Loading...
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(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)
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 int tclass = 0;
194 u32 mtu;
195
196 if (opt) {
197 unsigned int head_room;
198
199 /* First: exthdrs may take lots of space (~8K for now)
200 MAX_HEADER is not enough.
201 */
202 head_room = opt->opt_nflen + opt->opt_flen;
203 seg_len += head_room;
204 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
205
206 if (skb_headroom(skb) < head_room) {
207 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
208 if (skb2 == NULL) {
209 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
210 IPSTATS_MIB_OUTDISCARDS);
211 kfree_skb(skb);
212 return -ENOBUFS;
213 }
214 kfree_skb(skb);
215 skb = skb2;
216 skb_set_owner_w(skb, sk);
217 }
218 if (opt->opt_flen)
219 ipv6_push_frag_opts(skb, opt, &proto);
220 if (opt->opt_nflen)
221 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
222 }
223
224 skb_push(skb, sizeof(struct ipv6hdr));
225 skb_reset_network_header(skb);
226 hdr = ipv6_hdr(skb);
227
228 /*
229 * Fill in the IPv6 header
230 */
231 if (np) {
232 tclass = np->tclass;
233 hlimit = np->hop_limit;
234 }
235 if (hlimit < 0)
236 hlimit = ip6_dst_hoplimit(dst);
237
238 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl6->flowlabel;
239
240 hdr->payload_len = htons(seg_len);
241 hdr->nexthdr = proto;
242 hdr->hop_limit = hlimit;
243
244 ipv6_addr_copy(&hdr->saddr, &fl6->saddr);
245 ipv6_addr_copy(&hdr->daddr, first_hop);
246
247 skb->priority = sk->sk_priority;
248 skb->mark = sk->sk_mark;
249
250 mtu = dst_mtu(dst);
251 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
252 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
253 IPSTATS_MIB_OUT, skb->len);
254 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
255 dst->dev, dst_output);
256 }
257
258 if (net_ratelimit())
259 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
260 skb->dev = dst->dev;
261 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
262 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
263 kfree_skb(skb);
264 return -EMSGSIZE;
265}
266
267EXPORT_SYMBOL(ip6_xmit);
268
269/*
270 * To avoid extra problems ND packets are send through this
271 * routine. It's code duplication but I really want to avoid
272 * extra checks since ipv6_build_header is used by TCP (which
273 * is for us performance critical)
274 */
275
276int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
277 const struct in6_addr *saddr, const struct in6_addr *daddr,
278 int proto, int len)
279{
280 struct ipv6_pinfo *np = inet6_sk(sk);
281 struct ipv6hdr *hdr;
282
283 skb->protocol = htons(ETH_P_IPV6);
284 skb->dev = dev;
285
286 skb_reset_network_header(skb);
287 skb_put(skb, sizeof(struct ipv6hdr));
288 hdr = ipv6_hdr(skb);
289
290 *(__be32*)hdr = htonl(0x60000000);
291
292 hdr->payload_len = htons(len);
293 hdr->nexthdr = proto;
294 hdr->hop_limit = np->hop_limit;
295
296 ipv6_addr_copy(&hdr->saddr, saddr);
297 ipv6_addr_copy(&hdr->daddr, daddr);
298
299 return 0;
300}
301
302static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
303{
304 struct ip6_ra_chain *ra;
305 struct sock *last = NULL;
306
307 read_lock(&ip6_ra_lock);
308 for (ra = ip6_ra_chain; ra; ra = ra->next) {
309 struct sock *sk = ra->sk;
310 if (sk && ra->sel == sel &&
311 (!sk->sk_bound_dev_if ||
312 sk->sk_bound_dev_if == skb->dev->ifindex)) {
313 if (last) {
314 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
315 if (skb2)
316 rawv6_rcv(last, skb2);
317 }
318 last = sk;
319 }
320 }
321
322 if (last) {
323 rawv6_rcv(last, skb);
324 read_unlock(&ip6_ra_lock);
325 return 1;
326 }
327 read_unlock(&ip6_ra_lock);
328 return 0;
329}
330
331static int ip6_forward_proxy_check(struct sk_buff *skb)
332{
333 struct ipv6hdr *hdr = ipv6_hdr(skb);
334 u8 nexthdr = hdr->nexthdr;
335 int offset;
336
337 if (ipv6_ext_hdr(nexthdr)) {
338 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
339 if (offset < 0)
340 return 0;
341 } else
342 offset = sizeof(struct ipv6hdr);
343
344 if (nexthdr == IPPROTO_ICMPV6) {
345 struct icmp6hdr *icmp6;
346
347 if (!pskb_may_pull(skb, (skb_network_header(skb) +
348 offset + 1 - skb->data)))
349 return 0;
350
351 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
352
353 switch (icmp6->icmp6_type) {
354 case NDISC_ROUTER_SOLICITATION:
355 case NDISC_ROUTER_ADVERTISEMENT:
356 case NDISC_NEIGHBOUR_SOLICITATION:
357 case NDISC_NEIGHBOUR_ADVERTISEMENT:
358 case NDISC_REDIRECT:
359 /* For reaction involving unicast neighbor discovery
360 * message destined to the proxied address, pass it to
361 * input function.
362 */
363 return 1;
364 default:
365 break;
366 }
367 }
368
369 /*
370 * The proxying router can't forward traffic sent to a link-local
371 * address, so signal the sender and discard the packet. This
372 * behavior is clarified by the MIPv6 specification.
373 */
374 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
375 dst_link_failure(skb);
376 return -1;
377 }
378
379 return 0;
380}
381
382static inline int ip6_forward_finish(struct sk_buff *skb)
383{
384 return dst_output(skb);
385}
386
387int ip6_forward(struct sk_buff *skb)
388{
389 struct dst_entry *dst = skb_dst(skb);
390 struct ipv6hdr *hdr = ipv6_hdr(skb);
391 struct inet6_skb_parm *opt = IP6CB(skb);
392 struct net *net = dev_net(dst->dev);
393 struct neighbour *n;
394 u32 mtu;
395
396 if (net->ipv6.devconf_all->forwarding == 0)
397 goto error;
398
399 if (skb_warn_if_lro(skb))
400 goto drop;
401
402 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
403 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
404 goto drop;
405 }
406
407 if (skb->pkt_type != PACKET_HOST)
408 goto drop;
409
410 skb_forward_csum(skb);
411
412 /*
413 * We DO NOT make any processing on
414 * RA packets, pushing them to user level AS IS
415 * without ane WARRANTY that application will be able
416 * to interpret them. The reason is that we
417 * cannot make anything clever here.
418 *
419 * We are not end-node, so that if packet contains
420 * AH/ESP, we cannot make anything.
421 * Defragmentation also would be mistake, RA packets
422 * cannot be fragmented, because there is no warranty
423 * that different fragments will go along one path. --ANK
424 */
425 if (opt->ra) {
426 u8 *ptr = skb_network_header(skb) + opt->ra;
427 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
428 return 0;
429 }
430
431 /*
432 * check and decrement ttl
433 */
434 if (hdr->hop_limit <= 1) {
435 /* Force OUTPUT device used as source address */
436 skb->dev = dst->dev;
437 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
438 IP6_INC_STATS_BH(net,
439 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
440
441 kfree_skb(skb);
442 return -ETIMEDOUT;
443 }
444
445 /* XXX: idev->cnf.proxy_ndp? */
446 if (net->ipv6.devconf_all->proxy_ndp &&
447 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
448 int proxied = ip6_forward_proxy_check(skb);
449 if (proxied > 0)
450 return ip6_input(skb);
451 else if (proxied < 0) {
452 IP6_INC_STATS(net, ip6_dst_idev(dst),
453 IPSTATS_MIB_INDISCARDS);
454 goto drop;
455 }
456 }
457
458 if (!xfrm6_route_forward(skb)) {
459 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
460 goto drop;
461 }
462 dst = skb_dst(skb);
463
464 /* IPv6 specs say nothing about it, but it is clear that we cannot
465 send redirects to source routed frames.
466 We don't send redirects to frames decapsulated from IPsec.
467 */
468 n = dst_get_neighbour(dst);
469 if (skb->dev == dst->dev && n && opt->srcrt == 0 && !skb_sec_path(skb)) {
470 struct in6_addr *target = NULL;
471 struct rt6_info *rt;
472
473 /*
474 * incoming and outgoing devices are the same
475 * send a redirect.
476 */
477
478 rt = (struct rt6_info *) dst;
479 if ((rt->rt6i_flags & RTF_GATEWAY))
480 target = (struct in6_addr*)&n->primary_key;
481 else
482 target = &hdr->daddr;
483
484 if (!rt->rt6i_peer)
485 rt6_bind_peer(rt, 1);
486
487 /* Limit redirects both by destination (here)
488 and by source (inside ndisc_send_redirect)
489 */
490 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
491 ndisc_send_redirect(skb, n, target);
492 } else {
493 int addrtype = ipv6_addr_type(&hdr->saddr);
494
495 /* This check is security critical. */
496 if (addrtype == IPV6_ADDR_ANY ||
497 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
498 goto error;
499 if (addrtype & IPV6_ADDR_LINKLOCAL) {
500 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
501 ICMPV6_NOT_NEIGHBOUR, 0);
502 goto error;
503 }
504 }
505
506 mtu = dst_mtu(dst);
507 if (mtu < IPV6_MIN_MTU)
508 mtu = IPV6_MIN_MTU;
509
510 if (skb->len > mtu && !skb_is_gso(skb)) {
511 /* Again, force OUTPUT device used as source address */
512 skb->dev = dst->dev;
513 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
514 IP6_INC_STATS_BH(net,
515 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
516 IP6_INC_STATS_BH(net,
517 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
518 kfree_skb(skb);
519 return -EMSGSIZE;
520 }
521
522 if (skb_cow(skb, dst->dev->hard_header_len)) {
523 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
524 goto drop;
525 }
526
527 hdr = ipv6_hdr(skb);
528
529 /* Mangling hops number delayed to point after skb COW */
530
531 hdr->hop_limit--;
532
533 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
534 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
535 ip6_forward_finish);
536
537error:
538 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
539drop:
540 kfree_skb(skb);
541 return -EINVAL;
542}
543
544static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
545{
546 to->pkt_type = from->pkt_type;
547 to->priority = from->priority;
548 to->protocol = from->protocol;
549 skb_dst_drop(to);
550 skb_dst_set(to, dst_clone(skb_dst(from)));
551 to->dev = from->dev;
552 to->mark = from->mark;
553
554#ifdef CONFIG_NET_SCHED
555 to->tc_index = from->tc_index;
556#endif
557 nf_copy(to, from);
558#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
559 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
560 to->nf_trace = from->nf_trace;
561#endif
562 skb_copy_secmark(to, from);
563}
564
565int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
566{
567 u16 offset = sizeof(struct ipv6hdr);
568 struct ipv6_opt_hdr *exthdr =
569 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
570 unsigned int packet_len = skb->tail - skb->network_header;
571 int found_rhdr = 0;
572 *nexthdr = &ipv6_hdr(skb)->nexthdr;
573
574 while (offset + 1 <= packet_len) {
575
576 switch (**nexthdr) {
577
578 case NEXTHDR_HOP:
579 break;
580 case NEXTHDR_ROUTING:
581 found_rhdr = 1;
582 break;
583 case NEXTHDR_DEST:
584#if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
585 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
586 break;
587#endif
588 if (found_rhdr)
589 return offset;
590 break;
591 default :
592 return offset;
593 }
594
595 offset += ipv6_optlen(exthdr);
596 *nexthdr = &exthdr->nexthdr;
597 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
598 offset);
599 }
600
601 return offset;
602}
603
604void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
605{
606 static atomic_t ipv6_fragmentation_id;
607 int old, new;
608
609 if (rt) {
610 struct inet_peer *peer;
611
612 if (!rt->rt6i_peer)
613 rt6_bind_peer(rt, 1);
614 peer = rt->rt6i_peer;
615 if (peer) {
616 fhdr->identification = htonl(inet_getid(peer, 0));
617 return;
618 }
619 }
620 do {
621 old = atomic_read(&ipv6_fragmentation_id);
622 new = old + 1;
623 if (!new)
624 new = 1;
625 } while (atomic_cmpxchg(&ipv6_fragmentation_id, old, new) != old);
626 fhdr->identification = htonl(new);
627}
628
629int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
630{
631 struct sk_buff *frag;
632 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
633 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
634 struct ipv6hdr *tmp_hdr;
635 struct frag_hdr *fh;
636 unsigned int mtu, hlen, left, len;
637 __be32 frag_id = 0;
638 int ptr, offset = 0, err=0;
639 u8 *prevhdr, nexthdr = 0;
640 struct net *net = dev_net(skb_dst(skb)->dev);
641
642 hlen = ip6_find_1stfragopt(skb, &prevhdr);
643 nexthdr = *prevhdr;
644
645 mtu = ip6_skb_dst_mtu(skb);
646
647 /* We must not fragment if the socket is set to force MTU discovery
648 * or if the skb it not generated by a local socket.
