == frag)
997 break;
998 frag2->sk = NULL;
999 frag2->destructor = NULL;
1000 skb->truesize += frag2->truesize;
1001 }
1002 }
1003
1004slow_path:
1005 /*
1006 * Fragment the datagram.
1007 */
1008
1009 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
1010 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
1011 &state);
1012
1013 /*
1014 * Keep copying data until we run out.
1015 */
1016
1017 while (state.left > 0) {
1018 frag = ip6_frag_next(skb, &state);
1019 if (IS_ERR(frag)) {
1020 err = PTR_ERR(frag);
1021 goto fail;
1022 }
1023
1024 /*
1025 * Put this fragment into the sending queue.
1026 */
1027 skb_set_delivery_time(frag, tstamp, tstamp_type);
1028 err = output(net, sk, frag);
1029 if (err)
1030 goto fail;
1031
1032 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1033 IPSTATS_MIB_FRAGCREATES);
1034 }
1035 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1036 IPSTATS_MIB_FRAGOKS);
1037 consume_skb(skb);
1038 return err;
1039
1040fail_toobig:
1041 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1042 err = -EMSGSIZE;
1043
1044fail:
1045 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1046 IPSTATS_MIB_FRAGFAILS);
1047 kfree_skb(skb);
1048 return err;
1049}
1050
1051static inline int ip6_rt_check(const struct rt6key *rt_key,
1052 const struct in6_addr *fl_addr,
1053 const struct in6_addr *addr_cache)
1054{
1055 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1056 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1057}
1058
1059static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1060 struct dst_entry *dst,
1061 const struct flowi6 *fl6)
1062{
1063 struct ipv6_pinfo *np = inet6_sk(sk);
1064 struct rt6_info *rt;
1065
1066 if (!dst)
1067 goto out;
1068
1069 if (dst->ops->family != AF_INET6) {
1070 dst_release(dst);
1071 return NULL;
1072 }
1073
1074 rt = dst_rt6_info(dst);
1075 /* Yes, checking route validity in not connected
1076 * case is not very simple. Take into account,
1077 * that we do not support routing by source, TOS,
1078 * and MSG_DONTROUTE --ANK (980726)
1079 *
1080 * 1. ip6_rt_check(): If route was host route,
1081 * check that cached destination is current.
1082 * If it is network route, we still may
1083 * check its validity using saved pointer
1084 * to the last used address: daddr_cache.
1085 * We do not want to save whole address now,
1086 * (because main consumer of this service
1087 * is tcp, which has not this problem),
1088 * so that the last trick works only on connected
1089 * sockets.
1090 * 2. oif also should be the same.
1091 */
1092 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1093#ifdef CONFIG_IPV6_SUBTREES
1094 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1095#endif
1096 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1097 dst_release(dst);
1098 dst = NULL;
1099 }
1100
1101out:
1102 return dst;
1103}
1104
1105static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1106 struct dst_entry **dst, struct flowi6 *fl6)
1107{
1108#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1109 struct neighbour *n;
1110 struct rt6_info *rt;
1111#endif
1112 int err;
1113 int flags = 0;
1114
1115 /* The correct way to handle this would be to do
1116 * ip6_route_get_saddr, and then ip6_route_output; however,
1117 * the route-specific preferred source forces the
1118 * ip6_route_output call _before_ ip6_route_get_saddr.
1119 *
1120 * In source specific routing (no src=any default route),
1121 * ip6_route_output will fail given src=any saddr, though, so
1122 * that's why we try it again later.
1123 */
1124 if (ipv6_addr_any(&fl6->saddr)) {
1125 struct fib6_info *from;
1126 struct rt6_info *rt;
1127
1128 *dst = ip6_route_output(net, sk, fl6);
1129 rt = (*dst)->error ? NULL : dst_rt6_info(*dst);
1130
1131 rcu_read_lock();
1132 from = rt ? rcu_dereference(rt->from) : NULL;
1133 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1134 sk ? READ_ONCE(inet6_sk(sk)->srcprefs) : 0,
1135 fl6->flowi6_l3mdev,
1136 &fl6->saddr);
1137 rcu_read_unlock();
1138
1139 if (err)
1140 goto out_err_release;
1141
1142 /* If we had an erroneous initial result, pretend it
1143 * never existed and let the SA-enabled version take
1144 * over.
1145 */
1146 if ((*dst)->error) {
1147 dst_release(*dst);
1148 *dst = NULL;
1149 }
1150
1151 if (fl6->flowi6_oif)
1152 flags |= RT6_LOOKUP_F_IFACE;
1153 }
1154
1155 if (!*dst)
1156 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1157
1158 err = (*dst)->error;
1159 if (err)
1160 goto out_err_release;
1161
1162#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1163 /*
1164 * Here if the dst entry we've looked up
1165 * has a neighbour entry that is in the INCOMPLETE
1166 * state and the src address from the flow is
1167 * marked as OPTIMISTIC, we release the found
1168 * dst entry and replace it instead with the
1169 * dst entry of the nexthop router
1170 */
1171 rt = dst_rt6_info(*dst);
1172 rcu_read_lock();
1173 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1174 rt6_nexthop(rt, &fl6->daddr));
1175 err = n && !(READ_ONCE(n->nud_state) & NUD_VALID) ? -EINVAL : 0;
1176 rcu_read_unlock();
1177
1178 if (err) {
1179 struct inet6_ifaddr *ifp;
1180 struct flowi6 fl_gw6;
1181 int redirect;
1182
1183 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1184 (*dst)->dev, 1);
1185
1186 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1187 if (ifp)
1188 in6_ifa_put(ifp);
1189
1190 if (redirect) {
1191 /*
1192 * We need to get the dst entry for the
1193 * default router instead
1194 */
1195 dst_release(*dst);
1196 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1197 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1198 *dst = ip6_route_output(net, sk, &fl_gw6);
1199 err = (*dst)->error;
1200 if (err)
1201 goto out_err_release;
1202 }
1203 }
1204#endif
1205 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1206 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1207 err = -EAFNOSUPPORT;
1208 goto out_err_release;
1209 }
1210
1211 return 0;
1212
1213out_err_release:
1214 dst_release(*dst);
1215 *dst = NULL;
1216
1217 if (err == -ENETUNREACH)
1218 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1219 return err;
1220}
1221
1222/**
1223 * ip6_dst_lookup - perform route lookup on flow
1224 * @net: Network namespace to perform lookup in
1225 * @sk: socket which provides route info
1226 * @dst: pointer to dst_entry * for result
1227 * @fl6: flow to lookup
1228 *
1229 * This function performs a route lookup on the given flow.
1230 *
1231 * It returns zero on success, or a standard errno code on error.
1232 */
1233int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1234 struct flowi6 *fl6)
1235{
1236 *dst = NULL;
1237 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1238}
1239EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1240
1241/**
1242 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1243 * @net: Network namespace to perform lookup in
1244 * @sk: socket which provides route info
1245 * @fl6: flow to lookup
1246 * @final_dst: final destination address for ipsec lookup
1247 *
1248 * This function performs a route lookup on the given flow.
1249 *
1250 * It returns a valid dst pointer on success, or a pointer encoded
1251 * error code.
1252 */
1253struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1254 const struct in6_addr *final_dst)
1255{
1256 struct dst_entry *dst = NULL;
1257 int err;
1258
1259 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1260 if (err)
1261 return ERR_PTR(err);
1262 if (final_dst)
1263 fl6->daddr = *final_dst;
1264
1265 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1266}
1267EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1268
1269/**
1270 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1271 * @sk: socket which provides the dst cache and route info
1272 * @fl6: flow to lookup
1273 * @final_dst: final destination address for ipsec lookup
1274 * @connected: whether @sk is connected or not
1275 *
1276 * This function performs a route lookup on the given flow with the
1277 * possibility of using the cached route in the socket if it is valid.
1278 * It will take the socket dst lock when operating on the dst cache.
1279 * As a result, this function can only be used in process context.
1280 *
1281 * In addition, for a connected socket, cache the dst in the socket
1282 * if the current cache is not valid.
1283 *
1284 * It returns a valid dst pointer on success, or a pointer encoded
1285 * error code.
1286 */
1287struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1288 const struct in6_addr *final_dst,
1289 bool connected)
1290{
1291 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1292
1293 dst = ip6_sk_dst_check(sk, dst, fl6);
1294 if (dst)
1295 return dst;
1296
1297 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1298 if (connected && !IS_ERR(dst))
1299 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1300
1301 return dst;
1302}
1303EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1304
1305static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1306 gfp_t gfp)
1307{
1308 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1309}
1310
1311static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1312 gfp_t gfp)
1313{
1314 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1315}
1316
1317static void ip6_append_data_mtu(unsigned int *mtu,
1318 int *maxfraglen,
1319 unsigned int fragheaderlen,
1320 struct sk_buff *skb,
1321 struct rt6_info *rt,
1322 unsigned int orig_mtu)
1323{
1324 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1325 if (!skb) {
1326 /* first fragment, reserve header_len */
1327 *mtu = orig_mtu - rt->dst.header_len;
1328
1329 } else {
1330 /*
1331 * this fragment is not first, the headers
1332 * space is regarded as data space.
1333 */
1334 *mtu = orig_mtu;
1335 }
1336 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1337 + fragheaderlen - sizeof(struct frag_hdr);
1338 }
1339}
1340
1341static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1342 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1343 struct rt6_info *rt)
1344{
1345 struct ipv6_pinfo *np = inet6_sk(sk);
1346 unsigned int mtu, frag_size;
1347 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1348
1349 /* callers pass dst together with a reference, set it first so
1350 * ip6_cork_release() can put it down even in case of an error.
