?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'remotes/powerpc/topic/xive' into kvm-ppc-next
[linux/fpc-iii.git] / net / ipv6 / ip6_output.c
blob58f6288e9ba53e6964b74d71dde7615ead695c06
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/bpf-cgroup.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv6.h>
45
46 #include <net/sock.h>
47 #include <net/snmp.h>
48
49 #include <net/ipv6.h>
50 #include <net/ndisc.h>
51 #include <net/protocol.h>
52 #include <net/ip6_route.h>
53 #include <net/addrconf.h>
54 #include <net/rawv6.h>
55 #include <net/icmp.h>
56 #include <net/xfrm.h>
57 #include <net/checksum.h>
58 #include <linux/mroute6.h>
59 #include <net/l3mdev.h>
60 #include <net/lwtunnel.h>
61
62 static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
63 {
64 struct dst_entry *dst = skb_dst(skb);
65 struct net_device *dev = dst->dev;
66 struct neighbour *neigh;
67 struct in6_addr *nexthop;
68 int ret;
69
70 skb->protocol = htons(ETH_P_IPV6);
71 skb->dev = dev;
72
73 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
74 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
75
76 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
77 ((mroute6_socket(net, skb) &&
78 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
79 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
80 &ipv6_hdr(skb)->saddr))) {
81 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
82
83 /* Do not check for IFF_ALLMULTI; multicast routing
84 is not supported in any case.
85 */
86 if (newskb)
87 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
88 net, sk, newskb, NULL, newskb->dev,
89 dev_loopback_xmit);
90
91 if (ipv6_hdr(skb)->hop_limit == 0) {
92 IP6_INC_STATS(net, idev,
93 IPSTATS_MIB_OUTDISCARDS);
94 kfree_skb(skb);
95 return 0;
96 }
97 }
98
99 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
100
101 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
102 IPV6_ADDR_SCOPE_NODELOCAL &&
103 !(dev->flags & IFF_LOOPBACK)) {
104 kfree_skb(skb);
105 return 0;
106 }
107 }
108
109 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
110 int res = lwtunnel_xmit(skb);
111
112 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
113 return res;
114 }
115
116 rcu_read_lock_bh();
117 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
118 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
119 if (unlikely(!neigh))
120 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
121 if (!IS_ERR(neigh)) {
122 sock_confirm_neigh(skb, neigh);
123 ret = neigh_output(neigh, skb);
124 rcu_read_unlock_bh();
125 return ret;
126 }
127 rcu_read_unlock_bh();
128
129 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
130 kfree_skb(skb);
131 return -EINVAL;
132 }
133
134 static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
135 {
136 int ret;
137
138 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
139 if (ret) {
140 kfree_skb(skb);
141 return ret;
142 }
143
144 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
145 dst_allfrag(skb_dst(skb)) ||
146 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
147 return ip6_fragment(net, sk, skb, ip6_finish_output2);
148 else
149 return ip6_finish_output2(net, sk, skb);
150 }
151
152 int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
153 {
154 struct net_device *dev = skb_dst(skb)->dev;
155 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
156
157 if (unlikely(idev->cnf.disable_ipv6)) {
158 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
159 kfree_skb(skb);
160 return 0;
161 }
162
163 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
164 net, sk, skb, NULL, dev,
165 ip6_finish_output,
166 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
167 }
168
169 /*
170 * xmit an sk_buff (used by TCP, SCTP and DCCP)
171 * Note : socket lock is not held for SYNACK packets, but might be modified
172 * by calls to skb_set_owner_w() and ipv6_local_error(),
173 * which are using proper atomic operations or spinlocks.