649 */
650 if (!skb->local_df && skb->len > mtu) {
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 left = skb->len - hlen; /* Space per frame */
796 ptr = hlen; /* Where to start from */
797
798 /*
799 * Fragment the datagram.
800 */
801
802 *prevhdr = NEXTHDR_FRAGMENT;
803
804 /*
805 * Keep copying data until we run out.
806 */
807 while(left > 0) {
808 len = left;
809 /* IF: it doesn't fit, use 'mtu' - the data space left */
810 if (len > mtu)
811 len = mtu;
812 /* IF: we are not sending up to and including the packet end
813 then align the next start on an eight byte boundary */
814 if (len < left) {
815 len &= ~7;
816 }
817 /*
818 * Allocate buffer.
819 */
820
821 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_ALLOCATED_SPACE(rt->dst.dev), GFP_ATOMIC)) == NULL) {
822 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
823 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
824 IPSTATS_MIB_FRAGFAILS);
825 err = -ENOMEM;
826 goto fail;
827 }
828
829 /*
830 * Set up data on packet
831 */
832
833 ip6_copy_metadata(frag, skb);
834 skb_reserve(frag, LL_RESERVED_SPACE(rt->dst.dev));
835 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
836 skb_reset_network_header(frag);
837 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
838 frag->transport_header = (frag->network_header + hlen +
839 sizeof(struct frag_hdr));
840
841 /*
842 * Charge the memory for the fragment to any owner
843 * it might possess
844 */
845 if (skb->sk)
846 skb_set_owner_w(frag, skb->sk);
847
848 /*
849 * Copy the packet header into the new buffer.
850 */
851 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
852
853 /*
854 * Build fragment header.
855 */
856 fh->nexthdr = nexthdr;
857 fh->reserved = 0;
858 if (!frag_id) {
859 ipv6_select_ident(fh, rt);
860 frag_id = fh->identification;
861 } else
862 fh->identification = frag_id;
863
864 /*
865 * Copy a block of the IP datagram.
866 */
867 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
868 BUG();
869 left -= len;
870
871 fh->frag_off = htons(offset);
872 if (left > 0)
873 fh->frag_off |= htons(IP6_MF);
874 ipv6_hdr(frag)->payload_len = htons(frag->len -
875 sizeof(struct ipv6hdr));
876
877 ptr += len;
878 offset += len;
879
880 /*
881 * Put this fragment into the sending queue.
882 */
883 err = output(frag);
884 if (err)
885 goto fail;
886
887 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
888 IPSTATS_MIB_FRAGCREATES);
889 }
890 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
891 IPSTATS_MIB_FRAGOKS);
892 kfree_skb(skb);
893 return err;
894
895fail:
896 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
897 IPSTATS_MIB_FRAGFAILS);
898 kfree_skb(skb);
899 return err;
900}
901
902static inline int ip6_rt_check(const struct rt6key *rt_key,
903 const struct in6_addr *fl_addr,
904 const struct in6_addr *addr_cache)
905{
906 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
907 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
908}
909
910static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
911 struct dst_entry *dst,
912 const struct flowi6 *fl6)
913{
914 struct ipv6_pinfo *np = inet6_sk(sk);
915 struct rt6_info *rt = (struct rt6_info *)dst;
916
917 if (!dst)
918 goto out;
919
920 /* Yes, checking route validity in not connected
921 * case is not very simple. Take into account,
922 * that we do not support routing by source, TOS,
923 * and MSG_DONTROUTE --ANK (980726)
924 *
925 * 1. ip6_rt_check(): If route was host route,
926 * check that cached destination is current.
927 * If it is network route, we still may
928 * check its validity using saved pointer
929 * to the last used address: daddr_cache.
930 * We do not want to save whole address now,
931 * (because main consumer of this service
932 * is tcp, which has not this problem),
933 * so that the last trick works only on connected
934 * sockets.
935 * 2. oif also should be the same.
936 */
937 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
938#ifdef CONFIG_IPV6_SUBTREES
939 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
940#endif
941 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
942 dst_release(dst);
943 dst = NULL;
944 }
945
946out:
947 return dst;
948}
949
950static int ip6_dst_lookup_tail(struct sock *sk,
951 struct dst_entry **dst, struct flowi6 *fl6)
952{
953 struct net *net = sock_net(sk);
954#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
955 struct neighbour *n;
956#endif
957 int err;
958
959 if (*dst == NULL)
960 *dst = ip6_route_output(net, sk, fl6);
961
962 if ((err = (*dst)->error))
963 goto out_err_release;
964
965 if (ipv6_addr_any(&fl6->saddr)) {
966 struct rt6_info *rt = (struct rt6_info *) *dst;
967 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
968 sk ? inet6_sk(sk)->srcprefs : 0,
969 &fl6->saddr);
970 if (err)
971 goto out_err_release;
972 }
973
974#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
975 /*
976 * Here if the dst entry we've looked up
977 * has a neighbour entry that is in the INCOMPLETE
978 * state and the src address from the flow is
979 * marked as OPTIMISTIC, we release the found
980 * dst entry and replace it instead with the
981 * dst entry of the nexthop router
982 */
983 rcu_read_lock();
984 n = dst_get_neighbour(*dst);
985 if (n && !(n->nud_state & NUD_VALID)) {
986 struct inet6_ifaddr *ifp;
987 struct flowi6 fl_gw6;
988 int redirect;
989
990 rcu_read_unlock();
991 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
992 (*dst)->dev, 1);
993
994 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
995 if (ifp)
996 in6_ifa_put(ifp);
997
998 if (redirect) {
999 /*
1000 * We need to get the dst entry for the
1001 * default router instead
1002 */
1003 dst_release(*dst);
1004 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1005 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1006 *dst = ip6_route_output(net, sk, &fl_gw6);
1007 if ((err = (*dst)->error))
1008 goto out_err_release;
1009 }
1010 } else {
1011 rcu_read_unlock();
1012 }
1013#endif
1014
1015 return 0;
1016
1017out_err_release:
1018 if (err == -ENETUNREACH)
1019 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1020 dst_release(*dst);
1021 *dst = NULL;
1022 return err;
1023}
1024
1025/**
1026 * ip6_dst_lookup - perform route lookup on flow
1027 * @sk: socket which provides route info
1028 * @dst: pointer to dst_entry * for result
1029 * @fl6: flow to lookup
1030 *
1031 * This function performs a route lookup on the given flow.
1032 *
1033 * It returns zero on success, or a standard errno code on error.
1034 */
1035int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1036{
1037 *dst = NULL;
1038 return ip6_dst_lookup_tail(sk, dst, fl6);
1039}
1040EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1041
1042/**
1043 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1044 * @sk: socket which provides route info
1045 * @fl6: flow to lookup
1046 * @final_dst: final destination address for ipsec lookup
1047 * @can_sleep: we are in a sleepable context
1048 *
1049 * This function performs a route lookup on the given flow.
1050 *
1051 * It returns a valid dst pointer on success, or a pointer encoded
1052 * error code.
1053 */
1054struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1055 const struct in6_addr *final_dst,
1056 bool can_sleep)
1057{
1058 struct dst_entry *dst = NULL;
1059 int err;
1060
1061 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1062 if (err)
1063 return ERR_PTR(err);
1064 if (final_dst)
1065 ipv6_addr_copy(&fl6->daddr, final_dst);
1066 if (can_sleep)
1067 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1068
1069 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1070}
1071EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1072
1073/**
1074 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1075 * @sk: socket which provides the dst cache and route info
1076 * @fl6: flow to lookup
1077 * @final_dst: final destination address for ipsec lookup
1078 * @can_sleep: we are in a sleepable context
1079 *
1080 * This function performs a route lookup on the given flow with the
1081 * possibility of using the cached route in the socket if it is valid.
1082 * It will take the socket dst lock when operating on the dst cache.
1083 * As a result, this function can only be used in process context.
1084 *
1085 * It returns a valid dst pointer on success, or a pointer encoded
1086 * error code.
1087 */
1088struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1089 const struct in6_addr *final_dst,
1090 bool can_sleep)
1091{
1092 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1093 int err;
1094
1095 dst = ip6_sk_dst_check(sk, dst, fl6);
1096
1097 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1098 if (err)
1099 return ERR_PTR(err);
1100 if (final_dst)
1101 ipv6_addr_copy(&fl6->daddr, final_dst);
1102 if (can_sleep)
1103 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1104
1105 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1106}
1107EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1108
1109static inline int ip6_ufo_append_data(struct sock *sk,
1110 int getfrag(void *from, char *to, int offset, int len,
1111 int odd, struct sk_buff *skb),
1112 void *from, int length, int hh_len, int fragheaderlen,
1113 int transhdrlen, int mtu,unsigned int flags,
1114 struct rt6_info *rt)
1115
1116{
1117 struct sk_buff *skb;
1118 int err;
1119
1120 /* There is support for UDP large send offload by network
1121 * device, so create one single skb packet containing complete
1122 * udp datagram
1123 */
1124 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1125 skb = sock_alloc_send_skb(sk,
1126 hh_len + fragheaderlen + transhdrlen + 20,
1127 (flags & MSG_DONTWAIT), &err);
1128 if (skb == NULL)
1129 return -ENOMEM;
1130
1131 /* reserve space for Hardware header */
1132 skb_reserve(skb, hh_len);
1133
1134 /* create space for UDP/IP header */
1135 skb_put(skb,fragheaderlen + transhdrlen);
1136
1137 /* initialize network header pointer */
1138 skb_reset_network_header(skb);
1139
1140 /* initialize protocol header pointer */
1141 skb->transport_header = skb->network_header + fragheaderlen;
1142
1143 skb->ip_summed = CHECKSUM_PARTIAL;
1144 skb->csum = 0;
1145 }
1146
1147 err = skb_append_datato_frags(sk,skb, getfrag, from,
1148 (length - transhdrlen));
1149 if (!err) {
1150 struct frag_hdr fhdr;
1151
1152 /* Specify the length of each IPv6 datagram fragment.