1351 */
1352 cork->base.dst = &rt->dst;
1353
1354 /*
1355 * setup for corking
1356 */
1357 if (opt) {
1358 if (WARN_ON(v6_cork->opt))
1359 return -EINVAL;
1360
1361 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1362 if (unlikely(!nopt))
1363 return -ENOBUFS;
1364
1365 nopt->tot_len = sizeof(*opt);
1366 nopt->opt_flen = opt->opt_flen;
1367 nopt->opt_nflen = opt->opt_nflen;
1368
1369 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1370 if (opt->dst0opt && !nopt->dst0opt)
1371 return -ENOBUFS;
1372
1373 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1374 if (opt->dst1opt && !nopt->dst1opt)
1375 return -ENOBUFS;
1376
1377 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1378 if (opt->hopopt && !nopt->hopopt)
1379 return -ENOBUFS;
1380
1381 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1382 if (opt->srcrt && !nopt->srcrt)
1383 return -ENOBUFS;
1384
1385 /* need source address above miyazawa*/
1386 }
1387 v6_cork->hop_limit = ipc6->hlimit;
1388 v6_cork->tclass = ipc6->tclass;
1389 if (rt->dst.flags & DST_XFRM_TUNNEL)
1390 mtu = READ_ONCE(np->pmtudisc) >= IPV6_PMTUDISC_PROBE ?
1391 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1392 else
1393 mtu = READ_ONCE(np->pmtudisc) >= IPV6_PMTUDISC_PROBE ?
1394 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1395
1396 frag_size = READ_ONCE(np->frag_size);
1397 if (frag_size && frag_size < mtu)
1398 mtu = frag_size;
1399
1400 cork->base.fragsize = mtu;
1401 cork->base.gso_size = ipc6->gso_size;
1402 cork->base.tx_flags = 0;
1403 cork->base.mark = ipc6->sockc.mark;
1404 sock_tx_timestamp(sk, &ipc6->sockc, &cork->base.tx_flags);
1405 if (ipc6->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1406 cork->base.flags |= IPCORK_TS_OPT_ID;
1407 cork->base.ts_opt_id = ipc6->sockc.ts_opt_id;
1408 }
1409 cork->base.length = 0;
1410 cork->base.transmit_time = ipc6->sockc.transmit_time;
1411
1412 return 0;
1413}
1414
1415static int __ip6_append_data(struct sock *sk,
1416 struct sk_buff_head *queue,
1417 struct inet_cork_full *cork_full,
1418 struct inet6_cork *v6_cork,
1419 struct page_frag *pfrag,
1420 int getfrag(void *from, char *to, int offset,
1421 int len, int odd, struct sk_buff *skb),
1422 void *from, size_t length, int transhdrlen,
1423 unsigned int flags, struct ipcm6_cookie *ipc6)
1424{
1425 struct sk_buff *skb, *skb_prev = NULL;
1426 struct inet_cork *cork = &cork_full->base;
1427 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1428 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1429 struct ubuf_info *uarg = NULL;
1430 int exthdrlen = 0;
1431 int dst_exthdrlen = 0;
1432 int hh_len;
1433 int copy;
1434 int err;
1435 int offset = 0;
1436 bool zc = false;
1437 u32 tskey = 0;
1438 struct rt6_info *rt = dst_rt6_info(cork->dst);
1439 bool paged, hold_tskey = false, extra_uref = false;
1440 struct ipv6_txoptions *opt = v6_cork->opt;
1441 int csummode = CHECKSUM_NONE;
1442 unsigned int maxnonfragsize, headersize;
1443 unsigned int wmem_alloc_delta = 0;
1444
1445 skb = skb_peek_tail(queue);
1446 if (!skb) {
1447 exthdrlen = opt ? opt->opt_flen : 0;
1448 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1449 }
1450
1451 paged = !!cork->gso_size;
1452 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1453 orig_mtu = mtu;
1454
1455 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1456
1457 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1458 (opt ? opt->opt_nflen : 0);
1459
1460 headersize = sizeof(struct ipv6hdr) +
1461 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1462 rt->rt6i_nfheader_len;
1463
1464 if (mtu <= fragheaderlen ||
1465 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1466 goto emsgsize;
1467
1468 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1469 sizeof(struct frag_hdr);
1470
1471 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1472 * the first fragment
1473 */
1474 if (headersize + transhdrlen > mtu)
1475 goto emsgsize;
1476
1477 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1478 (sk->sk_protocol == IPPROTO_UDP ||
1479 sk->sk_protocol == IPPROTO_ICMPV6 ||
1480 sk->sk_protocol == IPPROTO_RAW)) {
1481 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1482 sizeof(struct ipv6hdr));
1483 goto emsgsize;
1484 }
1485
1486 if (ip6_sk_ignore_df(sk))
1487 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1488 else
1489 maxnonfragsize = mtu;
1490
1491 if (cork->length + length > maxnonfragsize - headersize) {
1492emsgsize:
1493 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1494 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1495 return -EMSGSIZE;
1496 }
1497
1498 /* CHECKSUM_PARTIAL only with no extension headers and when
1499 * we are not going to fragment
1500 */
1501 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1502 headersize == sizeof(struct ipv6hdr) &&
1503 length <= mtu - headersize &&
1504 (!(flags & MSG_MORE) || cork->gso_size) &&
1505 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1506 csummode = CHECKSUM_PARTIAL;
1507
1508 if ((flags & MSG_ZEROCOPY) && length) {
1509 struct msghdr *msg = from;
1510
1511 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1512 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1513 return -EINVAL;
1514
1515 /* Leave uarg NULL if can't zerocopy, callers should
1516 * be able to handle it.
1517 */
1518 if ((rt->dst.dev->features & NETIF_F_SG) &&
1519 csummode == CHECKSUM_PARTIAL) {
1520 paged = true;
1521 zc = true;
1522 uarg = msg->msg_ubuf;
1523 }
1524 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1525 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1526 if (!uarg)
1527 return -ENOBUFS;
1528 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1529 if (rt->dst.dev->features & NETIF_F_SG &&
1530 csummode == CHECKSUM_PARTIAL) {
1531 paged = true;
1532 zc = true;
1533 } else {
1534 uarg_to_msgzc(uarg)->zerocopy = 0;
1535 skb_zcopy_set(skb, uarg, &extra_uref);
1536 }
1537 }
1538 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1539 if (inet_test_bit(HDRINCL, sk))
1540 return -EPERM;
1541 if (rt->dst.dev->features & NETIF_F_SG &&
1542 getfrag == ip_generic_getfrag)
1543 /* We need an empty buffer to attach stuff to */
1544 paged = true;
1545 else
1546 flags &= ~MSG_SPLICE_PAGES;
1547 }
1548
1549 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1550 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1551 if (cork->flags & IPCORK_TS_OPT_ID) {
1552 tskey = cork->ts_opt_id;
1553 } else {
1554 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1555 hold_tskey = true;
1556 }
1557 }
1558
1559 /*
1560 * Let's try using as much space as possible.
1561 * Use MTU if total length of the message fits into the MTU.
1562 * Otherwise, we need to reserve fragment header and
1563 * fragment alignment (= 8-15 octects, in total).
1564 *
1565 * Note that we may need to "move" the data from the tail
1566 * of the buffer to the new fragment when we split
1567 * the message.
1568 *
1569 * FIXME: It may be fragmented into multiple chunks
1570 * at once if non-fragmentable extension headers
1571 * are too large.
1572 * --yoshfuji
1573 */
1574
1575 cork->length += length;
1576 if (!skb)
1577 goto alloc_new_skb;
1578
1579 while (length > 0) {
1580 /* Check if the remaining data fits into current packet. */
1581 copy = (cork->length <= mtu ? mtu : maxfraglen) - skb->len;
1582 if (copy < length)
1583 copy = maxfraglen - skb->len;
1584
1585 if (copy <= 0) {
1586 char *data;
1587 unsigned int datalen;
1588 unsigned int fraglen;
1589 unsigned int fraggap;
1590 unsigned int alloclen, alloc_extra;
1591 unsigned int pagedlen;
1592alloc_new_skb:
1593 /* There's no room in the current skb */
1594 if (skb)
1595 fraggap = skb->len - maxfraglen;
1596 else
1597 fraggap = 0;
1598 /* update mtu and maxfraglen if necessary */
1599 if (!skb || !skb_prev)
1600 ip6_append_data_mtu(&mtu, &maxfraglen,
1601 fragheaderlen, skb, rt,
1602 orig_mtu);
1603
1604 skb_prev = skb;
1605
1606 /*
1607 * If remaining data exceeds the mtu,
1608 * we know we need more fragment(s).
1609 */
1610 datalen = length + fraggap;
1611
1612 if (datalen > (cork->length <= mtu ? mtu : maxfraglen) - fragheaderlen)
1613 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1614 fraglen = datalen + fragheaderlen;
1615 pagedlen = 0;
1616
1617 alloc_extra = hh_len;
1618 alloc_extra += dst_exthdrlen;
1619 alloc_extra += rt->dst.trailer_len;
1620
1621 /* We just reserve space for fragment header.
1622 * Note: this may be overallocation if the message
1623 * (without MSG_MORE) fits into the MTU.
1624 */
1625 alloc_extra += sizeof(struct frag_hdr);
1626
1627 if ((flags & MSG_MORE) &&
1628 !(rt->dst.dev->features&NETIF_F_SG))
1629 alloclen = mtu;
1630 else if (!paged &&
1631 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1632 !(rt->dst.dev->features & NETIF_F_SG)))
1633 alloclen = fraglen;
1634 else {
1635 alloclen = fragheaderlen + transhdrlen;
1636 pagedlen = datalen - transhdrlen;
1637 }
1638 alloclen += alloc_extra;
1639
1640 if (datalen != length + fraggap) {
1641 /*
1642 * this is not the last fragment, the trailer
1643 * space is regarded as data space.
1644 */
1645 datalen += rt->dst.trailer_len;
1646 }
1647
1648 fraglen = datalen + fragheaderlen;
1649
1650 copy = datalen - transhdrlen - fraggap - pagedlen;
1651 /* [!] NOTE: copy may be negative if pagedlen>0
1652 * because then the equation may reduces to -fraggap.