174 */
175 int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
176 __u32 mark, struct ipv6_txoptions *opt, int tclass)
177 {
178 struct net *net = sock_net(sk);
179 const struct ipv6_pinfo *np = inet6_sk(sk);
180 struct in6_addr *first_hop = &fl6->daddr;
181 struct dst_entry *dst = skb_dst(skb);
182 struct ipv6hdr *hdr;
183 u8 proto = fl6->flowi6_proto;
184 int seg_len = skb->len;
185 int hlimit = -1;
186 u32 mtu;
187
188 if (opt) {
189 unsigned int head_room;
190
191 /* First: exthdrs may take lots of space (~8K for now)
192 MAX_HEADER is not enough.
193 */
194 head_room = opt->opt_nflen + opt->opt_flen;
195 seg_len += head_room;
196 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
197
198 if (skb_headroom(skb) < head_room) {
199 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
200 if (!skb2) {
201 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
202 IPSTATS_MIB_OUTDISCARDS);
203 kfree_skb(skb);
204 return -ENOBUFS;
205 }
206 consume_skb(skb);
207 skb = skb2;
208 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
209 * it is safe to call in our context (socket lock not held)
210 */
211 skb_set_owner_w(skb, (struct sock *)sk);
212 }
213 if (opt->opt_flen)
214 ipv6_push_frag_opts(skb, opt, &proto);
215 if (opt->opt_nflen)
216 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
217 &fl6->saddr);
218 }
219
220 skb_push(skb, sizeof(struct ipv6hdr));
221 skb_reset_network_header(skb);
222 hdr = ipv6_hdr(skb);
223
224 /*
225 * Fill in the IPv6 header
226 */
227 if (np)
228 hlimit = np->hop_limit;
229 if (hlimit < 0)
230 hlimit = ip6_dst_hoplimit(dst);
231
232 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
233 np->autoflowlabel, fl6));
234
235 hdr->payload_len = htons(seg_len);
236 hdr->nexthdr = proto;
237 hdr->hop_limit = hlimit;
238
239 hdr->saddr = fl6->saddr;
240 hdr->daddr = *first_hop;
241
242 skb->protocol = htons(ETH_P_IPV6);
243 skb->priority = sk->sk_priority;
244 skb->mark = mark;
245
246 mtu = dst_mtu(dst);
247 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
248 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
249 IPSTATS_MIB_OUT, skb->len);
250
251 /* if egress device is enslaved to an L3 master device pass the
252 * skb to its handler for processing
253 */
254 skb = l3mdev_ip6_out((struct sock *)sk, skb);
255 if (unlikely(!skb))
256 return 0;
257
258 /* hooks should never assume socket lock is held.
259 * we promote our socket to non const
260 */
261 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
262 net, (struct sock *)sk, skb, NULL, dst->dev,
263 dst_output);
264 }
265
266 skb->dev = dst->dev;
267 /* ipv6_local_error() does not require socket lock,
268 * we promote our socket to non const
269 */
270 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
271
272 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
273 kfree_skb(skb);
274 return -EMSGSIZE;
275 }
276 EXPORT_SYMBOL(ip6_xmit);
277
278 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
279 {
280 struct ip6_ra_chain *ra;
281 struct sock *last = NULL;
282
283 read_lock(&ip6_ra_lock);
284 for (ra = ip6_ra_chain; ra; ra = ra->next) {
285 struct sock *sk = ra->sk;
286 if (sk && ra->sel == sel &&
287 (!sk->sk_bound_dev_if ||
288 sk->sk_bound_dev_if == skb->dev->ifindex)) {
289 if (last) {
290 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
291 if (skb2)
292 rawv6_rcv(last, skb2);
293 }
294 last = sk;
295 }
296 }
297
298 if (last) {
299 rawv6_rcv(last, skb);
300 read_unlock(&ip6_ra_lock);
301 return 1;
302 }
303 read_unlock(&ip6_ra_lock);
304 return 0;
305 }
306
307 static int ip6_forward_proxy_check(struct sk_buff *skb)
308 {
309 struct ipv6hdr *hdr = ipv6_hdr(skb);
310 u8 nexthdr = hdr->nexthdr;
311 __be16 frag_off;
312 int offset;
313
314 if (ipv6_ext_hdr(nexthdr)) {
315 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
316 if (offset < 0)
317 return 0;
318 } else
319 offset = sizeof(struct ipv6hdr);
320
321 if (nexthdr == IPPROTO_ICMPV6) {
322 struct icmp6hdr *icmp6;
323
324 if (!pskb_may_pull(skb, (skb_network_header(skb) +
325 offset + 1 - skb->data)))
326 return 0;
327
328 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
329
330 switch (icmp6->icmp6_type) {
331 case NDISC_ROUTER_SOLICITATION:
332 case NDISC_ROUTER_ADVERTISEMENT:
333 case NDISC_NEIGHBOUR_SOLICITATION:
334 case NDISC_NEIGHBOUR_ADVERTISEMENT:
335 case NDISC_REDIRECT:
336 /* For reaction involving unicast neighbor discovery
337 * message destined to the proxied address, pass it to
338 * input function.