1153 * It has to be a multiple of 8.
1154 */
1155 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1156 sizeof(struct frag_hdr)) & ~7;
1157 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1158 ipv6_select_ident(&fhdr, rt);
1159 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1160 __skb_queue_tail(&sk->sk_write_queue, skb);
1161
1162 return 0;
1163 }
1164 /* There is not enough support do UPD LSO,
1165 * so follow normal path
1166 */
1167 kfree_skb(skb);
1168
1169 return err;
1170}
1171
1172static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1173 gfp_t gfp)
1174{
1175 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1176}
1177
1178static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1179 gfp_t gfp)
1180{
1181 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1182}
1183
1184int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1185 int offset, int len, int odd, struct sk_buff *skb),
1186 void *from, int length, int transhdrlen,
1187 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1188 struct rt6_info *rt, unsigned int flags, int dontfrag)
1189{
1190 struct inet_sock *inet = inet_sk(sk);
1191 struct ipv6_pinfo *np = inet6_sk(sk);
1192 struct inet_cork *cork;
1193 struct sk_buff *skb;
1194 unsigned int maxfraglen, fragheaderlen;
1195 int exthdrlen;
1196 int hh_len;
1197 int mtu;
1198 int copy;
1199 int err;
1200 int offset = 0;
1201 int csummode = CHECKSUM_NONE;
1202 __u8 tx_flags = 0;
1203
1204 if (flags&MSG_PROBE)
1205 return 0;
1206 cork = &inet->cork.base;
1207 if (skb_queue_empty(&sk->sk_write_queue)) {
1208 /*
1209 * setup for corking
1210 */
1211 if (opt) {
1212 if (WARN_ON(np->cork.opt))
1213 return -EINVAL;
1214
1215 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1216 if (unlikely(np->cork.opt == NULL))
1217 return -ENOBUFS;
1218
1219 np->cork.opt->tot_len = opt->tot_len;
1220 np->cork.opt->opt_flen = opt->opt_flen;
1221 np->cork.opt->opt_nflen = opt->opt_nflen;
1222
1223 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1224 sk->sk_allocation);
1225 if (opt->dst0opt && !np->cork.opt->dst0opt)
1226 return -ENOBUFS;
1227
1228 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1229 sk->sk_allocation);
1230 if (opt->dst1opt && !np->cork.opt->dst1opt)
1231 return -ENOBUFS;
1232
1233 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1234 sk->sk_allocation);
1235 if (opt->hopopt && !np->cork.opt->hopopt)
1236 return -ENOBUFS;
1237
1238 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1239 sk->sk_allocation);
1240 if (opt->srcrt && !np->cork.opt->srcrt)
1241 return -ENOBUFS;
1242
1243 /* need source address above miyazawa*/
1244 }
1245 dst_hold(&rt->dst);
1246 cork->dst = &rt->dst;
1247 inet->cork.fl.u.ip6 = *fl6;
1248 np->cork.hop_limit = hlimit;
1249 np->cork.tclass = tclass;
1250 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1251 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1252 if (np->frag_size < mtu) {
1253 if (np->frag_size)
1254 mtu = np->frag_size;
1255 }
1256 cork->fragsize = mtu;
1257 if (dst_allfrag(rt->dst.path))
1258 cork->flags |= IPCORK_ALLFRAG;
1259 cork->length = 0;
1260 sk->sk_sndmsg_page = NULL;
1261 sk->sk_sndmsg_off = 0;
1262 exthdrlen = rt->dst.header_len + (opt ? opt->opt_flen : 0) -
1263 rt->rt6i_nfheader_len;
1264 length += exthdrlen;
1265 transhdrlen += exthdrlen;
1266 } else {
1267 rt = (struct rt6_info *)cork->dst;
1268 fl6 = &inet->cork.fl.u.ip6;
1269 opt = np->cork.opt;
1270 transhdrlen = 0;
1271 exthdrlen = 0;
1272 mtu = cork->fragsize;
1273 }
1274
1275 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1276
1277 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1278 (opt ? opt->opt_nflen : 0);
1279 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1280
1281 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1282 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1283 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1284 return -EMSGSIZE;
1285 }
1286 }
1287
1288 /* For UDP, check if TX timestamp is enabled */
1289 if (sk->sk_type == SOCK_DGRAM) {
1290 err = sock_tx_timestamp(sk, &tx_flags);
1291 if (err)
1292 goto error;
1293 }
1294
1295 /*
1296 * Let's try using as much space as possible.
1297 * Use MTU if total length of the message fits into the MTU.
1298 * Otherwise, we need to reserve fragment header and
1299 * fragment alignment (= 8-15 octects, in total).
1300 *
1301 * Note that we may need to "move" the data from the tail of
1302 * of the buffer to the new fragment when we split
1303 * the message.
1304 *
1305 * FIXME: It may be fragmented into multiple chunks
1306 * at once if non-fragmentable extension headers
1307 * are too large.
1308 * --yoshfuji
1309 */
1310
1311 cork->length += length;
1312 if (length > mtu) {
1313 int proto = sk->sk_protocol;
1314 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1315 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1316 return -EMSGSIZE;
1317 }
1318
1319 if (proto == IPPROTO_UDP &&
1320 (rt->dst.dev->features & NETIF_F_UFO)) {
1321
1322 err = ip6_ufo_append_data(sk, getfrag, from, length,
1323 hh_len, fragheaderlen,
1324 transhdrlen, mtu, flags, rt);
1325 if (err)
1326 goto error;
1327 return 0;
1328 }
1329 }
1330
1331 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1332 goto alloc_new_skb;
1333
1334 while (length > 0) {
1335 /* Check if the remaining data fits into current packet. */
1336 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1337 if (copy < length)
1338 copy = maxfraglen - skb->len;
1339
1340 if (copy <= 0) {
1341 char *data;
1342 unsigned int datalen;
1343 unsigned int fraglen;
1344 unsigned int fraggap;
1345 unsigned int alloclen;
1346 struct sk_buff *skb_prev;
1347alloc_new_skb:
1348 skb_prev = skb;
1349
1350 /* There's no room in the current skb */
1351 if (skb_prev)
1352 fraggap = skb_prev->len - maxfraglen;
1353 else
1354 fraggap = 0;
1355
1356 /*
1357 * If remaining data exceeds the mtu,
1358 * we know we need more fragment(s).
1359 */
1360 datalen = length + fraggap;
1361 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1362 datalen = maxfraglen - fragheaderlen;
1363
1364 fraglen = datalen + fragheaderlen;
1365 if ((flags & MSG_MORE) &&
1366 !(rt->dst.dev->features&NETIF_F_SG))
1367 alloclen = mtu;
1368 else
1369 alloclen = datalen + fragheaderlen;
1370
1371 /*
1372 * The last fragment gets additional space at tail.
1373 * Note: we overallocate on fragments with MSG_MODE
1374 * because we have no idea if we're the last one.
1375 */
1376 if (datalen == length + fraggap)
1377 alloclen += rt->dst.trailer_len;
1378
1379 /*
1380 * We just reserve space for fragment header.
1381 * Note: this may be overallocation if the message
1382 * (without MSG_MORE) fits into the MTU.
1383 */
1384 alloclen += sizeof(struct frag_hdr);
1385
1386 if (transhdrlen) {
1387 skb = sock_alloc_send_skb(sk,
1388 alloclen + hh_len,
1389 (flags & MSG_DONTWAIT), &err);
1390 } else {
1391 skb = NULL;
1392 if (atomic_read(&sk->sk_wmem_alloc) <=
1393 2 * sk->sk_sndbuf)
1394 skb = sock_wmalloc(sk,
1395 alloclen + hh_len, 1,
1396 sk->sk_allocation);
1397 if (unlikely(skb == NULL))
1398 err = -ENOBUFS;
1399 else {
1400 /* Only the initial fragment
1401 * is time stamped.