1653 */
1654 if (copy < 0 && !(flags & MSG_SPLICE_PAGES)) {
1655 err = -EINVAL;
1656 goto error;
1657 }
1658 if (transhdrlen) {
1659 skb = sock_alloc_send_skb(sk, alloclen,
1660 (flags & MSG_DONTWAIT), &err);
1661 } else {
1662 skb = NULL;
1663 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1664 2 * sk->sk_sndbuf)
1665 skb = alloc_skb(alloclen,
1666 sk->sk_allocation);
1667 if (unlikely(!skb))
1668 err = -ENOBUFS;
1669 }
1670 if (!skb)
1671 goto error;
1672 /*
1673 * Fill in the control structures
1674 */
1675 skb->protocol = htons(ETH_P_IPV6);
1676 skb->ip_summed = csummode;
1677 skb->csum = 0;
1678 /* reserve for fragmentation and ipsec header */
1679 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1680 dst_exthdrlen);
1681
1682 /*
1683 * Find where to start putting bytes
1684 */
1685 data = skb_put(skb, fraglen - pagedlen);
1686 skb_set_network_header(skb, exthdrlen);
1687 data += fragheaderlen;
1688 skb->transport_header = (skb->network_header +
1689 fragheaderlen);
1690 if (fraggap) {
1691 skb->csum = skb_copy_and_csum_bits(
1692 skb_prev, maxfraglen,
1693 data + transhdrlen, fraggap);
1694 skb_prev->csum = csum_sub(skb_prev->csum,
1695 skb->csum);
1696 data += fraggap;
1697 pskb_trim_unique(skb_prev, maxfraglen);
1698 }
1699 if (copy > 0 &&
1700 getfrag(from, data + transhdrlen, offset,
1701 copy, fraggap, skb) < 0) {
1702 err = -EFAULT;
1703 kfree_skb(skb);
1704 goto error;
1705 } else if (flags & MSG_SPLICE_PAGES) {
1706 copy = 0;
1707 }
1708
1709 offset += copy;
1710 length -= copy + transhdrlen;
1711 transhdrlen = 0;
1712 exthdrlen = 0;
1713 dst_exthdrlen = 0;
1714
1715 /* Only the initial fragment is time stamped */
1716 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1717 cork->tx_flags = 0;
1718 skb_shinfo(skb)->tskey = tskey;
1719 tskey = 0;
1720 skb_zcopy_set(skb, uarg, &extra_uref);
1721
1722 if ((flags & MSG_CONFIRM) && !skb_prev)
1723 skb_set_dst_pending_confirm(skb, 1);
1724
1725 /*
1726 * Put the packet on the pending queue
1727 */
1728 if (!skb->destructor) {
1729 skb->destructor = sock_wfree;
1730 skb->sk = sk;
1731 wmem_alloc_delta += skb->truesize;
1732 }
1733 __skb_queue_tail(queue, skb);
1734 continue;
1735 }
1736
1737 if (copy > length)
1738 copy = length;
1739
1740 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1741 skb_tailroom(skb) >= copy) {
1742 unsigned int off;
1743
1744 off = skb->len;
1745 if (getfrag(from, skb_put(skb, copy),
1746 offset, copy, off, skb) < 0) {
1747 __skb_trim(skb, off);
1748 err = -EFAULT;
1749 goto error;
1750 }
1751 } else if (flags & MSG_SPLICE_PAGES) {
1752 struct msghdr *msg = from;
1753
1754 err = -EIO;
1755 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1756 goto error;
1757
1758 err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1759 sk->sk_allocation);
1760 if (err < 0)
1761 goto error;
1762 copy = err;
1763 wmem_alloc_delta += copy;
1764 } else if (!zc) {
1765 int i = skb_shinfo(skb)->nr_frags;
1766
1767 err = -ENOMEM;
1768 if (!sk_page_frag_refill(sk, pfrag))
1769 goto error;
1770
1771 skb_zcopy_downgrade_managed(skb);
1772 if (!skb_can_coalesce(skb, i, pfrag->page,
1773 pfrag->offset)) {
1774 err = -EMSGSIZE;
1775 if (i == MAX_SKB_FRAGS)
1776 goto error;
1777
1778 __skb_fill_page_desc(skb, i, pfrag->page,
1779 pfrag->offset, 0);
1780 skb_shinfo(skb)->nr_frags = ++i;
1781 get_page(pfrag->page);
1782 }
1783 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1784 if (getfrag(from,
1785 page_address(pfrag->page) + pfrag->offset,
1786 offset, copy, skb->len, skb) < 0)
1787 goto error_efault;
1788
1789 pfrag->offset += copy;
1790 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1791 skb->len += copy;
1792 skb->data_len += copy;
1793 skb->truesize += copy;
1794 wmem_alloc_delta += copy;
1795 } else {
1796 err = skb_zerocopy_iter_dgram(skb, from, copy);
1797 if (err < 0)
1798 goto error;
1799 }
1800 offset += copy;
1801 length -= copy;
1802 }
1803
1804 if (wmem_alloc_delta)
1805 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1806 return 0;
1807
1808error_efault:
1809 err = -EFAULT;
1810error:
1811 net_zcopy_put_abort(uarg, extra_uref);
1812 cork->length -= length;
1813 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1814 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1815 if (hold_tskey)
1816 atomic_dec(&sk->sk_tskey);
1817 return err;
1818}
1819
1820int ip6_append_data(struct sock *sk,
1821 int getfrag(void *from, char *to, int offset, int len,
1822 int odd, struct sk_buff *skb),
1823 void *from, size_t length, int transhdrlen,
1824 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1825 struct rt6_info *rt, unsigned int flags)
1826{
1827 struct inet_sock *inet = inet_sk(sk);
1828 struct ipv6_pinfo *np = inet6_sk(sk);
1829 int exthdrlen;
1830 int err;
1831
1832 if (flags&MSG_PROBE)
1833 return 0;
1834 if (skb_queue_empty(&sk->sk_write_queue)) {
1835 /*
1836 * setup for corking
1837 */
1838 dst_hold(&rt->dst);
1839 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1840 ipc6, rt);
1841 if (err)
1842 return err;
1843
1844 inet->cork.fl.u.ip6 = *fl6;
1845 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1846 length += exthdrlen;
1847 transhdrlen += exthdrlen;
1848 } else {
1849 transhdrlen = 0;
1850 }
1851
1852 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1853 &np->cork, sk_page_frag(sk), getfrag,
1854 from, length, transhdrlen, flags, ipc6);
1855}
1856EXPORT_SYMBOL_GPL(ip6_append_data);
1857
1858static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1859{
1860 struct dst_entry *dst = cork->base.dst;
1861
1862 cork->base.dst = NULL;
1863 skb_dst_set(skb, dst);
1864}
1865
1866static void ip6_cork_release(struct inet_cork_full *cork,
1867 struct inet6_cork *v6_cork)
1868{
1869 if (v6_cork->opt) {
1870 struct ipv6_txoptions *opt = v6_cork->opt;
1871
1872 kfree(opt->dst0opt);
1873 kfree(opt->dst1opt);
1874 kfree(opt->hopopt);
1875 kfree(opt->srcrt);
1876 kfree(opt);
1877 v6_cork->opt = NULL;
1878 }
1879
1880 if (cork->base.dst) {
1881 dst_release(cork->base.dst);
1882 cork->base.dst = NULL;
1883 }
1884}
1885
1886struct sk_buff *__ip6_make_skb(struct sock *sk,
1887 struct sk_buff_head *queue,
1888 struct inet_cork_full *cork,
1889 struct inet6_cork *v6_cork)
1890{
1891 struct sk_buff *skb, *tmp_skb;
1892 struct sk_buff **tail_skb;
1893 struct in6_addr *final_dst;
1894 struct net *net = sock_net(sk);
1895 struct ipv6hdr *hdr;
1896 struct ipv6_txoptions *opt = v6_cork->opt;
1897 struct rt6_info *rt = dst_rt6_info(cork->base.dst);
1898 struct flowi6 *fl6 = &cork->fl.u.ip6;
1899 unsigned char proto = fl6->flowi6_proto;
1900
1901 skb = __skb_dequeue(queue);
1902 if (!skb)
1903 goto out;
1904 tail_skb = &(skb_shinfo(skb)->frag_list);
1905
1906 /* move skb->data to ip header from ext header */
1907 if (skb->data < skb_network_header(skb))
1908 __skb_pull(skb, skb_network_offset(skb));
1909 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1910 __skb_pull(tmp_skb, skb_network_header_len(skb));
1911 *tail_skb = tmp_skb;
1912 tail_skb = &(tmp_skb->next);
1913 skb->len += tmp_skb->len;
1914 skb->data_len += tmp_skb->len;
1915 skb->truesize += tmp_skb->truesize;
1916 tmp_skb->destructor = NULL;
1917 tmp_skb->sk = NULL;
1918 }
1919
1920 /* Allow local fragmentation. */
1921 skb->ignore_df = ip6_sk_ignore_df(sk);
1922 __skb_pull(skb, skb_network_header_len(skb));
1923
1924 final_dst = &fl6->daddr;
1925 if (opt && opt->opt_flen)
1926 ipv6_push_frag_opts(skb, opt, &proto);
1927 if (opt && opt->opt_nflen)
1928 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1929
1930 skb_push(skb, sizeof(struct ipv6hdr));
1931 skb_reset_network_header(skb);
1932 hdr = ipv6_hdr(skb);
1933
1934 ip6_flow_hdr(hdr, v6_cork->tclass,
1935 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1936 ip6_autoflowlabel(net, sk), fl6));
1937 hdr->hop_limit = v6_cork->hop_limit;
1938 hdr->nexthdr = proto;
1939 hdr->saddr = fl6->saddr;
1940 hdr->daddr = *final_dst;
1941
1942 skb->priority = READ_ONCE(sk->sk_priority);
1943 skb->mark = cork->base.mark;
1944 if (sk_is_tcp(sk))
1945 skb_set_delivery_time(skb, cork->base.transmit_time, SKB_CLOCK_MONOTONIC);
1946 else
1947 skb_set_delivery_type_by_clockid(skb, cork->base.transmit_time, sk->sk_clockid);
1948
1949 ip6_cork_steal_dst(skb, cork);
1950 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1951 if (proto == IPPROTO_ICMPV6) {
1952 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1953 u8 icmp6_type;
1954
1955 if (sk->sk_socket->type == SOCK_RAW &&
1956 !(fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH))
1957 icmp6_type = fl6->fl6_icmp_type;
1958 else
1959 icmp6_type = icmp6_hdr(skb)->icmp6_type;
1960 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_type);
1961 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1962 }
1963
1964 ip6_cork_release(cork, v6_cork);
1965out:
1966 return skb;
1967}
1968
1969int ip6_send_skb(struct sk_buff *skb)
1970{
1971 struct net *net = sock_net(skb->sk);
1972 struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
1973 int err;
1974
1975 rcu_read_lock();
1976 err = ip6_local_out(net, skb->sk, skb);
1977 if (err) {
1978 if (err > 0)
1979 err = net_xmit_errno(err);
1980 if (err)
1981 IP6_INC_STATS(net, rt->rt6i_idev,
1982 IPSTATS_MIB_OUTDISCARDS);
1983 }
1984
1985 rcu_read_unlock();
1986 return err;
1987}
1988
1989int ip6_push_pending_frames(struct sock *sk)
1990{
1991 struct sk_buff *skb;
1992
1993 skb = ip6_finish_skb(sk);
1994 if (!skb)
1995 return 0;
1996
1997 return ip6_send_skb(skb);
1998}
1999EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2000
2001static void __ip6_flush_pending_frames(struct sock *sk,
2002 struct sk_buff_head *queue,
2003 struct inet_cork_full *cork,
2004 struct inet6_cork *v6_cork)
2005{
2006 struct sk_buff *skb;
2007
2008 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2009 if (skb_dst(skb))
2010 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2011 IPSTATS_MIB_OUTDISCARDS);
2012 kfree_skb(skb);
2013 }
2014
2015 ip6_cork_release(cork, v6_cork);
2016}
2017
2018void ip6_flush_pending_frames(struct sock *sk)
2019{
2020 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2021 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2022}
2023EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2024
2025struct sk_buff *ip6_make_skb(struct sock *sk,
2026 int getfrag(void *from, char *to, int offset,
2027 int len, int odd, struct sk_buff *skb),
2028 void *from, size_t length, int transhdrlen,
2029 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2030 unsigned int flags, struct inet_cork_full *cork)
2031{
2032 struct inet6_cork v6_cork;
2033 struct sk_buff_head queue;
2034 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2035 int err;
2036
2037 if (flags & MSG_PROBE) {
2038 dst_release(&rt->dst);
2039 return NULL;
2040 }
2041
2042 __skb_queue_head_init(&queue);
2043
2044 cork->base.flags = 0;
2045 cork->base.addr = 0;
2046 cork->base.opt = NULL;
2047 v6_cork.opt = NULL;
2048 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2049 if (err) {
2050 ip6_cork_release(cork, &v6_cork);
2051 return ERR_PTR(err);
2052 }
2053 if (ipc6->dontfrag < 0)
2054 ipc6->dontfrag = inet6_test_bit(DONTFRAG, sk);
2055
2056 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2057 ¤t->task_frag, getfrag, from,
2058 length + exthdrlen, transhdrlen + exthdrlen,
2059 flags, ipc6);
2060 if (err) {
2061 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2062 return ERR_PTR(err);
2063 }
2064
2065 return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2066}
1/*
2 * IPv6 output functions
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/net/ipv4/ip_output.c
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 *
15 * Changes:
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
20 * etc.