339 */
340 return 1;
341 default:
342 break;
343 }
344 }
345
346 /*
347 * The proxying router can't forward traffic sent to a link-local
348 * address, so signal the sender and discard the packet. This
349 * behavior is clarified by the MIPv6 specification.
350 */
351 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
352 dst_link_failure(skb);
353 return -1;
354 }
355
356 return 0;
357 }
358
359 static inline int ip6_forward_finish(struct net *net, struct sock *sk,
360 struct sk_buff *skb)
361 {
362 return dst_output(net, sk, skb);
363 }
364
365 static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
366 {
367 unsigned int mtu;
368 struct inet6_dev *idev;
369
370 if (dst_metric_locked(dst, RTAX_MTU)) {
371 mtu = dst_metric_raw(dst, RTAX_MTU);
372 if (mtu)
373 return mtu;
374 }
375
376 mtu = IPV6_MIN_MTU;
377 rcu_read_lock();
378 idev = __in6_dev_get(dst->dev);
379 if (idev)
380 mtu = idev->cnf.mtu6;
381 rcu_read_unlock();
382
383 return mtu;
384 }
385
386 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
387 {
388 if (skb->len <= mtu)
389 return false;
390
391 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
392 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
393 return true;
394
395 if (skb->ignore_df)
396 return false;
397
398 if (skb_is_gso(skb) && skb_gso_validate_mtu(skb, mtu))
399 return false;
400
401 return true;
402 }
403
404 int ip6_forward(struct sk_buff *skb)
405 {
406 struct dst_entry *dst = skb_dst(skb);
407 struct ipv6hdr *hdr = ipv6_hdr(skb);
408 struct inet6_skb_parm *opt = IP6CB(skb);
409 struct net *net = dev_net(dst->dev);
410 u32 mtu;
411
412 if (net->ipv6.devconf_all->forwarding == 0)
413 goto error;
414
415 if (skb->pkt_type != PACKET_HOST)
416 goto drop;
417
418 if (unlikely(skb->sk))
419 goto drop;
420
421 if (skb_warn_if_lro(skb))
422 goto drop;
423
424 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
425 __IP6_INC_STATS(net, ip6_dst_idev(dst),
426 IPSTATS_MIB_INDISCARDS);
427 goto drop;
428 }
429
430 skb_forward_csum(skb);
431
432 /*
433 * We DO NOT make any processing on
434 * RA packets, pushing them to user level AS IS
435 * without ane WARRANTY that application will be able
436 * to interpret them. The reason is that we
437 * cannot make anything clever here.
438 *
439 * We are not end-node, so that if packet contains
440 * AH/ESP, we cannot make anything.
441 * Defragmentation also would be mistake, RA packets
442 * cannot be fragmented, because there is no warranty
443 * that different fragments will go along one path. --ANK
444 */
445 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
446 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
447 return 0;
448 }
449
450 /*
451 * check and decrement ttl
452 */
453 if (hdr->hop_limit <= 1) {
454 /* Force OUTPUT device used as source address */
455 skb->dev = dst->dev;
456 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
457 __IP6_INC_STATS(net, ip6_dst_idev(dst),
458 IPSTATS_MIB_INHDRERRORS);
459
460 kfree_skb(skb);
461 return -ETIMEDOUT;
462 }
463
464 /* XXX: idev->cnf.proxy_ndp? */
465 if (net->ipv6.devconf_all->proxy_ndp &&
466 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
467 int proxied = ip6_forward_proxy_check(skb);
468 if (proxied > 0)
469 return ip6_input(skb);
470 else if (proxied < 0) {
471 __IP6_INC_STATS(net, ip6_dst_idev(dst),
472 IPSTATS_MIB_INDISCARDS);
473 goto drop;
474 }
475 }
476
477 if (!xfrm6_route_forward(skb)) {
478 __IP6_INC_STATS(net, ip6_dst_idev(dst),
479 IPSTATS_MIB_INDISCARDS);
480 goto drop;
481 }
482 dst = skb_dst(skb);
483
484 /* IPv6 specs say nothing about it, but it is clear that we cannot
485 send redirects to source routed frames.