1402 */
1403 tx_flags = 0;
1404 }
1405 }
1406 if (skb == NULL)
1407 goto error;
1408 /*
1409 * Fill in the control structures
1410 */
1411 skb->ip_summed = csummode;
1412 skb->csum = 0;
1413 /* reserve for fragmentation */
1414 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1415
1416 if (sk->sk_type == SOCK_DGRAM)
1417 skb_shinfo(skb)->tx_flags = tx_flags;
1418
1419 /*
1420 * Find where to start putting bytes
1421 */
1422 data = skb_put(skb, fraglen);
1423 skb_set_network_header(skb, exthdrlen);
1424 data += fragheaderlen;
1425 skb->transport_header = (skb->network_header +
1426 fragheaderlen);
1427 if (fraggap) {
1428 skb->csum = skb_copy_and_csum_bits(
1429 skb_prev, maxfraglen,
1430 data + transhdrlen, fraggap, 0);
1431 skb_prev->csum = csum_sub(skb_prev->csum,
1432 skb->csum);
1433 data += fraggap;
1434 pskb_trim_unique(skb_prev, maxfraglen);
1435 }
1436 copy = datalen - transhdrlen - fraggap;
1437 if (copy < 0) {
1438 err = -EINVAL;
1439 kfree_skb(skb);
1440 goto error;
1441 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1442 err = -EFAULT;
1443 kfree_skb(skb);
1444 goto error;
1445 }
1446
1447 offset += copy;
1448 length -= datalen - fraggap;
1449 transhdrlen = 0;
1450 exthdrlen = 0;
1451 csummode = CHECKSUM_NONE;
1452
1453 /*
1454 * Put the packet on the pending queue
1455 */
1456 __skb_queue_tail(&sk->sk_write_queue, skb);
1457 continue;
1458 }
1459
1460 if (copy > length)
1461 copy = length;
1462
1463 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1464 unsigned int off;
1465
1466 off = skb->len;
1467 if (getfrag(from, skb_put(skb, copy),
1468 offset, copy, off, skb) < 0) {
1469 __skb_trim(skb, off);
1470 err = -EFAULT;
1471 goto error;
1472 }
1473 } else {
1474 int i = skb_shinfo(skb)->nr_frags;
1475 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1476 struct page *page = sk->sk_sndmsg_page;
1477 int off = sk->sk_sndmsg_off;
1478 unsigned int left;
1479
1480 if (page && (left = PAGE_SIZE - off) > 0) {
1481 if (copy >= left)
1482 copy = left;
1483 if (page != frag->page) {
1484 if (i == MAX_SKB_FRAGS) {
1485 err = -EMSGSIZE;
1486 goto error;
1487 }
1488 get_page(page);
1489 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1490 frag = &skb_shinfo(skb)->frags[i];
1491 }
1492 } else if(i < MAX_SKB_FRAGS) {
1493 if (copy > PAGE_SIZE)
1494 copy = PAGE_SIZE;
1495 page = alloc_pages(sk->sk_allocation, 0);
1496 if (page == NULL) {
1497 err = -ENOMEM;
1498 goto error;
1499 }
1500 sk->sk_sndmsg_page = page;
1501 sk->sk_sndmsg_off = 0;
1502
1503 skb_fill_page_desc(skb, i, page, 0, 0);
1504 frag = &skb_shinfo(skb)->frags[i];
1505 } else {
1506 err = -EMSGSIZE;
1507 goto error;
1508 }
1509 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1510 err = -EFAULT;
1511 goto error;
1512 }
1513 sk->sk_sndmsg_off += copy;
1514 frag->size += copy;
1515 skb->len += copy;
1516 skb->data_len += copy;
1517 skb->truesize += copy;
1518 atomic_add(copy, &sk->sk_wmem_alloc);
1519 }
1520 offset += copy;
1521 length -= copy;
1522 }
1523 return 0;
1524error:
1525 cork->length -= length;
1526 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1527 return err;
1528}
1529
1530static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1531{
1532 if (np->cork.opt) {
1533 kfree(np->cork.opt->dst0opt);
1534 kfree(np->cork.opt->dst1opt);
1535 kfree(np->cork.opt->hopopt);
1536 kfree(np->cork.opt->srcrt);
1537 kfree(np->cork.opt);
1538 np->cork.opt = NULL;
1539 }
1540
1541 if (inet->cork.base.dst) {
1542 dst_release(inet->cork.base.dst);
1543 inet->cork.base.dst = NULL;
1544 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1545 }
1546 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1547}
1548
1549int ip6_push_pending_frames(struct sock *sk)
1550{
1551 struct sk_buff *skb, *tmp_skb;
1552 struct sk_buff **tail_skb;
1553 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1554 struct inet_sock *inet = inet_sk(sk);
1555 struct ipv6_pinfo *np = inet6_sk(sk);
1556 struct net *net = sock_net(sk);
1557 struct ipv6hdr *hdr;
1558 struct ipv6_txoptions *opt = np->cork.opt;
1559 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1560 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1561 unsigned char proto = fl6->flowi6_proto;
1562 int err = 0;
1563
1564 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1565 goto out;
1566 tail_skb = &(skb_shinfo(skb)->frag_list);
1567
1568 /* move skb->data to ip header from ext header */
1569 if (skb->data < skb_network_header(skb))
1570 __skb_pull(skb, skb_network_offset(skb));
1571 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1572 __skb_pull(tmp_skb, skb_network_header_len(skb));
1573 *tail_skb = tmp_skb;
1574 tail_skb = &(tmp_skb->next);
1575 skb->len += tmp_skb->len;
1576 skb->data_len += tmp_skb->len;
1577 skb->truesize += tmp_skb->truesize;
1578 tmp_skb->destructor = NULL;
1579 tmp_skb->sk = NULL;
1580 }
1581
1582 /* Allow local fragmentation. */
1583 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1584 skb->local_df = 1;
1585
1586 ipv6_addr_copy(final_dst, &fl6->daddr);
1587 __skb_pull(skb, skb_network_header_len(skb));
1588 if (opt && opt->opt_flen)
1589 ipv6_push_frag_opts(skb, opt, &proto);
1590 if (opt && opt->opt_nflen)
1591 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1592
1593 skb_push(skb, sizeof(struct ipv6hdr));
1594 skb_reset_network_header(skb);
1595 hdr = ipv6_hdr(skb);
1596
1597 *(__be32*)hdr = fl6->flowlabel |
1598 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1599
1600 hdr->hop_limit = np->cork.hop_limit;
1601 hdr->nexthdr = proto;
1602 ipv6_addr_copy(&hdr->saddr, &fl6->saddr);
1603 ipv6_addr_copy(&hdr->daddr, final_dst);
1604
1605 skb->priority = sk->sk_priority;
1606 skb->mark = sk->sk_mark;
1607
1608 skb_dst_set(skb, dst_clone(&rt->dst));
1609 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1610 if (proto == IPPROTO_ICMPV6) {
1611 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1612
1613 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1614 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1615 }
1616
1617 err = ip6_local_out(skb);
1618 if (err) {
1619 if (err > 0)
1620 err = net_xmit_errno(err);
1621 if (err)
1622 goto error;
1623 }
1624
1625out:
1626 ip6_cork_release(inet, np);
1627 return err;
1628error:
1629 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1630 goto out;
1631}
1632
1633void ip6_flush_pending_frames(struct sock *sk)
1634{
1635 struct sk_buff *skb;
1636
1637 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1638 if (skb_dst(skb))
1639 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1640 IPSTATS_MIB_OUTDISCARDS);
1641 kfree_skb(skb);
1642 }
1643
1644 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1645}
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#include <net/l3mdev.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 neighbour *neigh;
65 struct in6_addr *nexthop;
66 int ret;
67
68 skb->protocol = htons(ETH_P_IPV6);
69 skb->dev = dev;
70
71 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
72 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
73
74 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
75 ((mroute6_socket(net, skb) &&
76 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
77 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
78 &ipv6_hdr(skb)->saddr))) {
79 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
80
81 /* Do not check for IFF_ALLMULTI; multicast routing
82 is not supported in any case.
83 */
84 if (newskb)
85 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
86 net, sk, newskb, NULL, newskb->dev,
87 dev_loopback_xmit);
88
89 if (ipv6_hdr(skb)->hop_limit == 0) {
90 IP6_INC_STATS(net, idev,
91 IPSTATS_MIB_OUTDISCARDS);
92 kfree_skb(skb);
93 return 0;
94 }
95 }
96
97 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
98
99 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 IPV6_ADDR_SCOPE_NODELOCAL &&
101 !(dev->flags & IFF_LOOPBACK)) {
102 kfree_skb(skb);
103 return 0;
104 }
105 }
106
107 rcu_read_lock_bh();
108 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
109 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 if (unlikely(!neigh))
111 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 if (!IS_ERR(neigh)) {
113 ret = dst_neigh_output(dst, neigh, skb);
114 rcu_read_unlock_bh();
115 return ret;
116 }
117 rcu_read_unlock_bh();
118
119 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
120 kfree_skb(skb);
121 return -EINVAL;
122}
123
124static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
125{
126 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
127 dst_allfrag(skb_dst(skb)) ||
128 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
129 return ip6_fragment(net, sk, skb, ip6_finish_output2);
130 else
131 return ip6_finish_output2(net, sk, skb);
132}
133
134int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
135{
136 struct net_device *dev = skb_dst(skb)->dev;
137 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
138
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
141 kfree_skb(skb);
142 return 0;
143 }
144
145 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
146 net, sk, skb, NULL, dev,
147 ip6_finish_output,
148 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
149}
150
151/*
152 * xmit an sk_buff (used by TCP, SCTP and DCCP)
153 * Note : socket lock is not held for SYNACK packets, but might be modified
154 * by calls to skb_set_owner_w() and ipv6_local_error(),
155 * which are using proper atomic operations or spinlocks.
156 */
157int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
158 struct ipv6_txoptions *opt, int tclass)
159{
160 struct net *net = sock_net(sk);
161 const struct ipv6_pinfo *np = inet6_sk(sk);
162 struct in6_addr *first_hop = &fl6->daddr;
163 struct dst_entry *dst = skb_dst(skb);
164 struct ipv6hdr *hdr;
165 u8 proto = fl6->flowi6_proto;
166 int seg_len = skb->len;
167 int hlimit = -1;
168 u32 mtu;
169
170 if (opt) {
171 unsigned int head_room;
172
173 /* First: exthdrs may take lots of space (~8K for now)
174 MAX_HEADER is not enough.
175 */
176 head_room = opt->opt_nflen + opt->opt_flen;
177 seg_len += head_room;
178 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
179
180 if (skb_headroom(skb) < head_room) {
181 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
182 if (!skb2) {
183 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
184 IPSTATS_MIB_OUTDISCARDS);
185 kfree_skb(skb);
186 return -ENOBUFS;
187 }
188 consume_skb(skb);
189 skb = skb2;
190 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
191 * it is safe to call in our context (socket lock not held)
192 */
193 skb_set_owner_w(skb, (struct sock *)sk);
194 }
195 if (opt->opt_flen)
196 ipv6_push_frag_opts(skb, opt, &proto);
197 if (opt->opt_nflen)
198 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
199 }
200
201 skb_push(skb, sizeof(struct ipv6hdr));
202 skb_reset_network_header(skb);
203 hdr = ipv6_hdr(skb);
204
205 /*
206 * Fill in the IPv6 header
207 */
208 if (np)
209 hlimit = np->hop_limit;
210 if (hlimit < 0)
211 hlimit = ip6_dst_hoplimit(dst);
212
213 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
214 np->autoflowlabel, fl6));
215
216 hdr->payload_len = htons(seg_len);
217 hdr->nexthdr = proto;
218 hdr->hop_limit = hlimit;
219
220 hdr->saddr = fl6->saddr;
221 hdr->daddr = *first_hop;
222
223 skb->protocol = htons(ETH_P_IPV6);
224 skb->priority = sk->sk_priority;
225 skb->mark = sk->sk_mark;
226
227 mtu = dst_mtu(dst);
228 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
229 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
230 IPSTATS_MIB_OUT, skb->len);
231 /* hooks should never assume socket lock is held.
232 * we promote our socket to non const
233 */
234 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
235 net, (struct sock *)sk, skb, NULL, dst->dev,
236 dst_output);
237 }
238
239 skb->dev = dst->dev;
240 /* ipv6_local_error() does not require socket lock,
241 * we promote our socket to non const
242 */
243 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
244
245 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
246 kfree_skb(skb);
247 return -EMSGSIZE;
248}
249EXPORT_SYMBOL(ip6_xmit);
250
251static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
252{
253 struct ip6_ra_chain *ra;
254 struct sock *last = NULL;
255
256 read_lock(&ip6_ra_lock);
257 for (ra = ip6_ra_chain; ra; ra = ra->next) {
258 struct sock *sk = ra->sk;
259 if (sk && ra->sel == sel &&
260 (!sk->sk_bound_dev_if ||
261 sk->sk_bound_dev_if == skb->dev->ifindex)) {
262 if (last) {
263 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
264 if (skb2)
265 rawv6_rcv(last, skb2);
266 }
267 last = sk;
268 }
269 }
270
271 if (last) {
272 rawv6_rcv(last, skb);
273 read_unlock(&ip6_ra_lock);
274 return 1;
275 }
276 read_unlock(&ip6_ra_lock);
277 return 0;
278}
279
280static int ip6_forward_proxy_check(struct sk_buff *skb)
281{
282 struct ipv6hdr *hdr = ipv6_hdr(skb);
283 u8 nexthdr = hdr->nexthdr;
284 __be16 frag_off;
285 int offset;
286
287 if (ipv6_ext_hdr(nexthdr)) {
288 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
289 if (offset < 0)
290 return 0;
291 } else
292 offset = sizeof(struct ipv6hdr);
293
294 if (nexthdr == IPPROTO_ICMPV6) {
295 struct icmp6hdr *icmp6;
296
297 if (!pskb_may_pull(skb, (skb_network_header(skb) +
298 offset + 1 - skb->data)))
299 return 0;
300
301 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
302
303 switch (icmp6->icmp6_type) {
304 case NDISC_ROUTER_SOLICITATION:
305 case NDISC_ROUTER_ADVERTISEMENT:
306 case NDISC_NEIGHBOUR_SOLICITATION:
307 case NDISC_NEIGHBOUR_ADVERTISEMENT:
308 case NDISC_REDIRECT:
309 /* For reaction involving unicast neighbor discovery
310 * message destined to the proxied address, pass it to
311 * input function.