21 *
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
26 * for datagram xmit
27 */
28
29#include <linux/errno.h>
30#include <linux/kernel.h>
31#include <linux/string.h>
32#include <linux/socket.h>
33#include <linux/net.h>
34#include <linux/netdevice.h>
35#include <linux/if_arp.h>
36#include <linux/in6.h>
37#include <linux/tcp.h>
38#include <linux/route.h>
39#include <linux/module.h>
40#include <linux/slab.h>
41
42#include <linux/netfilter.h>
43#include <linux/netfilter_ipv6.h>
44
45#include <net/sock.h>
46#include <net/snmp.h>
47
48#include <net/ipv6.h>
49#include <net/ndisc.h>
50#include <net/protocol.h>
51#include <net/ip6_route.h>
52#include <net/addrconf.h>
53#include <net/rawv6.h>
54#include <net/icmp.h>
55#include <net/xfrm.h>
56#include <net/checksum.h>
57#include <linux/mroute6.h>
58
59int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
60
61int __ip6_local_out(struct sk_buff *skb)
62{
63 int len;
64
65 len = skb->len - sizeof(struct ipv6hdr);
66 if (len > IPV6_MAXPLEN)
67 len = 0;
68 ipv6_hdr(skb)->payload_len = htons(len);
69
70 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
71 skb_dst(skb)->dev, dst_output);
72}
73
74int ip6_local_out(struct sk_buff *skb)
75{
76 int err;
77
78 err = __ip6_local_out(skb);
79 if (likely(err == 1))
80 err = dst_output(skb);
81
82 return err;
83}
84EXPORT_SYMBOL_GPL(ip6_local_out);
85
86/* dev_loopback_xmit for use with netfilter. */
87static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
88{
89 skb_reset_mac_header(newskb);
90 __skb_pull(newskb, skb_network_offset(newskb));
91 newskb->pkt_type = PACKET_LOOPBACK;
92 newskb->ip_summed = CHECKSUM_UNNECESSARY;
93 WARN_ON(!skb_dst(newskb));
94
95 netif_rx_ni(newskb);
96 return 0;
97}
98
99static int ip6_finish_output2(struct sk_buff *skb)
100{
101 struct dst_entry *dst = skb_dst(skb);
102 struct net_device *dev = dst->dev;
103 struct neighbour *neigh;
104
105 skb->protocol = htons(ETH_P_IPV6);
106 skb->dev = dev;
107
108 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
109 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
110
111 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
112 ((mroute6_socket(dev_net(dev), skb) &&
113 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
114 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
115 &ipv6_hdr(skb)->saddr))) {
116 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
117
118 /* Do not check for IFF_ALLMULTI; multicast routing
119 is not supported in any case.
120 */
121 if (newskb)
122 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
123 newskb, NULL, newskb->dev,
124 ip6_dev_loopback_xmit);
125
126 if (ipv6_hdr(skb)->hop_limit == 0) {
127 IP6_INC_STATS(dev_net(dev), idev,
128 IPSTATS_MIB_OUTDISCARDS);
129 kfree_skb(skb);
130 return 0;
131 }
132 }
133
134 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
135 skb->len);
136 }
137
138 rcu_read_lock();
139 neigh = dst_get_neighbour_noref(dst);
140 if (neigh) {
141 int res = neigh_output(neigh, skb);
142
143 rcu_read_unlock();
144 return res;
145 }
146 rcu_read_unlock();
147 IP6_INC_STATS_BH(dev_net(dst->dev),
148 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
149 kfree_skb(skb);
150 return -EINVAL;
151}
152
153static int ip6_finish_output(struct sk_buff *skb)
154{
155 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
156 dst_allfrag(skb_dst(skb)))
157 return ip6_fragment(skb, ip6_finish_output2);
158 else
159 return ip6_finish_output2(skb);
160}
161
162int ip6_output(struct sk_buff *skb)
163{
164 struct net_device *dev = skb_dst(skb)->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
166 if (unlikely(idev->cnf.disable_ipv6)) {
167 IP6_INC_STATS(dev_net(dev), idev,
168 IPSTATS_MIB_OUTDISCARDS);
169 kfree_skb(skb);
170 return 0;
171 }
172
173 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
174 ip6_finish_output,
175 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
176}
177
178/*
179 * xmit an sk_buff (used by TCP, SCTP and DCCP)
180 */
181
182int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
183 struct ipv6_txoptions *opt, int tclass)
184{
185 struct net *net = sock_net(sk);
186 struct ipv6_pinfo *np = inet6_sk(sk);
187 struct in6_addr *first_hop = &fl6->daddr;
188 struct dst_entry *dst = skb_dst(skb);
189 struct ipv6hdr *hdr;
190 u8 proto = fl6->flowi6_proto;
191 int seg_len = skb->len;
192 int hlimit = -1;
193 u32 mtu;
194
195 if (opt) {
196 unsigned int head_room;
197
198 /* First: exthdrs may take lots of space (~8K for now)
199 MAX_HEADER is not enough.
200 */
201 head_room = opt->opt_nflen + opt->opt_flen;
202 seg_len += head_room;
203 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
204
205 if (skb_headroom(skb) < head_room) {
206 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
207 if (skb2 == NULL) {
208 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
209 IPSTATS_MIB_OUTDISCARDS);
210 kfree_skb(skb);
211 return -ENOBUFS;
212 }
213 consume_skb(skb);
214 skb = skb2;
215 skb_set_owner_w(skb, sk);
216 }
217 if (opt->opt_flen)
218 ipv6_push_frag_opts(skb, opt, &proto);
219 if (opt->opt_nflen)
220 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
221 }
222
223 skb_push(skb, sizeof(struct ipv6hdr));
224 skb_reset_network_header(skb);
225 hdr = ipv6_hdr(skb);
226
227 /*
228 * Fill in the IPv6 header
229 */
230 if (np)
231 hlimit = np->hop_limit;
232 if (hlimit < 0)
233 hlimit = ip6_dst_hoplimit(dst);
234
235 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl6->flowlabel;
236
237 hdr->payload_len = htons(seg_len);
238 hdr->nexthdr = proto;
239 hdr->hop_limit = hlimit;
240
241 hdr->saddr = fl6->saddr;
242 hdr->daddr = *first_hop;
243
244 skb->priority = sk->sk_priority;
245 skb->mark = sk->sk_mark;
246
247 mtu = dst_mtu(dst);
248 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
249 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
250 IPSTATS_MIB_OUT, skb->len);
251 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
252 dst->dev, dst_output);
253 }
254
255 net_dbg_ratelimited("IPv6: sending pkt_too_big to self\n");
256 skb->dev = dst->dev;
257 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
258 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
259 kfree_skb(skb);
260 return -EMSGSIZE;
261}
262
263EXPORT_SYMBOL(ip6_xmit);
264
265/*
266 * To avoid extra problems ND packets are send through this
267 * routine. It's code duplication but I really want to avoid
268 * extra checks since ipv6_build_header is used by TCP (which
269 * is for us performance critical)
270 */
271
272int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
273 const struct in6_addr *saddr, const struct in6_addr *daddr,
274 int proto, int len)
275{
276 struct ipv6_pinfo *np = inet6_sk(sk);
277 struct ipv6hdr *hdr;
278
279 skb->protocol = htons(ETH_P_IPV6);
280 skb->dev = dev;
281
282 skb_reset_network_header(skb);
283 skb_put(skb, sizeof(struct ipv6hdr));
284 hdr = ipv6_hdr(skb);
285
286 *(__be32*)hdr = htonl(0x60000000);
287
288 hdr->payload_len = htons(len);
289 hdr->nexthdr = proto;
290 hdr->hop_limit = np->hop_limit;
291
292 hdr->saddr = *saddr;
293 hdr->daddr = *daddr;
294
295 return 0;
296}
297
298static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
299{
300 struct ip6_ra_chain *ra;
301 struct sock *last = NULL;
302
303 read_lock(&ip6_ra_lock);
304 for (ra = ip6_ra_chain; ra; ra = ra->next) {
305 struct sock *sk = ra->sk;
306 if (sk && ra->sel == sel &&
307 (!sk->sk_bound_dev_if ||
308 sk->sk_bound_dev_if == skb->dev->ifindex)) {
309 if (last) {
310 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
311 if (skb2)
312 rawv6_rcv(last, skb2);
313 }
314 last = sk;
315 }
316 }
317
318 if (last) {
319 rawv6_rcv(last, skb);
320 read_unlock(&ip6_ra_lock);
321 return 1;
322 }
323 read_unlock(&ip6_ra_lock);
324 return 0;
325}
326
327static int ip6_forward_proxy_check(struct sk_buff *skb)
328{
329 struct ipv6hdr *hdr = ipv6_hdr(skb);
330 u8 nexthdr = hdr->nexthdr;
331 __be16 frag_off;
332 int offset;
333
334 if (ipv6_ext_hdr(nexthdr)) {
335 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
336 if (offset < 0)
337 return 0;
338 } else
339 offset = sizeof(struct ipv6hdr);
340
341 if (nexthdr == IPPROTO_ICMPV6) {
342 struct icmp6hdr *icmp6;
343
344 if (!pskb_may_pull(skb, (skb_network_header(skb) +
345 offset + 1 - skb->data)))
346 return 0;
347
348 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
349
350 switch (icmp6->icmp6_type) {
351 case NDISC_ROUTER_SOLICITATION:
352 case NDISC_ROUTER_ADVERTISEMENT:
353 case NDISC_NEIGHBOUR_SOLICITATION:
354 case NDISC_NEIGHBOUR_ADVERTISEMENT:
355 case NDISC_REDIRECT:
356 /* For reaction involving unicast neighbor discovery
357 * message destined to the proxied address, pass it to
358 * input function.