486 We don't send redirects to frames decapsulated from IPsec.
487 */
488 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
489 struct in6_addr *target = NULL;
490 struct inet_peer *peer;
491 struct rt6_info *rt;
492
493 /*
494 * incoming and outgoing devices are the same
495 * send a redirect.
496 */
497
498 rt = (struct rt6_info *) dst;
499 if (rt->rt6i_flags & RTF_GATEWAY)
500 target = &rt->rt6i_gateway;
501 else
502 target = &hdr->daddr;
503
504 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
505
506 /* Limit redirects both by destination (here)
507 and by source (inside ndisc_send_redirect)
508 */
509 if (inet_peer_xrlim_allow(peer, 1*HZ))
510 ndisc_send_redirect(skb, target);
511 if (peer)
512 inet_putpeer(peer);
513 } else {
514 int addrtype = ipv6_addr_type(&hdr->saddr);
515
516 /* This check is security critical. */
517 if (addrtype == IPV6_ADDR_ANY ||
518 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
519 goto error;
520 if (addrtype & IPV6_ADDR_LINKLOCAL) {
521 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
522 ICMPV6_NOT_NEIGHBOUR, 0);
523 goto error;
524 }
525 }
526
527 mtu = ip6_dst_mtu_forward(dst);
528 if (mtu < IPV6_MIN_MTU)
529 mtu = IPV6_MIN_MTU;
530
531 if (ip6_pkt_too_big(skb, mtu)) {
532 /* Again, force OUTPUT device used as source address */
533 skb->dev = dst->dev;
534 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
535 __IP6_INC_STATS(net, ip6_dst_idev(dst),
536 IPSTATS_MIB_INTOOBIGERRORS);
537 __IP6_INC_STATS(net, ip6_dst_idev(dst),
538 IPSTATS_MIB_FRAGFAILS);
539 kfree_skb(skb);
540 return -EMSGSIZE;
541 }
542
543 if (skb_cow(skb, dst->dev->hard_header_len)) {
544 __IP6_INC_STATS(net, ip6_dst_idev(dst),
545 IPSTATS_MIB_OUTDISCARDS);
546 goto drop;
547 }
548
549 hdr = ipv6_hdr(skb);
550
551 /* Mangling hops number delayed to point after skb COW */
552
553 hdr->hop_limit--;
554
555 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
556 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
557 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
558 net, NULL, skb, skb->dev, dst->dev,
559 ip6_forward_finish);
560
561 error:
562 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
563 drop:
564 kfree_skb(skb);
565 return -EINVAL;
566 }
567
568 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
569 {
570 to->pkt_type = from->pkt_type;
571 to->priority = from->priority;
572 to->protocol = from->protocol;
573 skb_dst_drop(to);
574 skb_dst_set(to, dst_clone(skb_dst(from)));
575 to->dev = from->dev;
576 to->mark = from->mark;
577
578 #ifdef CONFIG_NET_SCHED
579 to->tc_index = from->tc_index;
580 #endif
581 nf_copy(to, from);
582 skb_copy_secmark(to, from);
583 }
584
585 int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
586 int (*output)(struct net *, struct sock *, struct sk_buff *))
587 {
588 struct sk_buff *frag;
589 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
590 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
591 inet6_sk(skb->sk) : NULL;
592 struct ipv6hdr *tmp_hdr;
593 struct frag_hdr *fh;
594 unsigned int mtu, hlen, left, len;
595 int hroom, troom;
596 __be32 frag_id;
597 int ptr, offset = 0, err = 0;
598 u8 *prevhdr, nexthdr = 0;
599
600 hlen = ip6_find_1stfragopt(skb, &prevhdr);
601 nexthdr = *prevhdr;
602
603 mtu = ip6_skb_dst_mtu(skb);
604
605 /* We must not fragment if the socket is set to force MTU discovery
606 * or if the skb it not generated by a local socket.