312 */
313 return 1;
314 default:
315 break;
316 }
317 }
318
319 /*
320 * The proxying router can't forward traffic sent to a link-local
321 * address, so signal the sender and discard the packet. This
322 * behavior is clarified by the MIPv6 specification.
323 */
324 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
325 dst_link_failure(skb);
326 return -1;
327 }
328
329 return 0;
330}
331
332static inline int ip6_forward_finish(struct net *net, struct sock *sk,
333 struct sk_buff *skb)
334{
335 return dst_output(net, sk, skb);
336}
337
338static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
339{
340 unsigned int mtu;
341 struct inet6_dev *idev;
342
343 if (dst_metric_locked(dst, RTAX_MTU)) {
344 mtu = dst_metric_raw(dst, RTAX_MTU);
345 if (mtu)
346 return mtu;
347 }
348
349 mtu = IPV6_MIN_MTU;
350 rcu_read_lock();
351 idev = __in6_dev_get(dst->dev);
352 if (idev)
353 mtu = idev->cnf.mtu6;
354 rcu_read_unlock();
355
356 return mtu;
357}
358
359static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
360{
361 if (skb->len <= mtu)
362 return false;
363
364 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
365 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
366 return true;
367
368 if (skb->ignore_df)
369 return false;
370
371 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
372 return false;
373
374 return true;
375}
376
377int ip6_forward(struct sk_buff *skb)
378{
379 struct dst_entry *dst = skb_dst(skb);
380 struct ipv6hdr *hdr = ipv6_hdr(skb);
381 struct inet6_skb_parm *opt = IP6CB(skb);
382 struct net *net = dev_net(dst->dev);
383 u32 mtu;
384
385 if (net->ipv6.devconf_all->forwarding == 0)
386 goto error;
387
388 if (skb->pkt_type != PACKET_HOST)
389 goto drop;
390
391 if (unlikely(skb->sk))
392 goto drop;
393
394 if (skb_warn_if_lro(skb))
395 goto drop;
396
397 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
398 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
399 IPSTATS_MIB_INDISCARDS);
400 goto drop;
401 }
402
403 skb_forward_csum(skb);
404
405 /*
406 * We DO NOT make any processing on
407 * RA packets, pushing them to user level AS IS
408 * without ane WARRANTY that application will be able
409 * to interpret them. The reason is that we
410 * cannot make anything clever here.
411 *
412 * We are not end-node, so that if packet contains
413 * AH/ESP, we cannot make anything.
414 * Defragmentation also would be mistake, RA packets
415 * cannot be fragmented, because there is no warranty
416 * that different fragments will go along one path. --ANK
417 */
418 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
419 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
420 return 0;
421 }
422
423 /*
424 * check and decrement ttl
425 */
426 if (hdr->hop_limit <= 1) {
427 /* Force OUTPUT device used as source address */
428 skb->dev = dst->dev;
429 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
430 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
431 IPSTATS_MIB_INHDRERRORS);
432
433 kfree_skb(skb);
434 return -ETIMEDOUT;
435 }
436
437 /* XXX: idev->cnf.proxy_ndp? */
438 if (net->ipv6.devconf_all->proxy_ndp &&
439 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
440 int proxied = ip6_forward_proxy_check(skb);
441 if (proxied > 0)
442 return ip6_input(skb);
443 else if (proxied < 0) {
444 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
445 IPSTATS_MIB_INDISCARDS);
446 goto drop;
447 }
448 }
449
450 if (!xfrm6_route_forward(skb)) {
451 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
452 IPSTATS_MIB_INDISCARDS);
453 goto drop;
454 }
455 dst = skb_dst(skb);
456
457 /* IPv6 specs say nothing about it, but it is clear that we cannot
458 send redirects to source routed frames.
459 We don't send redirects to frames decapsulated from IPsec.
460 */
461 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
462 struct in6_addr *target = NULL;
463 struct inet_peer *peer;
464 struct rt6_info *rt;
465
466 /*
467 * incoming and outgoing devices are the same
468 * send a redirect.
469 */
470
471 rt = (struct rt6_info *) dst;
472 if (rt->rt6i_flags & RTF_GATEWAY)
473 target = &rt->rt6i_gateway;
474 else
475 target = &hdr->daddr;
476
477 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
478
479 /* Limit redirects both by destination (here)
480 and by source (inside ndisc_send_redirect)
481 */
482 if (inet_peer_xrlim_allow(peer, 1*HZ))
483 ndisc_send_redirect(skb, target);
484 if (peer)
485 inet_putpeer(peer);
486 } else {
487 int addrtype = ipv6_addr_type(&hdr->saddr);
488
489 /* This check is security critical. */
490 if (addrtype == IPV6_ADDR_ANY ||
491 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
492 goto error;
493 if (addrtype & IPV6_ADDR_LINKLOCAL) {
494 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
495 ICMPV6_NOT_NEIGHBOUR, 0);
496 goto error;
497 }
498 }
499
500 mtu = ip6_dst_mtu_forward(dst);
501 if (mtu < IPV6_MIN_MTU)
502 mtu = IPV6_MIN_MTU;
503
504 if (ip6_pkt_too_big(skb, mtu)) {
505 /* Again, force OUTPUT device used as source address */
506 skb->dev = dst->dev;
507 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
508 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
509 IPSTATS_MIB_INTOOBIGERRORS);
510 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
511 IPSTATS_MIB_FRAGFAILS);
512 kfree_skb(skb);
513 return -EMSGSIZE;
514 }
515
516 if (skb_cow(skb, dst->dev->hard_header_len)) {
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
518 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,
531 net, NULL, skb, skb->dev, dst->dev,
532 ip6_forward_finish);
533
534error:
535 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
536drop:
537 kfree_skb(skb);
538 return -EINVAL;
539}
540
541static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
542{
543 to->pkt_type = from->pkt_type;
544 to->priority = from->priority;
545 to->protocol = from->protocol;
546 skb_dst_drop(to);
547 skb_dst_set(to, dst_clone(skb_dst(from)));
548 to->dev = from->dev;
549 to->mark = from->mark;
550
551#ifdef CONFIG_NET_SCHED
552 to->tc_index = from->tc_index;
553#endif
554 nf_copy(to, from);
555 skb_copy_secmark(to, from);
556}
557
558int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
559 int (*output)(struct net *, struct sock *, struct sk_buff *))
560{
561 struct sk_buff *frag;
562 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
563 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
564 inet6_sk(skb->sk) : NULL;
565 struct ipv6hdr *tmp_hdr;
566 struct frag_hdr *fh;
567 unsigned int mtu, hlen, left, len;
568 int hroom, troom;
569 __be32 frag_id;
570 int ptr, offset = 0, err = 0;
571 u8 *prevhdr, nexthdr = 0;
572
573 hlen = ip6_find_1stfragopt(skb, &prevhdr);
574 nexthdr = *prevhdr;
575
576 mtu = ip6_skb_dst_mtu(skb);
577
578 /* We must not fragment if the socket is set to force MTU discovery
579 * or if the skb it not generated by a local socket.
580 */
581 if (unlikely(!skb->ignore_df && skb->len > mtu))
582 goto fail_toobig;
583
584 if (IP6CB(skb)->frag_max_size) {
585 if (IP6CB(skb)->frag_max_size > mtu)
586 goto fail_toobig;
587
588 /* don't send fragments larger than what we received */
589 mtu = IP6CB(skb)->frag_max_size;
590 if (mtu < IPV6_MIN_MTU)
591 mtu = IPV6_MIN_MTU;
592 }
593
594 if (np && np->frag_size < mtu) {
595 if (np->frag_size)
596 mtu = np->frag_size;
597 }
598 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
599 goto fail_toobig;
600 mtu -= hlen + sizeof(struct frag_hdr);
601
602 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
603 &ipv6_hdr(skb)->saddr);
604
605 if (skb->ip_summed == CHECKSUM_PARTIAL &&
606 (err = skb_checksum_help(skb)))
607 goto fail;
608
609 hroom = LL_RESERVED_SPACE(rt->dst.dev);
610 if (skb_has_frag_list(skb)) {
611 int first_len = skb_pagelen(skb);
612 struct sk_buff *frag2;
613
614 if (first_len - hlen > mtu ||
615 ((first_len - hlen) & 7) ||
616 skb_cloned(skb) ||
617 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
618 goto slow_path;
619
620 skb_walk_frags(skb, frag) {
621 /* Correct geometry. */
622 if (frag->len > mtu ||
623 ((frag->len & 7) && frag->next) ||
624 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
625 goto slow_path_clean;
626
627 /* Partially cloned skb? */
628 if (skb_shared(frag))
629 goto slow_path_clean;
630
631 BUG_ON(frag->sk);
632 if (skb->sk) {
633 frag->sk = skb->sk;
634 frag->destructor = sock_wfree;
635 }
636 skb->truesize -= frag->truesize;
637 }
638
639 err = 0;
640 offset = 0;
641 /* BUILD HEADER */
642
643 *prevhdr = NEXTHDR_FRAGMENT;
644 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
645 if (!tmp_hdr) {
646 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
647 IPSTATS_MIB_FRAGFAILS);
648 err = -ENOMEM;
649 goto fail;
650 }
651 frag = skb_shinfo(skb)->frag_list;
652 skb_frag_list_init(skb);
653
654 __skb_pull(skb, hlen);
655 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
656 __skb_push(skb, hlen);
657 skb_reset_network_header(skb);
658 memcpy(skb_network_header(skb), tmp_hdr, hlen);
659
660 fh->nexthdr = nexthdr;
661 fh->reserved = 0;
662 fh->frag_off = htons(IP6_MF);
663 fh->identification = frag_id;
664
665 first_len = skb_pagelen(skb);
666 skb->data_len = first_len - skb_headlen(skb);
667 skb->len = first_len;
668 ipv6_hdr(skb)->payload_len = htons(first_len -
669 sizeof(struct ipv6hdr));
670
671 dst_hold(&rt->dst);
672
673 for (;;) {
674 /* Prepare header of the next frame,
675 * before previous one went down. */
676 if (frag) {
677 frag->ip_summed = CHECKSUM_NONE;
678 skb_reset_transport_header(frag);
679 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
680 __skb_push(frag, hlen);
681 skb_reset_network_header(frag);
682 memcpy(skb_network_header(frag), tmp_hdr,
683 hlen);
684 offset += skb->len - hlen - sizeof(struct frag_hdr);
685 fh->nexthdr = nexthdr;
686 fh->reserved = 0;
687 fh->frag_off = htons(offset);
688 if (frag->next)
689 fh->frag_off |= htons(IP6_MF);
690 fh->identification = frag_id;
691 ipv6_hdr(frag)->payload_len =
692 htons(frag->len -
693 sizeof(struct ipv6hdr));
694 ip6_copy_metadata(frag, skb);
695 }
696
697 err = output(net, sk, skb);
698 if (!err)
699 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
700 IPSTATS_MIB_FRAGCREATES);
701
702 if (err || !frag)
703 break;
704
705 skb = frag;
706 frag = skb->next;
707 skb->next = NULL;
708 }
709
710 kfree(tmp_hdr);
711
712 if (err == 0) {
713 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
714 IPSTATS_MIB_FRAGOKS);
715 ip6_rt_put(rt);
716 return 0;
717 }
718
719 kfree_skb_list(frag);
720
721 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
722 IPSTATS_MIB_FRAGFAILS);
723 ip6_rt_put(rt);
724 return err;
725
726slow_path_clean:
727 skb_walk_frags(skb, frag2) {
728 if (frag2 == frag)
729 break;
730 frag2->sk = NULL;
731 frag2->destructor = NULL;
732 skb->truesize += frag2->truesize;
733 }
734 }
735
736slow_path:
737 left = skb->len - hlen; /* Space per frame */
738 ptr = hlen; /* Where to start from */
739
740 /*
741 * Fragment the datagram.