359 */
360 return 1;
361 default:
362 break;
363 }
364 }
365
366 /*
367 * The proxying router can't forward traffic sent to a link-local
368 * address, so signal the sender and discard the packet. This
369 * behavior is clarified by the MIPv6 specification.
370 */
371 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
372 dst_link_failure(skb);
373 return -1;
374 }
375
376 return 0;
377}
378
379static inline int ip6_forward_finish(struct sk_buff *skb)
380{
381 return dst_output(skb);
382}
383
384int ip6_forward(struct sk_buff *skb)
385{
386 struct dst_entry *dst = skb_dst(skb);
387 struct ipv6hdr *hdr = ipv6_hdr(skb);
388 struct inet6_skb_parm *opt = IP6CB(skb);
389 struct net *net = dev_net(dst->dev);
390 u32 mtu;
391
392 if (net->ipv6.devconf_all->forwarding == 0)
393 goto error;
394
395 if (skb_warn_if_lro(skb))
396 goto drop;
397
398 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
399 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
400 goto drop;
401 }
402
403 if (skb->pkt_type != PACKET_HOST)
404 goto drop;
405
406 skb_forward_csum(skb);
407
408 /*
409 * We DO NOT make any processing on
410 * RA packets, pushing them to user level AS IS
411 * without ane WARRANTY that application will be able
412 * to interpret them. The reason is that we
413 * cannot make anything clever here.
414 *
415 * We are not end-node, so that if packet contains
416 * AH/ESP, we cannot make anything.
417 * Defragmentation also would be mistake, RA packets
418 * cannot be fragmented, because there is no warranty
419 * that different fragments will go along one path. --ANK
420 */
421 if (opt->ra) {
422 u8 *ptr = skb_network_header(skb) + opt->ra;
423 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
424 return 0;
425 }
426
427 /*
428 * check and decrement ttl
429 */
430 if (hdr->hop_limit <= 1) {
431 /* Force OUTPUT device used as source address */
432 skb->dev = dst->dev;
433 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
434 IP6_INC_STATS_BH(net,
435 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
436
437 kfree_skb(skb);
438 return -ETIMEDOUT;
439 }
440
441 /* XXX: idev->cnf.proxy_ndp? */
442 if (net->ipv6.devconf_all->proxy_ndp &&
443 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
444 int proxied = ip6_forward_proxy_check(skb);
445 if (proxied > 0)
446 return ip6_input(skb);
447 else if (proxied < 0) {
448 IP6_INC_STATS(net, ip6_dst_idev(dst),
449 IPSTATS_MIB_INDISCARDS);
450 goto drop;
451 }
452 }
453
454 if (!xfrm6_route_forward(skb)) {
455 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
456 goto drop;
457 }
458 dst = skb_dst(skb);
459
460 /* IPv6 specs say nothing about it, but it is clear that we cannot
461 send redirects to source routed frames.
462 We don't send redirects to frames decapsulated from IPsec.
463 */
464 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
465 struct in6_addr *target = NULL;
466 struct rt6_info *rt;
467
468 /*
469 * incoming and outgoing devices are the same
470 * send a redirect.
471 */
472
473 rt = (struct rt6_info *) dst;
474 if (rt->rt6i_flags & RTF_GATEWAY)
475 target = &rt->rt6i_gateway;
476 else
477 target = &hdr->daddr;
478
479 if (!rt->rt6i_peer)
480 rt6_bind_peer(rt, 1);
481
482 /* Limit redirects both by destination (here)
483 and by source (inside ndisc_send_redirect)
484 */
485 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
486 ndisc_send_redirect(skb, target);
487 } else {
488 int addrtype = ipv6_addr_type(&hdr->saddr);
489
490 /* This check is security critical. */
491 if (addrtype == IPV6_ADDR_ANY ||
492 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
493 goto error;
494 if (addrtype & IPV6_ADDR_LINKLOCAL) {
495 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
496 ICMPV6_NOT_NEIGHBOUR, 0);
497 goto error;
498 }
499 }
500
501 mtu = dst_mtu(dst);
502 if (mtu < IPV6_MIN_MTU)
503 mtu = IPV6_MIN_MTU;
504
505 if (skb->len > mtu && !skb_is_gso(skb)) {
506 /* Again, force OUTPUT device used as source address */
507 skb->dev = dst->dev;
508 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
509 IP6_INC_STATS_BH(net,
510 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
511 IP6_INC_STATS_BH(net,
512 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
513 kfree_skb(skb);
514 return -EMSGSIZE;
515 }
516
517 if (skb_cow(skb, dst->dev->hard_header_len)) {
518 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
519 goto drop;
520 }
521
522 hdr = ipv6_hdr(skb);
523
524 /* Mangling hops number delayed to point after skb COW */
525
526 hdr->hop_limit--;
527
528 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
529 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
530 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
531 ip6_forward_finish);
532
533error:
534 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
535drop:
536 kfree_skb(skb);
537 return -EINVAL;
538}
539
540static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
541{
542 to->pkt_type = from->pkt_type;
543 to->priority = from->priority;
544 to->protocol = from->protocol;
545 skb_dst_drop(to);
546 skb_dst_set(to, dst_clone(skb_dst(from)));
547 to->dev = from->dev;
548 to->mark = from->mark;
549
550#ifdef CONFIG_NET_SCHED
551 to->tc_index = from->tc_index;
552#endif
553 nf_copy(to, from);
554#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
555 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
556 to->nf_trace = from->nf_trace;
557#endif
558 skb_copy_secmark(to, from);
559}
560
561int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
562{
563 u16 offset = sizeof(struct ipv6hdr);
564 struct ipv6_opt_hdr *exthdr =
565 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
566 unsigned int packet_len = skb->tail - skb->network_header;
567 int found_rhdr = 0;
568 *nexthdr = &ipv6_hdr(skb)->nexthdr;
569
570 while (offset + 1 <= packet_len) {
571
572 switch (**nexthdr) {
573
574 case NEXTHDR_HOP:
575 break;
576 case NEXTHDR_ROUTING:
577 found_rhdr = 1;
578 break;
579 case NEXTHDR_DEST:
580#if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
581 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
582 break;
583#endif
584 if (found_rhdr)
585 return offset;
586 break;
587 default :
588 return offset;
589 }
590
591 offset += ipv6_optlen(exthdr);
592 *nexthdr = &exthdr->nexthdr;
593 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
594 offset);
595 }
596
597 return offset;
598}
599
600void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
601{
602 static atomic_t ipv6_fragmentation_id;
603 int old, new;
604
605 if (rt && !(rt->dst.flags & DST_NOPEER)) {
606 struct inet_peer *peer;
607
608 if (!rt->rt6i_peer)
609 rt6_bind_peer(rt, 1);
610 peer = rt->rt6i_peer;
611 if (peer) {
612 fhdr->identification = htonl(inet_getid(peer, 0));
613 return;
614 }
615 }
616 do {
617 old = atomic_read(&ipv6_fragmentation_id);
618 new = old + 1;
619 if (!new)
620 new = 1;
621 } while (atomic_cmpxchg(&ipv6_fragmentation_id, old, new) != old);
622 fhdr->identification = htonl(new);
623}
624
625int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
626{
627 struct sk_buff *frag;
628 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
629 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
630 struct ipv6hdr *tmp_hdr;
631 struct frag_hdr *fh;
632 unsigned int mtu, hlen, left, len;
633 int hroom, troom;
634 __be32 frag_id = 0;
635 int ptr, offset = 0, err=0;
636 u8 *prevhdr, nexthdr = 0;
637 struct net *net = dev_net(skb_dst(skb)->dev);
638
639 hlen = ip6_find_1stfragopt(skb, &prevhdr);
640 nexthdr = *prevhdr;
641
642 mtu = ip6_skb_dst_mtu(skb);
643
644 /* We must not fragment if the socket is set to force MTU discovery
645 * or if the skb it not generated by a local socket.