607 */
608 if (unlikely(!skb->ignore_df && skb->len > mtu))
609 goto fail_toobig;
610
611 if (IP6CB(skb)->frag_max_size) {
612 if (IP6CB(skb)->frag_max_size > mtu)
613 goto fail_toobig;
614
615 /* don't send fragments larger than what we received */
616 mtu = IP6CB(skb)->frag_max_size;
617 if (mtu < IPV6_MIN_MTU)
618 mtu = IPV6_MIN_MTU;
619 }
620
621 if (np && np->frag_size < mtu) {
622 if (np->frag_size)
623 mtu = np->frag_size;
624 }
625 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
626 goto fail_toobig;
627 mtu -= hlen + sizeof(struct frag_hdr);
628
629 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
630 &ipv6_hdr(skb)->saddr);
631
632 if (skb->ip_summed == CHECKSUM_PARTIAL &&
633 (err = skb_checksum_help(skb)))
634 goto fail;
635
636 hroom = LL_RESERVED_SPACE(rt->dst.dev);
637 if (skb_has_frag_list(skb)) {
638 unsigned int first_len = skb_pagelen(skb);
639 struct sk_buff *frag2;
640
641 if (first_len - hlen > mtu ||
642 ((first_len - hlen) & 7) ||
643 skb_cloned(skb) ||
644 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
645 goto slow_path;
646
647 skb_walk_frags(skb, frag) {
648 /* Correct geometry. */
649 if (frag->len > mtu ||
650 ((frag->len & 7) && frag->next) ||
651 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
652 goto slow_path_clean;
653
654 /* Partially cloned skb? */
655 if (skb_shared(frag))
656 goto slow_path_clean;
657
658 BUG_ON(frag->sk);
659 if (skb->sk) {
660 frag->sk = skb->sk;
661 frag->destructor = sock_wfree;
662 }
663 skb->truesize -= frag->truesize;
664 }
665
666 err = 0;
667 offset = 0;
668 /* BUILD HEADER */
669
670 *prevhdr = NEXTHDR_FRAGMENT;
671 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
672 if (!tmp_hdr) {
673 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
674 IPSTATS_MIB_FRAGFAILS);
675 err = -ENOMEM;
676 goto fail;
677 }
678 frag = skb_shinfo(skb)->frag_list;
679 skb_frag_list_init(skb);
680
681 __skb_pull(skb, hlen);
682 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
683 __skb_push(skb, hlen);
684 skb_reset_network_header(skb);
685 memcpy(skb_network_header(skb), tmp_hdr, hlen);
686
687 fh->nexthdr = nexthdr;
688 fh->reserved = 0;
689 fh->frag_off = htons(IP6_MF);
690 fh->identification = frag_id;
691
692 first_len = skb_pagelen(skb);
693 skb->data_len = first_len - skb_headlen(skb);
694 skb->len = first_len;
695 ipv6_hdr(skb)->payload_len = htons(first_len -
696 sizeof(struct ipv6hdr));
697
698 dst_hold(&rt->dst);
699
700 for (;;) {
701 /* Prepare header of the next frame,
702 * before previous one went down. */
703 if (frag) {
704 frag->ip_summed = CHECKSUM_NONE;
705 skb_reset_transport_header(frag);
706 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
707 __skb_push(frag, hlen);
708 skb_reset_network_header(frag);
709 memcpy(skb_network_header(frag), tmp_hdr,
710 hlen);
711 offset += skb->len - hlen - sizeof(struct frag_hdr);
712 fh->nexthdr = nexthdr;
713 fh->reserved = 0;
714 fh->frag_off = htons(offset);
715 if (frag->next)
716 fh->frag_off |= htons(IP6_MF);
717 fh->identification = frag_id;
718 ipv6_hdr(frag)->payload_len =
719 htons(frag->len -
720 sizeof(struct ipv6hdr));
721 ip6_copy_metadata(frag, skb);
722 }
723
724 err = output(net, sk, skb);
725 if (!err)
726 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
727 IPSTATS_MIB_FRAGCREATES);
728
729 if (err || !frag)
730 break;
731
732 skb = frag;
733 frag = skb->next;
734 skb->next = NULL;
735 }
736
737 kfree(tmp_hdr);
738
739 if (err == 0) {
740 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
741 IPSTATS_MIB_FRAGOKS);
742 ip6_rt_put(rt);
743 return 0;
744 }
745
746 kfree_skb_list(frag);
747
748 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
749 IPSTATS_MIB_FRAGFAILS);
750 ip6_rt_put(rt);
751 return err;
752
753 slow_path_clean:
754 skb_walk_frags(skb, frag2) {
755 if (frag2 == frag)
756 break;
757 frag2->sk = NULL;
758 frag2->destructor = NULL;
759 skb->truesize += frag2->truesize;
760 }
761 }
762
763 slow_path:
764 left = skb->len - hlen; /* Space per frame */
765 ptr = hlen; /* Where to start from */
766
767 /*
768 * Fragment the datagram.