742 */
743
744 *prevhdr = NEXTHDR_FRAGMENT;
745 troom = rt->dst.dev->needed_tailroom;
746
747 /*
748 * Keep copying data until we run out.
749 */
750 while (left > 0) {
751 len = left;
752 /* IF: it doesn't fit, use 'mtu' - the data space left */
753 if (len > mtu)
754 len = mtu;
755 /* IF: we are not sending up to and including the packet end
756 then align the next start on an eight byte boundary */
757 if (len < left) {
758 len &= ~7;
759 }
760
761 /* Allocate buffer */
762 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
763 hroom + troom, GFP_ATOMIC);
764 if (!frag) {
765 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
766 IPSTATS_MIB_FRAGFAILS);
767 err = -ENOMEM;
768 goto fail;
769 }
770
771 /*
772 * Set up data on packet
773 */
774
775 ip6_copy_metadata(frag, skb);
776 skb_reserve(frag, hroom);
777 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
778 skb_reset_network_header(frag);
779 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
780 frag->transport_header = (frag->network_header + hlen +
781 sizeof(struct frag_hdr));
782
783 /*
784 * Charge the memory for the fragment to any owner
785 * it might possess
786 */
787 if (skb->sk)
788 skb_set_owner_w(frag, skb->sk);
789
790 /*
791 * Copy the packet header into the new buffer.
792 */
793 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
794
795 /*
796 * Build fragment header.
797 */
798 fh->nexthdr = nexthdr;
799 fh->reserved = 0;
800 fh->identification = frag_id;
801
802 /*
803 * Copy a block of the IP datagram.
804 */
805 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
806 len));
807 left -= len;
808
809 fh->frag_off = htons(offset);
810 if (left > 0)
811 fh->frag_off |= htons(IP6_MF);
812 ipv6_hdr(frag)->payload_len = htons(frag->len -
813 sizeof(struct ipv6hdr));
814
815 ptr += len;
816 offset += len;
817
818 /*
819 * Put this fragment into the sending queue.
820 */
821 err = output(net, sk, frag);
822 if (err)
823 goto fail;
824
825 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
826 IPSTATS_MIB_FRAGCREATES);
827 }
828 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
829 IPSTATS_MIB_FRAGOKS);
830 consume_skb(skb);
831 return err;
832
833fail_toobig:
834 if (skb->sk && dst_allfrag(skb_dst(skb)))
835 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
836
837 skb->dev = skb_dst(skb)->dev;
838 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
839 err = -EMSGSIZE;
840
841fail:
842 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
843 IPSTATS_MIB_FRAGFAILS);
844 kfree_skb(skb);
845 return err;
846}
847
848static inline int ip6_rt_check(const struct rt6key *rt_key,
849 const struct in6_addr *fl_addr,
850 const struct in6_addr *addr_cache)
851{
852 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
853 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
854}
855
856static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
857 struct dst_entry *dst,
858 const struct flowi6 *fl6)
859{
860 struct ipv6_pinfo *np = inet6_sk(sk);
861 struct rt6_info *rt;
862
863 if (!dst)
864 goto out;
865
866 if (dst->ops->family != AF_INET6) {
867 dst_release(dst);
868 return NULL;
869 }
870
871 rt = (struct rt6_info *)dst;
872 /* Yes, checking route validity in not connected
873 * case is not very simple. Take into account,
874 * that we do not support routing by source, TOS,
875 * and MSG_DONTROUTE --ANK (980726)
876 *
877 * 1. ip6_rt_check(): If route was host route,
878 * check that cached destination is current.
879 * If it is network route, we still may
880 * check its validity using saved pointer
881 * to the last used address: daddr_cache.
882 * We do not want to save whole address now,
883 * (because main consumer of this service
884 * is tcp, which has not this problem),
885 * so that the last trick works only on connected
886 * sockets.
887 * 2. oif also should be the same.
888 */
889 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
890#ifdef CONFIG_IPV6_SUBTREES
891 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
892#endif
893 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
894 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
895 dst_release(dst);
896 dst = NULL;
897 }
898
899out:
900 return dst;
901}
902
903static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
904 struct dst_entry **dst, struct flowi6 *fl6)
905{
906#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
907 struct neighbour *n;
908 struct rt6_info *rt;
909#endif
910 int err;
911 int flags = 0;
912
913 /* The correct way to handle this would be to do
914 * ip6_route_get_saddr, and then ip6_route_output; however,
915 * the route-specific preferred source forces the
916 * ip6_route_output call _before_ ip6_route_get_saddr.
917 *
918 * In source specific routing (no src=any default route),
919 * ip6_route_output will fail given src=any saddr, though, so
920 * that's why we try it again later.
921 */
922 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
923 struct rt6_info *rt;
924 bool had_dst = *dst != NULL;
925
926 if (!had_dst)
927 *dst = ip6_route_output(net, sk, fl6);
928 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
929 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
930 sk ? inet6_sk(sk)->srcprefs : 0,
931 &fl6->saddr);
932 if (err)
933 goto out_err_release;
934
935 /* If we had an erroneous initial result, pretend it
936 * never existed and let the SA-enabled version take
937 * over.
938 */
939 if (!had_dst && (*dst)->error) {
940 dst_release(*dst);
941 *dst = NULL;
942 }
943
944 if (fl6->flowi6_oif)
945 flags |= RT6_LOOKUP_F_IFACE;
946 }
947
948 if (!*dst)
949 *dst = ip6_route_output_flags(net, sk, fl6, flags);
950
951 err = (*dst)->error;
952 if (err)
953 goto out_err_release;
954
955#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
956 /*
957 * Here if the dst entry we've looked up
958 * has a neighbour entry that is in the INCOMPLETE
959 * state and the src address from the flow is
960 * marked as OPTIMISTIC, we release the found
961 * dst entry and replace it instead with the
962 * dst entry of the nexthop router
963 */
964 rt = (struct rt6_info *) *dst;
965 rcu_read_lock_bh();
966 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
967 rt6_nexthop(rt, &fl6->daddr));
968 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
969 rcu_read_unlock_bh();
970
971 if (err) {
972 struct inet6_ifaddr *ifp;
973 struct flowi6 fl_gw6;
974 int redirect;
975
976 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
977 (*dst)->dev, 1);
978
979 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
980 if (ifp)
981 in6_ifa_put(ifp);
982
983 if (redirect) {
984 /*
985 * We need to get the dst entry for the
986 * default router instead
987 */
988 dst_release(*dst);
989 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
990 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
991 *dst = ip6_route_output(net, sk, &fl_gw6);
992 err = (*dst)->error;
993 if (err)
994 goto out_err_release;
995 }
996 }
997#endif
998
999 return 0;
1000
1001out_err_release:
1002 if (err == -ENETUNREACH)
1003 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1004 dst_release(*dst);
1005 *dst = NULL;
1006 return err;
1007}
1008
1009/**
1010 * ip6_dst_lookup - perform route lookup on flow
1011 * @sk: socket which provides route info
1012 * @dst: pointer to dst_entry * for result
1013 * @fl6: flow to lookup
1014 *
1015 * This function performs a route lookup on the given flow.
1016 *
1017 * It returns zero on success, or a standard errno code on error.
1018 */
1019int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1020 struct flowi6 *fl6)
1021{
1022 *dst = NULL;
1023 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1024}
1025EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1026
1027/**
1028 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1029 * @sk: socket which provides route info
1030 * @fl6: flow to lookup
1031 * @final_dst: final destination address for ipsec lookup
1032 *
1033 * This function performs a route lookup on the given flow.
1034 *
1035 * It returns a valid dst pointer on success, or a pointer encoded
1036 * error code.
1037 */
1038struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1039 const struct in6_addr *final_dst)
1040{
1041 struct dst_entry *dst = NULL;
1042 int err;
1043
1044 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1045 if (err)
1046 return ERR_PTR(err);
1047 if (final_dst)
1048 fl6->daddr = *final_dst;
1049 if (!fl6->flowi6_oif)
1050 fl6->flowi6_oif = l3mdev_fib_oif(dst->dev);
1051
1052 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1053}
1054EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1055
1056/**
1057 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1058 * @sk: socket which provides the dst cache and route info
1059 * @fl6: flow to lookup
1060 * @final_dst: final destination address for ipsec lookup
1061 *
1062 * This function performs a route lookup on the given flow with the
1063 * possibility of using the cached route in the socket if it is valid.
1064 * It will take the socket dst lock when operating on the dst cache.
1065 * As a result, this function can only be used in process context.
1066 *
1067 * It returns a valid dst pointer on success, or a pointer encoded
1068 * error code.