646 */
647 if (unlikely(!skb->local_df && skb->len > mtu)) {
648 if (skb->sk && dst_allfrag(skb_dst(skb)))
649 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
650
651 skb->dev = skb_dst(skb)->dev;
652 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
653 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
654 IPSTATS_MIB_FRAGFAILS);
655 kfree_skb(skb);
656 return -EMSGSIZE;
657 }
658
659 if (np && np->frag_size < mtu) {
660 if (np->frag_size)
661 mtu = np->frag_size;
662 }
663 mtu -= hlen + sizeof(struct frag_hdr);
664
665 if (skb_has_frag_list(skb)) {
666 int first_len = skb_pagelen(skb);
667 struct sk_buff *frag2;
668
669 if (first_len - hlen > mtu ||
670 ((first_len - hlen) & 7) ||
671 skb_cloned(skb))
672 goto slow_path;
673
674 skb_walk_frags(skb, frag) {
675 /* Correct geometry. */
676 if (frag->len > mtu ||
677 ((frag->len & 7) && frag->next) ||
678 skb_headroom(frag) < hlen)
679 goto slow_path_clean;
680
681 /* Partially cloned skb? */
682 if (skb_shared(frag))
683 goto slow_path_clean;
684
685 BUG_ON(frag->sk);
686 if (skb->sk) {
687 frag->sk = skb->sk;
688 frag->destructor = sock_wfree;
689 }
690 skb->truesize -= frag->truesize;
691 }
692
693 err = 0;
694 offset = 0;
695 frag = skb_shinfo(skb)->frag_list;
696 skb_frag_list_init(skb);
697 /* BUILD HEADER */
698
699 *prevhdr = NEXTHDR_FRAGMENT;
700 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
701 if (!tmp_hdr) {
702 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
703 IPSTATS_MIB_FRAGFAILS);
704 return -ENOMEM;
705 }
706
707 __skb_pull(skb, hlen);
708 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
709 __skb_push(skb, hlen);
710 skb_reset_network_header(skb);
711 memcpy(skb_network_header(skb), tmp_hdr, hlen);
712
713 ipv6_select_ident(fh, rt);
714 fh->nexthdr = nexthdr;
715 fh->reserved = 0;
716 fh->frag_off = htons(IP6_MF);
717 frag_id = fh->identification;
718
719 first_len = skb_pagelen(skb);
720 skb->data_len = first_len - skb_headlen(skb);
721 skb->len = first_len;
722 ipv6_hdr(skb)->payload_len = htons(first_len -
723 sizeof(struct ipv6hdr));
724
725 dst_hold(&rt->dst);
726
727 for (;;) {
728 /* Prepare header of the next frame,
729 * before previous one went down. */
730 if (frag) {
731 frag->ip_summed = CHECKSUM_NONE;
732 skb_reset_transport_header(frag);
733 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
734 __skb_push(frag, hlen);
735 skb_reset_network_header(frag);
736 memcpy(skb_network_header(frag), tmp_hdr,
737 hlen);
738 offset += skb->len - hlen - sizeof(struct frag_hdr);
739 fh->nexthdr = nexthdr;
740 fh->reserved = 0;
741 fh->frag_off = htons(offset);
742 if (frag->next != NULL)
743 fh->frag_off |= htons(IP6_MF);
744 fh->identification = frag_id;
745 ipv6_hdr(frag)->payload_len =
746 htons(frag->len -
747 sizeof(struct ipv6hdr));
748 ip6_copy_metadata(frag, skb);
749 }
750
751 err = output(skb);
752 if(!err)
753 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
754 IPSTATS_MIB_FRAGCREATES);
755
756 if (err || !frag)
757 break;
758
759 skb = frag;
760 frag = skb->next;
761 skb->next = NULL;
762 }
763
764 kfree(tmp_hdr);
765
766 if (err == 0) {
767 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
768 IPSTATS_MIB_FRAGOKS);
769 dst_release(&rt->dst);
770 return 0;
771 }
772
773 while (frag) {
774 skb = frag->next;
775 kfree_skb(frag);
776 frag = skb;
777 }
778
779 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
780 IPSTATS_MIB_FRAGFAILS);
781 dst_release(&rt->dst);
782 return err;
783
784slow_path_clean:
785 skb_walk_frags(skb, frag2) {
786 if (frag2 == frag)
787 break;
788 frag2->sk = NULL;
789 frag2->destructor = NULL;
790 skb->truesize += frag2->truesize;
791 }
792 }
793
794slow_path:
795 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
796 skb_checksum_help(skb))
797 goto fail;
798
799 left = skb->len - hlen; /* Space per frame */
800 ptr = hlen; /* Where to start from */
801
802 /*
803 * Fragment the datagram.
804 */
805
806 *prevhdr = NEXTHDR_FRAGMENT;
807 hroom = LL_RESERVED_SPACE(rt->dst.dev);
808 troom = rt->dst.dev->needed_tailroom;
809
810 /*
811 * Keep copying data until we run out.
812 */
813 while(left > 0) {
814 len = left;
815 /* IF: it doesn't fit, use 'mtu' - the data space left */
816 if (len > mtu)
817 len = mtu;
818 /* IF: we are not sending up to and including the packet end
819 then align the next start on an eight byte boundary */
820 if (len < left) {
821 len &= ~7;
822 }
823 /*
824 * Allocate buffer.
825 */
826
827 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
828 hroom + troom, GFP_ATOMIC)) == NULL) {
829 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
830 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
831 IPSTATS_MIB_FRAGFAILS);
832 err = -ENOMEM;
833 goto fail;
834 }
835
836 /*
837 * Set up data on packet
838 */
839
840 ip6_copy_metadata(frag, skb);
841 skb_reserve(frag, hroom);
842 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
843 skb_reset_network_header(frag);
844 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
845 frag->transport_header = (frag->network_header + hlen +
846 sizeof(struct frag_hdr));
847
848 /*
849 * Charge the memory for the fragment to any owner
850 * it might possess
851 */
852 if (skb->sk)
853 skb_set_owner_w(frag, skb->sk);
854
855 /*
856 * Copy the packet header into the new buffer.
857 */
858 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
859
860 /*
861 * Build fragment header.
862 */
863 fh->nexthdr = nexthdr;
864 fh->reserved = 0;
865 if (!frag_id) {
866 ipv6_select_ident(fh, rt);
867 frag_id = fh->identification;
868 } else
869 fh->identification = frag_id;
870
871 /*
872 * Copy a block of the IP datagram.
873 */
874 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
875 BUG();
876 left -= len;
877
878 fh->frag_off = htons(offset);
879 if (left > 0)
880 fh->frag_off |= htons(IP6_MF);
881 ipv6_hdr(frag)->payload_len = htons(frag->len -
882 sizeof(struct ipv6hdr));
883
884 ptr += len;
885 offset += len;
886
887 /*
888 * Put this fragment into the sending queue.
889 */
890 err = output(frag);
891 if (err)
892 goto fail;
893
894 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
895 IPSTATS_MIB_FRAGCREATES);
896 }
897 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
898 IPSTATS_MIB_FRAGOKS);
899 consume_skb(skb);
900 return err;
901
902fail:
903 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
904 IPSTATS_MIB_FRAGFAILS);
905 kfree_skb(skb);
906 return err;
907}
908
909static inline int ip6_rt_check(const struct rt6key *rt_key,
910 const struct in6_addr *fl_addr,
911 const struct in6_addr *addr_cache)
912{
913 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
914 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
915}
916
917static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
918 struct dst_entry *dst,
919 const struct flowi6 *fl6)
920{
921 struct ipv6_pinfo *np = inet6_sk(sk);
922 struct rt6_info *rt = (struct rt6_info *)dst;
923
924 if (!dst)
925 goto out;
926
927 /* Yes, checking route validity in not connected
928 * case is not very simple. Take into account,
929 * that we do not support routing by source, TOS,
930 * and MSG_DONTROUTE --ANK (980726)
931 *
932 * 1. ip6_rt_check(): If route was host route,
933 * check that cached destination is current.
934 * If it is network route, we still may
935 * check its validity using saved pointer
936 * to the last used address: daddr_cache.
937 * We do not want to save whole address now,
938 * (because main consumer of this service
939 * is tcp, which has not this problem),
940 * so that the last trick works only on connected
941 * sockets.
942 * 2. oif also should be the same.
943 */
944 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
945#ifdef CONFIG_IPV6_SUBTREES
946 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
947#endif
948 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
949 dst_release(dst);
950 dst = NULL;
951 }
952
953out:
954 return dst;
955}
956
957static int ip6_dst_lookup_tail(struct sock *sk,
958 struct dst_entry **dst, struct flowi6 *fl6)
959{
960 struct net *net = sock_net(sk);
961#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
962 struct neighbour *n;
963#endif
964 int err;
965
966 if (*dst == NULL)
967 *dst = ip6_route_output(net, sk, fl6);
968
969 if ((err = (*dst)->error))
970 goto out_err_release;
971
972 if (ipv6_addr_any(&fl6->saddr)) {
973 struct rt6_info *rt = (struct rt6_info *) *dst;
974 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
975 sk ? inet6_sk(sk)->srcprefs : 0,
976 &fl6->saddr);
977 if (err)
978 goto out_err_release;
979 }
980
981#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
982 /*
983 * Here if the dst entry we've looked up
984 * has a neighbour entry that is in the INCOMPLETE
985 * state and the src address from the flow is
986 * marked as OPTIMISTIC, we release the found
987 * dst entry and replace it instead with the
988 * dst entry of the nexthop router
989 */
990 rcu_read_lock();
991 n = dst_get_neighbour_noref(*dst);
992 if (n && !(n->nud_state & NUD_VALID)) {
993 struct inet6_ifaddr *ifp;
994 struct flowi6 fl_gw6;
995 int redirect;
996
997 rcu_read_unlock();
998 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
999 (*dst)->dev, 1);
1000
1001 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1002 if (ifp)
1003 in6_ifa_put(ifp);
1004
1005 if (redirect) {
1006 /*
1007 * We need to get the dst entry for the
1008 * default router instead
1009 */
1010 dst_release(*dst);
1011 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1012 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1013 *dst = ip6_route_output(net, sk, &fl_gw6);
1014 if ((err = (*dst)->error))
1015 goto out_err_release;
1016 }
1017 } else {
1018 rcu_read_unlock();
1019 }
1020#endif
1021
1022 return 0;
1023
1024out_err_release:
1025 if (err == -ENETUNREACH)
1026 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1027 dst_release(*dst);
1028 *dst = NULL;
1029 return err;
1030}
1031
1032/**
1033 * ip6_dst_lookup - perform route lookup on flow
1034 * @sk: socket which provides route info
1035 * @dst: pointer to dst_entry * for result
1036 * @fl6: flow to lookup
1037 *
1038 * This function performs a route lookup on the given flow.