769 */
770
771 troom = rt->dst.dev->needed_tailroom;
772
773 /*
774 * Keep copying data until we run out.
775 */
776 while (left > 0) {
777 u8 *fragnexthdr_offset;
778
779 len = left;
780 /* IF: it doesn't fit, use 'mtu' - the data space left */
781 if (len > mtu)
782 len = mtu;
783 /* IF: we are not sending up to and including the packet end
784 then align the next start on an eight byte boundary */
785 if (len < left) {
786 len &= ~7;
787 }
788
789 /* Allocate buffer */
790 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
791 hroom + troom, GFP_ATOMIC);
792 if (!frag) {
793 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
794 IPSTATS_MIB_FRAGFAILS);
795 err = -ENOMEM;
796 goto fail;
797 }
798
799 /*
800 * Set up data on packet
801 */
802
803 ip6_copy_metadata(frag, skb);
804 skb_reserve(frag, hroom);
805 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
806 skb_reset_network_header(frag);
807 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
808 frag->transport_header = (frag->network_header + hlen +
809 sizeof(struct frag_hdr));
810
811 /*
812 * Charge the memory for the fragment to any owner
813 * it might possess
814 */
815 if (skb->sk)
816 skb_set_owner_w(frag, skb->sk);
817
818 /*
819 * Copy the packet header into the new buffer.
820 */
821 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
822
823 fragnexthdr_offset = skb_network_header(frag);
824 fragnexthdr_offset += prevhdr - skb_network_header(skb);
825 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
826
827 /*
828 * Build fragment header.
829 */
830 fh->nexthdr = nexthdr;
831 fh->reserved = 0;
832 fh->identification = frag_id;
833
834 /*
835 * Copy a block of the IP datagram.
836 */
837 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
838 len));
839 left -= len;
840
841 fh->frag_off = htons(offset);
842 if (left > 0)
843 fh->frag_off |= htons(IP6_MF);
844 ipv6_hdr(frag)->payload_len = htons(frag->len -
845 sizeof(struct ipv6hdr));
846
847 ptr += len;
848 offset += len;
849
850 /*
851 * Put this fragment into the sending queue.
852 */
853 err = output(net, sk, frag);
854 if (err)
855 goto fail;
856
857 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
858 IPSTATS_MIB_FRAGCREATES);
859 }
860 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
861 IPSTATS_MIB_FRAGOKS);
862 consume_skb(skb);
863 return err;
864
865 fail_toobig:
866 if (skb->sk && dst_allfrag(skb_dst(skb)))
867 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
868
869 skb->dev = skb_dst(skb)->dev;
870 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
871 err = -EMSGSIZE;
872
873 fail:
874 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
875 IPSTATS_MIB_FRAGFAILS);
876 kfree_skb(skb);
877 return err;
878 }
879
880 static inline int ip6_rt_check(const struct rt6key *rt_key,
881 const struct in6_addr *fl_addr,
882 const struct in6_addr *addr_cache)
883 {
884 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
885 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
886 }
887
888 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
889 struct dst_entry *dst,
890 const struct flowi6 *fl6)
891 {
892 struct ipv6_pinfo *np = inet6_sk(sk);
893 struct rt6_info *rt;
894
895 if (!dst)
896 goto out;
897
898 if (dst->ops->family != AF_INET6) {
899 dst_release(dst);
900 return NULL;
901 }
902
903 rt = (struct rt6_info *)dst;
904 /* Yes, checking route validity in not connected
905 * case is not very simple. Take into account,
906 * that we do not support routing by source, TOS,
907 * and MSG_DONTROUTE --ANK (980726)
908 *
909 * 1. ip6_rt_check(): If route was host route,
910 * check that cached destination is current.