1069 */
1070struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1071 const struct in6_addr *final_dst)
1072{
1073 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1074 int err;
1075
1076 dst = ip6_sk_dst_check(sk, dst, fl6);
1077
1078 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1079 if (err)
1080 return ERR_PTR(err);
1081 if (final_dst)
1082 fl6->daddr = *final_dst;
1083
1084 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1085}
1086EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1087
1088static inline int ip6_ufo_append_data(struct sock *sk,
1089 struct sk_buff_head *queue,
1090 int getfrag(void *from, char *to, int offset, int len,
1091 int odd, struct sk_buff *skb),
1092 void *from, int length, int hh_len, int fragheaderlen,
1093 int exthdrlen, int transhdrlen, int mtu,
1094 unsigned int flags, const struct flowi6 *fl6)
1095
1096{
1097 struct sk_buff *skb;
1098 int err;
1099
1100 /* There is support for UDP large send offload by network
1101 * device, so create one single skb packet containing complete
1102 * udp datagram
1103 */
1104 skb = skb_peek_tail(queue);
1105 if (!skb) {
1106 skb = sock_alloc_send_skb(sk,
1107 hh_len + fragheaderlen + transhdrlen + 20,
1108 (flags & MSG_DONTWAIT), &err);
1109 if (!skb)
1110 return err;
1111
1112 /* reserve space for Hardware header */
1113 skb_reserve(skb, hh_len);
1114
1115 /* create space for UDP/IP header */
1116 skb_put(skb, fragheaderlen + transhdrlen);
1117
1118 /* initialize network header pointer */
1119 skb_set_network_header(skb, exthdrlen);
1120
1121 /* initialize protocol header pointer */
1122 skb->transport_header = skb->network_header + fragheaderlen;
1123
1124 skb->protocol = htons(ETH_P_IPV6);
1125 skb->csum = 0;
1126
1127 __skb_queue_tail(queue, skb);
1128 } else if (skb_is_gso(skb)) {
1129 goto append;
1130 }
1131
1132 skb->ip_summed = CHECKSUM_PARTIAL;
1133 /* Specify the length of each IPv6 datagram fragment.
1134 * It has to be a multiple of 8.
1135 */
1136 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1137 sizeof(struct frag_hdr)) & ~7;
1138 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1139 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1140 &fl6->daddr,
1141 &fl6->saddr);
1142
1143append:
1144 return skb_append_datato_frags(sk, skb, getfrag, from,
1145 (length - transhdrlen));
1146}
1147
1148static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1149 gfp_t gfp)
1150{
1151 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1152}
1153
1154static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1155 gfp_t gfp)
1156{
1157 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1158}
1159
1160static void ip6_append_data_mtu(unsigned int *mtu,
1161 int *maxfraglen,
1162 unsigned int fragheaderlen,
1163 struct sk_buff *skb,
1164 struct rt6_info *rt,
1165 unsigned int orig_mtu)
1166{
1167 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1168 if (!skb) {
1169 /* first fragment, reserve header_len */
1170 *mtu = orig_mtu - rt->dst.header_len;
1171
1172 } else {
1173 /*
1174 * this fragment is not first, the headers
1175 * space is regarded as data space.
1176 */
1177 *mtu = orig_mtu;
1178 }
1179 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1180 + fragheaderlen - sizeof(struct frag_hdr);
1181 }
1182}
1183
1184static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1185 struct inet6_cork *v6_cork,
1186 int hlimit, int tclass, struct ipv6_txoptions *opt,
1187 struct rt6_info *rt, struct flowi6 *fl6)
1188{
1189 struct ipv6_pinfo *np = inet6_sk(sk);
1190 unsigned int mtu;
1191
1192 /*
1193 * setup for corking
1194 */
1195 if (opt) {
1196 if (WARN_ON(v6_cork->opt))
1197 return -EINVAL;
1198
1199 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1200 if (unlikely(!v6_cork->opt))
1201 return -ENOBUFS;
1202
1203 v6_cork->opt->tot_len = opt->tot_len;
1204 v6_cork->opt->opt_flen = opt->opt_flen;
1205 v6_cork->opt->opt_nflen = opt->opt_nflen;
1206
1207 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1208 sk->sk_allocation);
1209 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1210 return -ENOBUFS;
1211
1212 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1213 sk->sk_allocation);
1214 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1215 return -ENOBUFS;
1216
1217 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1218 sk->sk_allocation);
1219 if (opt->hopopt && !v6_cork->opt->hopopt)
1220 return -ENOBUFS;
1221
1222 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1223 sk->sk_allocation);
1224 if (opt->srcrt && !v6_cork->opt->srcrt)
1225 return -ENOBUFS;
1226
1227 /* need source address above miyazawa*/
1228 }
1229 dst_hold(&rt->dst);
1230 cork->base.dst = &rt->dst;
1231 cork->fl.u.ip6 = *fl6;
1232 v6_cork->hop_limit = hlimit;
1233 v6_cork->tclass = tclass;
1234 if (rt->dst.flags & DST_XFRM_TUNNEL)
1235 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1236 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1237 else
1238 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1239 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1240 if (np->frag_size < mtu) {
1241 if (np->frag_size)
1242 mtu = np->frag_size;
1243 }
1244 cork->base.fragsize = mtu;
1245 if (dst_allfrag(rt->dst.path))
1246 cork->base.flags |= IPCORK_ALLFRAG;
1247 cork->base.length = 0;
1248
1249 return 0;
1250}
1251
1252static int __ip6_append_data(struct sock *sk,
1253 struct flowi6 *fl6,
1254 struct sk_buff_head *queue,
1255 struct inet_cork *cork,
1256 struct inet6_cork *v6_cork,
1257 struct page_frag *pfrag,
1258 int getfrag(void *from, char *to, int offset,
1259 int len, int odd, struct sk_buff *skb),
1260 void *from, int length, int transhdrlen,
1261 unsigned int flags, int dontfrag)
1262{
1263 struct sk_buff *skb, *skb_prev = NULL;
1264 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1265 int exthdrlen = 0;
1266 int dst_exthdrlen = 0;
1267 int hh_len;
1268 int copy;
1269 int err;
1270 int offset = 0;
1271 __u8 tx_flags = 0;
1272 u32 tskey = 0;
1273 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1274 struct ipv6_txoptions *opt = v6_cork->opt;
1275 int csummode = CHECKSUM_NONE;
1276 unsigned int maxnonfragsize, headersize;
1277
1278 skb = skb_peek_tail(queue);
1279 if (!skb) {
1280 exthdrlen = opt ? opt->opt_flen : 0;
1281 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1282 }
1283
1284 mtu = cork->fragsize;
1285 orig_mtu = mtu;
1286
1287 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1288
1289 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1290 (opt ? opt->opt_nflen : 0);
1291 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1292 sizeof(struct frag_hdr);
1293
1294 headersize = sizeof(struct ipv6hdr) +
1295 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1296 (dst_allfrag(&rt->dst) ?
1297 sizeof(struct frag_hdr) : 0) +
1298 rt->rt6i_nfheader_len;
1299
1300 if (cork->length + length > mtu - headersize && dontfrag &&
1301 (sk->sk_protocol == IPPROTO_UDP ||
1302 sk->sk_protocol == IPPROTO_RAW)) {
1303 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1304 sizeof(struct ipv6hdr));
1305 goto emsgsize;
1306 }
1307
1308 if (ip6_sk_ignore_df(sk))
1309 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1310 else
1311 maxnonfragsize = mtu;
1312
1313 if (cork->length + length > maxnonfragsize - headersize) {
1314emsgsize:
1315 ipv6_local_error(sk, EMSGSIZE, fl6,
1316 mtu - headersize +
1317 sizeof(struct ipv6hdr));
1318 return -EMSGSIZE;
1319 }
1320
1321 /* CHECKSUM_PARTIAL only with no extension headers and when
1322 * we are not going to fragment
1323 */
1324 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1325 headersize == sizeof(struct ipv6hdr) &&
1326 length < mtu - headersize &&
1327 !(flags & MSG_MORE) &&
1328 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1329 csummode = CHECKSUM_PARTIAL;
1330
1331 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1332 sock_tx_timestamp(sk, &tx_flags);
1333 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1334 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1335 tskey = sk->sk_tskey++;
1336 }
1337
1338 /*
1339 * Let's try using as much space as possible.
1340 * Use MTU if total length of the message fits into the MTU.
1341 * Otherwise, we need to reserve fragment header and
1342 * fragment alignment (= 8-15 octects, in total).
1343 *
1344 * Note that we may need to "move" the data from the tail of
1345 * of the buffer to the new fragment when we split
1346 * the message.
1347 *
1348 * FIXME: It may be fragmented into multiple chunks
1349 * at once if non-fragmentable extension headers
1350 * are too large.
1351 * --yoshfuji
1352 */
1353
1354 cork->length += length;
1355 if (((length > mtu) ||
1356 (skb && skb_is_gso(skb))) &&
1357 (sk->sk_protocol == IPPROTO_UDP) &&
1358 (rt->dst.dev->features & NETIF_F_UFO) &&
1359 (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) {
1360 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1361 hh_len, fragheaderlen, exthdrlen,
1362 transhdrlen, mtu, flags, fl6);
1363 if (err)
1364 goto error;
1365 return 0;
1366 }
1367
1368 if (!skb)
1369 goto alloc_new_skb;
1370
1371 while (length > 0) {
1372 /* Check if the remaining data fits into current packet. */
1373 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1374 if (copy < length)
1375 copy = maxfraglen - skb->len;
1376
1377 if (copy <= 0) {
1378 char *data;
1379 unsigned int datalen;
1380 unsigned int fraglen;
1381 unsigned int fraggap;
1382 unsigned int alloclen;
1383alloc_new_skb:
1384 /* There's no room in the current skb */
1385 if (skb)
1386 fraggap = skb->len - maxfraglen;
1387 else
1388 fraggap = 0;
1389 /* update mtu and maxfraglen if necessary */
1390 if (!skb || !skb_prev)
1391 ip6_append_data_mtu(&mtu, &maxfraglen,
1392 fragheaderlen, skb, rt,
1393 orig_mtu);
1394
1395 skb_prev = skb;
1396
1397 /*
1398 * If remaining data exceeds the mtu,
1399 * we know we need more fragment(s).
1400 */
1401 datalen = length + fraggap;
1402
1403 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1404 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1405 if ((flags & MSG_MORE) &&
1406 !(rt->dst.dev->features&NETIF_F_SG))
1407 alloclen = mtu;
1408 else
1409 alloclen = datalen + fragheaderlen;
1410
1411 alloclen += dst_exthdrlen;
1412
1413 if (datalen != length + fraggap) {
1414 /*
1415 * this is not the last fragment, the trailer
1416 * space is regarded as data space.
1417 */
1418 datalen += rt->dst.trailer_len;
1419 }
1420
1421 alloclen += rt->dst.trailer_len;
1422 fraglen = datalen + fragheaderlen;
1423
1424 /*
1425 * We just reserve space for fragment header.
1426 * Note: this may be overallocation if the message
1427 * (without MSG_MORE) fits into the MTU.