1039 *
1040 * It returns zero on success, or a standard errno code on error.
1041 */
1042int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1043{
1044 *dst = NULL;
1045 return ip6_dst_lookup_tail(sk, dst, fl6);
1046}
1047EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1048
1049/**
1050 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1051 * @sk: socket which provides route info
1052 * @fl6: flow to lookup
1053 * @final_dst: final destination address for ipsec lookup
1054 * @can_sleep: we are in a sleepable context
1055 *
1056 * This function performs a route lookup on the given flow.
1057 *
1058 * It returns a valid dst pointer on success, or a pointer encoded
1059 * error code.
1060 */
1061struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1062 const struct in6_addr *final_dst,
1063 bool can_sleep)
1064{
1065 struct dst_entry *dst = NULL;
1066 int err;
1067
1068 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1069 if (err)
1070 return ERR_PTR(err);
1071 if (final_dst)
1072 fl6->daddr = *final_dst;
1073 if (can_sleep)
1074 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1075
1076 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1077}
1078EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1079
1080/**
1081 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1082 * @sk: socket which provides the dst cache and route info
1083 * @fl6: flow to lookup
1084 * @final_dst: final destination address for ipsec lookup
1085 * @can_sleep: we are in a sleepable context
1086 *
1087 * This function performs a route lookup on the given flow with the
1088 * possibility of using the cached route in the socket if it is valid.
1089 * It will take the socket dst lock when operating on the dst cache.
1090 * As a result, this function can only be used in process context.
1091 *
1092 * It returns a valid dst pointer on success, or a pointer encoded
1093 * error code.
1094 */
1095struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1096 const struct in6_addr *final_dst,
1097 bool can_sleep)
1098{
1099 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1100 int err;
1101
1102 dst = ip6_sk_dst_check(sk, dst, fl6);
1103
1104 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1105 if (err)
1106 return ERR_PTR(err);
1107 if (final_dst)
1108 fl6->daddr = *final_dst;
1109 if (can_sleep)
1110 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1111
1112 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1113}
1114EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1115
1116static inline int ip6_ufo_append_data(struct sock *sk,
1117 int getfrag(void *from, char *to, int offset, int len,
1118 int odd, struct sk_buff *skb),
1119 void *from, int length, int hh_len, int fragheaderlen,
1120 int transhdrlen, int mtu,unsigned int flags,
1121 struct rt6_info *rt)
1122
1123{
1124 struct sk_buff *skb;
1125 int err;
1126
1127 /* There is support for UDP large send offload by network
1128 * device, so create one single skb packet containing complete
1129 * udp datagram
1130 */
1131 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1132 skb = sock_alloc_send_skb(sk,
1133 hh_len + fragheaderlen + transhdrlen + 20,
1134 (flags & MSG_DONTWAIT), &err);
1135 if (skb == NULL)
1136 return err;
1137
1138 /* reserve space for Hardware header */
1139 skb_reserve(skb, hh_len);
1140
1141 /* create space for UDP/IP header */
1142 skb_put(skb,fragheaderlen + transhdrlen);
1143
1144 /* initialize network header pointer */
1145 skb_reset_network_header(skb);
1146
1147 /* initialize protocol header pointer */
1148 skb->transport_header = skb->network_header + fragheaderlen;
1149
1150 skb->ip_summed = CHECKSUM_PARTIAL;
1151 skb->csum = 0;
1152 }
1153
1154 err = skb_append_datato_frags(sk,skb, getfrag, from,
1155 (length - transhdrlen));
1156 if (!err) {
1157 struct frag_hdr fhdr;
1158
1159 /* Specify the length of each IPv6 datagram fragment.
1160 * It has to be a multiple of 8.
1161 */
1162 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1163 sizeof(struct frag_hdr)) & ~7;
1164 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1165 ipv6_select_ident(&fhdr, rt);
1166 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1167 __skb_queue_tail(&sk->sk_write_queue, skb);
1168
1169 return 0;
1170 }
1171 /* There is not enough support do UPD LSO,
1172 * so follow normal path
1173 */
1174 kfree_skb(skb);
1175
1176 return err;
1177}
1178
1179static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1180 gfp_t gfp)
1181{
1182 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1183}
1184
1185static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1186 gfp_t gfp)
1187{
1188 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1189}
1190
1191static void ip6_append_data_mtu(int *mtu,
1192 int *maxfraglen,
1193 unsigned int fragheaderlen,
1194 struct sk_buff *skb,
1195 struct rt6_info *rt)
1196{
1197 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1198 if (skb == NULL) {
1199 /* first fragment, reserve header_len */
1200 *mtu = *mtu - rt->dst.header_len;
1201
1202 } else {
1203 /*
1204 * this fragment is not first, the headers
1205 * space is regarded as data space.
1206 */
1207 *mtu = dst_mtu(rt->dst.path);
1208 }
1209 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1210 + fragheaderlen - sizeof(struct frag_hdr);
1211 }
1212}
1213
1214int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1215 int offset, int len, int odd, struct sk_buff *skb),
1216 void *from, int length, int transhdrlen,
1217 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1218 struct rt6_info *rt, unsigned int flags, int dontfrag)
1219{
1220 struct inet_sock *inet = inet_sk(sk);
1221 struct ipv6_pinfo *np = inet6_sk(sk);
1222 struct inet_cork *cork;
1223 struct sk_buff *skb, *skb_prev = NULL;
1224 unsigned int maxfraglen, fragheaderlen;
1225 int exthdrlen;
1226 int dst_exthdrlen;
1227 int hh_len;
1228 int mtu;
1229 int copy;
1230 int err;
1231 int offset = 0;
1232 __u8 tx_flags = 0;
1233
1234 if (flags&MSG_PROBE)
1235 return 0;
1236 cork = &inet->cork.base;
1237 if (skb_queue_empty(&sk->sk_write_queue)) {
1238 /*
1239 * setup for corking
1240 */
1241 if (opt) {
1242 if (WARN_ON(np->cork.opt))
1243 return -EINVAL;
1244
1245 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1246 if (unlikely(np->cork.opt == NULL))
1247 return -ENOBUFS;
1248
1249 np->cork.opt->tot_len = opt->tot_len;
1250 np->cork.opt->opt_flen = opt->opt_flen;
1251 np->cork.opt->opt_nflen = opt->opt_nflen;
1252
1253 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1254 sk->sk_allocation);
1255 if (opt->dst0opt && !np->cork.opt->dst0opt)
1256 return -ENOBUFS;
1257
1258 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1259 sk->sk_allocation);
1260 if (opt->dst1opt && !np->cork.opt->dst1opt)
1261 return -ENOBUFS;
1262
1263 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1264 sk->sk_allocation);
1265 if (opt->hopopt && !np->cork.opt->hopopt)
1266 return -ENOBUFS;
1267
1268 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1269 sk->sk_allocation);
1270 if (opt->srcrt && !np->cork.opt->srcrt)
1271 return -ENOBUFS;
1272
1273 /* need source address above miyazawa*/
1274 }
1275 dst_hold(&rt->dst);
1276 cork->dst = &rt->dst;
1277 inet->cork.fl.u.ip6 = *fl6;
1278 np->cork.hop_limit = hlimit;
1279 np->cork.tclass = tclass;
1280 if (rt->dst.flags & DST_XFRM_TUNNEL)
1281 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1282 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1283 else
1284 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1285 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1286 if (np->frag_size < mtu) {
1287 if (np->frag_size)
1288 mtu = np->frag_size;
1289 }
1290 cork->fragsize = mtu;
1291 if (dst_allfrag(rt->dst.path))
1292 cork->flags |= IPCORK_ALLFRAG;
1293 cork->length = 0;
1294 sk->sk_sndmsg_page = NULL;
1295 sk->sk_sndmsg_off = 0;
1296 exthdrlen = (opt ? opt->opt_flen : 0) - rt->rt6i_nfheader_len;
1297 length += exthdrlen;
1298 transhdrlen += exthdrlen;
1299 dst_exthdrlen = rt->dst.header_len;
1300 } else {
1301 rt = (struct rt6_info *)cork->dst;
1302 fl6 = &inet->cork.fl.u.ip6;
1303 opt = np->cork.opt;
1304 transhdrlen = 0;
1305 exthdrlen = 0;
1306 dst_exthdrlen = 0;
1307 mtu = cork->fragsize;
1308 }
1309
1310 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1311
1312 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1313 (opt ? opt->opt_nflen : 0);
1314 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1315
1316 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1317 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1318 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1319 return -EMSGSIZE;
1320 }
1321 }
1322
1323 /* For UDP, check if TX timestamp is enabled */
1324 if (sk->sk_type == SOCK_DGRAM) {
1325 err = sock_tx_timestamp(sk, &tx_flags);
1326 if (err)
1327 goto error;
1328 }
1329
1330 /*
1331 * Let's try using as much space as possible.
1332 * Use MTU if total length of the message fits into the MTU.
1333 * Otherwise, we need to reserve fragment header and
1334 * fragment alignment (= 8-15 octects, in total).
1335 *
1336 * Note that we may need to "move" the data from the tail of
1337 * of the buffer to the new fragment when we split
1338 * the message.
1339 *
1340 * FIXME: It may be fragmented into multiple chunks
1341 * at once if non-fragmentable extension headers
1342 * are too large.