911 * If it is network route, we still may
912 * check its validity using saved pointer
913 * to the last used address: daddr_cache.
914 * We do not want to save whole address now,
915 * (because main consumer of this service
916 * is tcp, which has not this problem),
917 * so that the last trick works only on connected
918 * sockets.
919 * 2. oif also should be the same.
920 */
921 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
922 #ifdef CONFIG_IPV6_SUBTREES
923 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
924 #endif
925 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
926 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
927 dst_release(dst);
928 dst = NULL;
929 }
930
931 out:
932 return dst;
933 }
934
935 static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
936 struct dst_entry **dst, struct flowi6 *fl6)
937 {
938 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
939 struct neighbour *n;
940 struct rt6_info *rt;
941 #endif
942 int err;
943 int flags = 0;
944
945 /* The correct way to handle this would be to do
946 * ip6_route_get_saddr, and then ip6_route_output; however,
947 * the route-specific preferred source forces the
948 * ip6_route_output call _before_ ip6_route_get_saddr.
949 *
950 * In source specific routing (no src=any default route),
951 * ip6_route_output will fail given src=any saddr, though, so
952 * that's why we try it again later.
953 */
954 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
955 struct rt6_info *rt;
956 bool had_dst = *dst != NULL;
957
958 if (!had_dst)
959 *dst = ip6_route_output(net, sk, fl6);
960 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
961 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
962 sk ? inet6_sk(sk)->srcprefs : 0,
963 &fl6->saddr);
964 if (err)
965 goto out_err_release;
966
967 /* If we had an erroneous initial result, pretend it
968 * never existed and let the SA-enabled version take
969 * over.
970 */
971 if (!had_dst && (*dst)->error) {
972 dst_release(*dst);
973 *dst = NULL;
974 }
975
976 if (fl6->flowi6_oif)
977 flags |= RT6_LOOKUP_F_IFACE;
978 }
979
980 if (!*dst)
981 *dst = ip6_route_output_flags(net, sk, fl6, flags);
982
983 err = (*dst)->error;
984 if (err)
985 goto out_err_release;
986
987 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
988 /*
989 * Here if the dst entry we've looked up
990 * has a neighbour entry that is in the INCOMPLETE
991 * state and the src address from the flow is
992 * marked as OPTIMISTIC, we release the found
993 * dst entry and replace it instead with the
994 * dst entry of the nexthop router
995 */
996 rt = (struct rt6_info *) *dst;
997 rcu_read_lock_bh();
998 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
999 rt6_nexthop(rt, &fl6->daddr));
1000 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1001 rcu_read_unlock_bh();
1002
1003 if (err) {
1004 struct inet6_ifaddr *ifp;
1005 struct flowi6 fl_gw6;
1006 int redirect;
1007
1008 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1009 (*dst)->dev, 1);
1010
1011 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1012 if (ifp)
1013 in6_ifa_put(ifp);
1014
1015 if (redirect) {
1016 /*
1017 * We need to get the dst entry for the
1018 * default router instead
1019 */
1020 dst_release(*dst);
1021 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1022 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1023 *dst = ip6_route_output(net, sk, &fl_gw6);
1024 err = (*dst)->error;
1025 if (err)
1026 goto out_err_release;
1027 }
1028 }
1029 #endif
1030 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1031 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1032 err = -EAFNOSUPPORT;
1033 goto out_err_release;
1034 }
1035
1036 return 0;
1037
1038 out_err_release:
1039 dst_release(*dst);
1040 *dst = NULL;
1041
1042 if (err == -ENETUNREACH)
1043 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1044 return err;
1045 }
1046
1047 /**
1048 * ip6_dst_lookup - perform route lookup on flow
1049 * @sk: socket which provides route info
1050 * @dst: pointer to dst_entry * for result
1051 * @fl6: flow to lookup
1052 *
1053 * This function performs a route lookup on the given flow.