1428 */
1429 alloclen += sizeof(struct frag_hdr);
1430
1431 if (transhdrlen) {
1432 skb = sock_alloc_send_skb(sk,
1433 alloclen + hh_len,
1434 (flags & MSG_DONTWAIT), &err);
1435 } else {
1436 skb = NULL;
1437 if (atomic_read(&sk->sk_wmem_alloc) <=
1438 2 * sk->sk_sndbuf)
1439 skb = sock_wmalloc(sk,
1440 alloclen + hh_len, 1,
1441 sk->sk_allocation);
1442 if (unlikely(!skb))
1443 err = -ENOBUFS;
1444 }
1445 if (!skb)
1446 goto error;
1447 /*
1448 * Fill in the control structures
1449 */
1450 skb->protocol = htons(ETH_P_IPV6);
1451 skb->ip_summed = csummode;
1452 skb->csum = 0;
1453 /* reserve for fragmentation and ipsec header */
1454 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1455 dst_exthdrlen);
1456
1457 /* Only the initial fragment is time stamped */
1458 skb_shinfo(skb)->tx_flags = tx_flags;
1459 tx_flags = 0;
1460 skb_shinfo(skb)->tskey = tskey;
1461 tskey = 0;
1462
1463 /*
1464 * Find where to start putting bytes
1465 */
1466 data = skb_put(skb, fraglen);
1467 skb_set_network_header(skb, exthdrlen);
1468 data += fragheaderlen;
1469 skb->transport_header = (skb->network_header +
1470 fragheaderlen);
1471 if (fraggap) {
1472 skb->csum = skb_copy_and_csum_bits(
1473 skb_prev, maxfraglen,
1474 data + transhdrlen, fraggap, 0);
1475 skb_prev->csum = csum_sub(skb_prev->csum,
1476 skb->csum);
1477 data += fraggap;
1478 pskb_trim_unique(skb_prev, maxfraglen);
1479 }
1480 copy = datalen - transhdrlen - fraggap;
1481
1482 if (copy < 0) {
1483 err = -EINVAL;
1484 kfree_skb(skb);
1485 goto error;
1486 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1487 err = -EFAULT;
1488 kfree_skb(skb);
1489 goto error;
1490 }
1491
1492 offset += copy;
1493 length -= datalen - fraggap;
1494 transhdrlen = 0;
1495 exthdrlen = 0;
1496 dst_exthdrlen = 0;
1497
1498 /*
1499 * Put the packet on the pending queue
1500 */
1501 __skb_queue_tail(queue, skb);
1502 continue;
1503 }
1504
1505 if (copy > length)
1506 copy = length;
1507
1508 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1509 unsigned int off;
1510
1511 off = skb->len;
1512 if (getfrag(from, skb_put(skb, copy),
1513 offset, copy, off, skb) < 0) {
1514 __skb_trim(skb, off);
1515 err = -EFAULT;
1516 goto error;
1517 }
1518 } else {
1519 int i = skb_shinfo(skb)->nr_frags;
1520
1521 err = -ENOMEM;
1522 if (!sk_page_frag_refill(sk, pfrag))
1523 goto error;
1524
1525 if (!skb_can_coalesce(skb, i, pfrag->page,
1526 pfrag->offset)) {
1527 err = -EMSGSIZE;
1528 if (i == MAX_SKB_FRAGS)
1529 goto error;
1530
1531 __skb_fill_page_desc(skb, i, pfrag->page,
1532 pfrag->offset, 0);
1533 skb_shinfo(skb)->nr_frags = ++i;
1534 get_page(pfrag->page);
1535 }
1536 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1537 if (getfrag(from,
1538 page_address(pfrag->page) + pfrag->offset,
1539 offset, copy, skb->len, skb) < 0)
1540 goto error_efault;
1541
1542 pfrag->offset += copy;
1543 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1544 skb->len += copy;
1545 skb->data_len += copy;
1546 skb->truesize += copy;
1547 atomic_add(copy, &sk->sk_wmem_alloc);
1548 }
1549 offset += copy;
1550 length -= copy;
1551 }
1552
1553 return 0;
1554
1555error_efault:
1556 err = -EFAULT;
1557error:
1558 cork->length -= length;
1559 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1560 return err;
1561}
1562
1563int ip6_append_data(struct sock *sk,
1564 int getfrag(void *from, char *to, int offset, int len,
1565 int odd, struct sk_buff *skb),
1566 void *from, int length, int transhdrlen, int hlimit,
1567 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1568 struct rt6_info *rt, unsigned int flags, int dontfrag)
1569{
1570 struct inet_sock *inet = inet_sk(sk);
1571 struct ipv6_pinfo *np = inet6_sk(sk);
1572 int exthdrlen;
1573 int err;
1574
1575 if (flags&MSG_PROBE)
1576 return 0;
1577 if (skb_queue_empty(&sk->sk_write_queue)) {
1578 /*
1579 * setup for corking
1580 */
1581 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1582 tclass, opt, rt, fl6);
1583 if (err)
1584 return err;
1585
1586 exthdrlen = (opt ? opt->opt_flen : 0);
1587 length += exthdrlen;
1588 transhdrlen += exthdrlen;
1589 } else {
1590 fl6 = &inet->cork.fl.u.ip6;
1591 transhdrlen = 0;
1592 }
1593
1594 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1595 &np->cork, sk_page_frag(sk), getfrag,
1596 from, length, transhdrlen, flags, dontfrag);
1597}
1598EXPORT_SYMBOL_GPL(ip6_append_data);
1599
1600static void ip6_cork_release(struct inet_cork_full *cork,
1601 struct inet6_cork *v6_cork)
1602{
1603 if (v6_cork->opt) {
1604 kfree(v6_cork->opt->dst0opt);
1605 kfree(v6_cork->opt->dst1opt);
1606 kfree(v6_cork->opt->hopopt);
1607 kfree(v6_cork->opt->srcrt);
1608 kfree(v6_cork->opt);
1609 v6_cork->opt = NULL;
1610 }
1611
1612 if (cork->base.dst) {
1613 dst_release(cork->base.dst);
1614 cork->base.dst = NULL;
1615 cork->base.flags &= ~IPCORK_ALLFRAG;
1616 }
1617 memset(&cork->fl, 0, sizeof(cork->fl));
1618}
1619
1620struct sk_buff *__ip6_make_skb(struct sock *sk,
1621 struct sk_buff_head *queue,
1622 struct inet_cork_full *cork,
1623 struct inet6_cork *v6_cork)
1624{
1625 struct sk_buff *skb, *tmp_skb;
1626 struct sk_buff **tail_skb;
1627 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1628 struct ipv6_pinfo *np = inet6_sk(sk);
1629 struct net *net = sock_net(sk);
1630 struct ipv6hdr *hdr;
1631 struct ipv6_txoptions *opt = v6_cork->opt;
1632 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1633 struct flowi6 *fl6 = &cork->fl.u.ip6;
1634 unsigned char proto = fl6->flowi6_proto;
1635
1636 skb = __skb_dequeue(queue);
1637 if (!skb)
1638 goto out;
1639 tail_skb = &(skb_shinfo(skb)->frag_list);
1640
1641 /* move skb->data to ip header from ext header */
1642 if (skb->data < skb_network_header(skb))
1643 __skb_pull(skb, skb_network_offset(skb));
1644 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1645 __skb_pull(tmp_skb, skb_network_header_len(skb));
1646 *tail_skb = tmp_skb;
1647 tail_skb = &(tmp_skb->next);
1648 skb->len += tmp_skb->len;
1649 skb->data_len += tmp_skb->len;
1650 skb->truesize += tmp_skb->truesize;
1651 tmp_skb->destructor = NULL;
1652 tmp_skb->sk = NULL;
1653 }
1654
1655 /* Allow local fragmentation. */
1656 skb->ignore_df = ip6_sk_ignore_df(sk);
1657
1658 *final_dst = fl6->daddr;
1659 __skb_pull(skb, skb_network_header_len(skb));
1660 if (opt && opt->opt_flen)
1661 ipv6_push_frag_opts(skb, opt, &proto);
1662 if (opt && opt->opt_nflen)
1663 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1664
1665 skb_push(skb, sizeof(struct ipv6hdr));
1666 skb_reset_network_header(skb);
1667 hdr = ipv6_hdr(skb);
1668
1669 ip6_flow_hdr(hdr, v6_cork->tclass,
1670 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1671 np->autoflowlabel, fl6));
1672 hdr->hop_limit = v6_cork->hop_limit;
1673 hdr->nexthdr = proto;
1674 hdr->saddr = fl6->saddr;
1675 hdr->daddr = *final_dst;
1676
1677 skb->priority = sk->sk_priority;
1678 skb->mark = sk->sk_mark;
1679
1680 skb_dst_set(skb, dst_clone(&rt->dst));
1681 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1682 if (proto == IPPROTO_ICMPV6) {
1683 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1684
1685 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1686 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1687 }
1688
1689 ip6_cork_release(cork, v6_cork);
1690out:
1691 return skb;
1692}
1693
1694int ip6_send_skb(struct sk_buff *skb)
1695{
1696 struct net *net = sock_net(skb->sk);
1697 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1698 int err;
1699
1700 err = ip6_local_out(net, skb->sk, skb);
1701 if (err) {
1702 if (err > 0)
1703 err = net_xmit_errno(err);
1704 if (err)
1705 IP6_INC_STATS(net, rt->rt6i_idev,
1706 IPSTATS_MIB_OUTDISCARDS);
1707 }
1708
1709 return err;
1710}
1711
1712int ip6_push_pending_frames(struct sock *sk)
1713{
1714 struct sk_buff *skb;
1715
1716 skb = ip6_finish_skb(sk);
1717 if (!skb)
1718 return 0;
1719
1720 return ip6_send_skb(skb);
1721}
1722EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1723
1724static void __ip6_flush_pending_frames(struct sock *sk,
1725 struct sk_buff_head *queue,
1726 struct inet_cork_full *cork,
1727 struct inet6_cork *v6_cork)
1728{
1729 struct sk_buff *skb;
1730
1731 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1732 if (skb_dst(skb))
1733 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1734 IPSTATS_MIB_OUTDISCARDS);
1735 kfree_skb(skb);
1736 }
1737
1738 ip6_cork_release(cork, v6_cork);
1739}
1740
1741void ip6_flush_pending_frames(struct sock *sk)
1742{
1743 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1744 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1745}
1746EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1747
1748struct sk_buff *ip6_make_skb(struct sock *sk,
1749 int getfrag(void *from, char *to, int offset,
1750 int len, int odd, struct sk_buff *skb),
1751 void *from, int length, int transhdrlen,
1752 int hlimit, int tclass,
1753 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1754 struct rt6_info *rt, unsigned int flags,
1755 int dontfrag)
1756{
1757 struct inet_cork_full cork;
1758 struct inet6_cork v6_cork;
1759 struct sk_buff_head queue;
1760 int exthdrlen = (opt ? opt->opt_flen : 0);
1761 int err;
1762
1763 if (flags & MSG_PROBE)
1764 return NULL;
1765
1766 __skb_queue_head_init(&queue);
1767
1768 cork.base.flags = 0;
1769 cork.base.addr = 0;
1770 cork.base.opt = NULL;
1771 v6_cork.opt = NULL;
1772 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1773 if (err)
1774 return ERR_PTR(err);
1775
1776 if (dontfrag < 0)
1777 dontfrag = inet6_sk(sk)->dontfrag;
1778
1779 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1780 ¤t->task_frag, getfrag, from,
1781 length + exthdrlen, transhdrlen + exthdrlen,
1782 flags, dontfrag);
1783 if (err) {
1784 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1785 return ERR_PTR(err);
1786 }
1787
1788 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1789}