1343 * --yoshfuji
1344 */
1345
1346 cork->length += length;
1347 if (length > mtu) {
1348 int proto = sk->sk_protocol;
1349 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1350 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1351 return -EMSGSIZE;
1352 }
1353
1354 if (proto == IPPROTO_UDP &&
1355 (rt->dst.dev->features & NETIF_F_UFO)) {
1356
1357 err = ip6_ufo_append_data(sk, getfrag, from, length,
1358 hh_len, fragheaderlen,
1359 transhdrlen, mtu, flags, rt);
1360 if (err)
1361 goto error;
1362 return 0;
1363 }
1364 }
1365
1366 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1367 goto alloc_new_skb;
1368
1369 while (length > 0) {
1370 /* Check if the remaining data fits into current packet. */
1371 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1372 if (copy < length)
1373 copy = maxfraglen - skb->len;
1374
1375 if (copy <= 0) {
1376 char *data;
1377 unsigned int datalen;
1378 unsigned int fraglen;
1379 unsigned int fraggap;
1380 unsigned int alloclen;
1381alloc_new_skb:
1382 /* There's no room in the current skb */
1383 if (skb)
1384 fraggap = skb->len - maxfraglen;
1385 else
1386 fraggap = 0;
1387 /* update mtu and maxfraglen if necessary */
1388 if (skb == NULL || skb_prev == NULL)
1389 ip6_append_data_mtu(&mtu, &maxfraglen,
1390 fragheaderlen, skb, rt);
1391
1392 skb_prev = skb;
1393
1394 /*
1395 * If remaining data exceeds the mtu,
1396 * we know we need more fragment(s).
1397 */
1398 datalen = length + fraggap;
1399
1400 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1401 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1402 if ((flags & MSG_MORE) &&
1403 !(rt->dst.dev->features&NETIF_F_SG))
1404 alloclen = mtu;
1405 else
1406 alloclen = datalen + fragheaderlen;
1407
1408 alloclen += dst_exthdrlen;
1409
1410 if (datalen != length + fraggap) {
1411 /*
1412 * this is not the last fragment, the trailer
1413 * space is regarded as data space.
1414 */
1415 datalen += rt->dst.trailer_len;
1416 }
1417
1418 alloclen += rt->dst.trailer_len;
1419 fraglen = datalen + fragheaderlen;
1420
1421 /*
1422 * We just reserve space for fragment header.
1423 * Note: this may be overallocation if the message
1424 * (without MSG_MORE) fits into the MTU.
1425 */
1426 alloclen += sizeof(struct frag_hdr);
1427
1428 if (transhdrlen) {
1429 skb = sock_alloc_send_skb(sk,
1430 alloclen + hh_len,
1431 (flags & MSG_DONTWAIT), &err);
1432 } else {
1433 skb = NULL;
1434 if (atomic_read(&sk->sk_wmem_alloc) <=
1435 2 * sk->sk_sndbuf)
1436 skb = sock_wmalloc(sk,
1437 alloclen + hh_len, 1,
1438 sk->sk_allocation);
1439 if (unlikely(skb == NULL))
1440 err = -ENOBUFS;
1441 else {
1442 /* Only the initial fragment
1443 * is time stamped.
1444 */
1445 tx_flags = 0;
1446 }
1447 }
1448 if (skb == NULL)
1449 goto error;
1450 /*
1451 * Fill in the control structures
1452 */
1453 skb->ip_summed = CHECKSUM_NONE;
1454 skb->csum = 0;
1455 /* reserve for fragmentation and ipsec header */
1456 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1457 dst_exthdrlen);
1458
1459 if (sk->sk_type == SOCK_DGRAM)
1460 skb_shinfo(skb)->tx_flags = tx_flags;
1461
1462 /*
1463 * Find where to start putting bytes
1464 */
1465 data = skb_put(skb, fraglen);
1466 skb_set_network_header(skb, exthdrlen);
1467 data += fragheaderlen;
1468 skb->transport_header = (skb->network_header +
1469 fragheaderlen);
1470 if (fraggap) {
1471 skb->csum = skb_copy_and_csum_bits(
1472 skb_prev, maxfraglen,
1473 data + transhdrlen, fraggap, 0);
1474 skb_prev->csum = csum_sub(skb_prev->csum,
1475 skb->csum);
1476 data += fraggap;
1477 pskb_trim_unique(skb_prev, maxfraglen);
1478 }
1479 copy = datalen - transhdrlen - fraggap;
1480
1481 if (copy < 0) {
1482 err = -EINVAL;
1483 kfree_skb(skb);
1484 goto error;
1485 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1486 err = -EFAULT;
1487 kfree_skb(skb);
1488 goto error;
1489 }
1490
1491 offset += copy;
1492 length -= datalen - fraggap;
1493 transhdrlen = 0;
1494 exthdrlen = 0;
1495 dst_exthdrlen = 0;
1496
1497 /*
1498 * Put the packet on the pending queue
1499 */
1500 __skb_queue_tail(&sk->sk_write_queue, skb);
1501 continue;
1502 }
1503
1504 if (copy > length)
1505 copy = length;
1506
1507 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1508 unsigned int off;
1509
1510 off = skb->len;
1511 if (getfrag(from, skb_put(skb, copy),
1512 offset, copy, off, skb) < 0) {
1513 __skb_trim(skb, off);
1514 err = -EFAULT;
1515 goto error;
1516 }
1517 } else {
1518 int i = skb_shinfo(skb)->nr_frags;
1519 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1520 struct page *page = sk->sk_sndmsg_page;
1521 int off = sk->sk_sndmsg_off;
1522 unsigned int left;
1523
1524 if (page && (left = PAGE_SIZE - off) > 0) {
1525 if (copy >= left)
1526 copy = left;
1527 if (page != skb_frag_page(frag)) {
1528 if (i == MAX_SKB_FRAGS) {
1529 err = -EMSGSIZE;
1530 goto error;
1531 }
1532 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1533 skb_frag_ref(skb, i);
1534 frag = &skb_shinfo(skb)->frags[i];
1535 }
1536 } else if(i < MAX_SKB_FRAGS) {
1537 if (copy > PAGE_SIZE)
1538 copy = PAGE_SIZE;
1539 page = alloc_pages(sk->sk_allocation, 0);
1540 if (page == NULL) {
1541 err = -ENOMEM;
1542 goto error;
1543 }
1544 sk->sk_sndmsg_page = page;
1545 sk->sk_sndmsg_off = 0;
1546
1547 skb_fill_page_desc(skb, i, page, 0, 0);
1548 frag = &skb_shinfo(skb)->frags[i];
1549 } else {
1550 err = -EMSGSIZE;
1551 goto error;
1552 }
1553 if (getfrag(from,
1554 skb_frag_address(frag) + skb_frag_size(frag),
1555 offset, copy, skb->len, skb) < 0) {
1556 err = -EFAULT;
1557 goto error;
1558 }
1559 sk->sk_sndmsg_off += copy;
1560 skb_frag_size_add(frag, copy);
1561 skb->len += copy;
1562 skb->data_len += copy;
1563 skb->truesize += copy;
1564 atomic_add(copy, &sk->sk_wmem_alloc);
1565 }
1566 offset += copy;
1567 length -= copy;
1568 }
1569 return 0;
1570error:
1571 cork->length -= length;
1572 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1573 return err;
1574}
1575EXPORT_SYMBOL_GPL(ip6_append_data);
1576
1577static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1578{
1579 if (np->cork.opt) {
1580 kfree(np->cork.opt->dst0opt);
1581 kfree(np->cork.opt->dst1opt);
1582 kfree(np->cork.opt->hopopt);
1583 kfree(np->cork.opt->srcrt);
1584 kfree(np->cork.opt);
1585 np->cork.opt = NULL;
1586 }
1587
1588 if (inet->cork.base.dst) {
1589 dst_release(inet->cork.base.dst);
1590 inet->cork.base.dst = NULL;
1591 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1592 }
1593 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1594}
1595
1596int ip6_push_pending_frames(struct sock *sk)
1597{
1598 struct sk_buff *skb, *tmp_skb;
1599 struct sk_buff **tail_skb;
1600 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1601 struct inet_sock *inet = inet_sk(sk);
1602 struct ipv6_pinfo *np = inet6_sk(sk);
1603 struct net *net = sock_net(sk);
1604 struct ipv6hdr *hdr;
1605 struct ipv6_txoptions *opt = np->cork.opt;
1606 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1607 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1608 unsigned char proto = fl6->flowi6_proto;
1609 int err = 0;
1610
1611 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1612 goto out;
1613 tail_skb = &(skb_shinfo(skb)->frag_list);
1614
1615 /* move skb->data to ip header from ext header */
1616 if (skb->data < skb_network_header(skb))
1617 __skb_pull(skb, skb_network_offset(skb));
1618 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1619 __skb_pull(tmp_skb, skb_network_header_len(skb));
1620 *tail_skb = tmp_skb;
1621 tail_skb = &(tmp_skb->next);
1622 skb->len += tmp_skb->len;
1623 skb->data_len += tmp_skb->len;
1624 skb->truesize += tmp_skb->truesize;
1625 tmp_skb->destructor = NULL;
1626 tmp_skb->sk = NULL;
1627 }
1628
1629 /* Allow local fragmentation. */
1630 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1631 skb->local_df = 1;
1632
1633 *final_dst = fl6->daddr;
1634 __skb_pull(skb, skb_network_header_len(skb));
1635 if (opt && opt->opt_flen)
1636 ipv6_push_frag_opts(skb, opt, &proto);
1637 if (opt && opt->opt_nflen)
1638 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1639
1640 skb_push(skb, sizeof(struct ipv6hdr));
1641 skb_reset_network_header(skb);
1642 hdr = ipv6_hdr(skb);
1643
1644 *(__be32*)hdr = fl6->flowlabel |
1645 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1646
1647 hdr->hop_limit = np->cork.hop_limit;
1648 hdr->nexthdr = proto;
1649 hdr->saddr = fl6->saddr;
1650 hdr->daddr = *final_dst;
1651
1652 skb->priority = sk->sk_priority;
1653 skb->mark = sk->sk_mark;
1654
1655 skb_dst_set(skb, dst_clone(&rt->dst));
1656 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1657 if (proto == IPPROTO_ICMPV6) {
1658 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1659
1660 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1661 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1662 }
1663
1664 err = ip6_local_out(skb);
1665 if (err) {
1666 if (err > 0)
1667 err = net_xmit_errno(err);
1668 if (err)
1669 goto error;
1670 }
1671
1672out:
1673 ip6_cork_release(inet, np);
1674 return err;
1675error:
1676 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1677 goto out;
1678}
1679EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1680
1681void ip6_flush_pending_frames(struct sock *sk)
1682{
1683 struct sk_buff *skb;
1684
1685 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1686 if (skb_dst(skb))
1687 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1688 IPSTATS_MIB_OUTDISCARDS);
1689 kfree_skb(skb);
1690 }
1691
1692 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1693}
1694EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);