1054 *
1055 * It returns zero on success, or a standard errno code on error.
1056 */
1057 int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1058 struct flowi6 *fl6)
1059 {
1060 *dst = NULL;
1061 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1062 }
1063 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1064
1065 /**
1066 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1067 * @sk: socket which provides route info
1068 * @fl6: flow to lookup
1069 * @final_dst: final destination address for ipsec lookup
1070 *
1071 * This function performs a route lookup on the given flow.
1072 *
1073 * It returns a valid dst pointer on success, or a pointer encoded
1074 * error code.
1075 */
1076 struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1077 const struct in6_addr *final_dst)
1078 {
1079 struct dst_entry *dst = NULL;
1080 int err;
1081
1082 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1083 if (err)
1084 return ERR_PTR(err);
1085 if (final_dst)
1086 fl6->daddr = *final_dst;
1087
1088 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1089 }
1090 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1091
1092 /**
1093 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1094 * @sk: socket which provides the dst cache and route info
1095 * @fl6: flow to lookup
1096 * @final_dst: final destination address for ipsec lookup
1097 *
1098 * This function performs a route lookup on the given flow with the
1099 * possibility of using the cached route in the socket if it is valid.
1100 * It will take the socket dst lock when operating on the dst cache.
1101 * As a result, this function can only be used in process context.
1102 *
1103 * It returns a valid dst pointer on success, or a pointer encoded
1104 * error code.
1105 */
1106 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1107 const struct in6_addr *final_dst)
1108 {
1109 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1110
1111 dst = ip6_sk_dst_check(sk, dst, fl6);
1112 if (!dst)
1113 dst = ip6_dst_lookup_flow(sk, fl6, final_dst);
1114
1115 return dst;
1116 }
1117 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1118
1119 static inline int ip6_ufo_append_data(struct sock *sk,
1120 struct sk_buff_head *queue,
1121 int getfrag(void *from, char *to, int offset, int len,
1122 int odd, struct sk_buff *skb),
1123 void *from, int length, int hh_len, int fragheaderlen,
1124 int exthdrlen, int transhdrlen, int mtu,
1125 unsigned int flags, const struct flowi6 *fl6)
1126
1127 {
1128 struct sk_buff *skb;
1129 int err;
1130
1131 /* There is support for UDP large send offload by network
1132 * device, so create one single skb packet containing complete
1133 * udp datagram
1134 */
1135 skb = skb_peek_tail(queue);
1136 if (!skb) {
1137 skb = sock_alloc_send_skb(sk,
1138 hh_len + fragheaderlen + transhdrlen + 20,
1139 (flags & MSG_DONTWAIT), &err);
1140 if (!skb)
1141 return err;
1142
1143 /* reserve space for Hardware header */
1144 skb_reserve(skb, hh_len);
1145
1146 /* create space for UDP/IP header */
1147 skb_put(skb, fragheaderlen + transhdrlen);
1148
1149 /* initialize network header pointer */
1150 skb_set_network_header(skb, exthdrlen);
1151
1152 /* initialize protocol header pointer */
1153 skb->transport_header = skb->network_header + fragheaderlen;
1154
1155 skb->protocol = htons(ETH_P_IPV6);
1156 skb->csum = 0;
1157
1158 if (flags & MSG_CONFIRM)
1159 skb_set_dst_pending_confirm(skb, 1);
1160
1161 __skb_queue_tail(queue, skb);
1162 } else if (skb_is_gso(skb)) {
1163 goto append;
1164 }
1165
1166 skb->ip_summed = CHECKSUM_PARTIAL;
1167 /* Specify the length of each IPv6 datagram fragment.
1168 * It has to be a multiple of 8.
1169 */
1170 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1171 sizeof(struct frag_hdr)) & ~7;
1172 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1173 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1174 &fl6->daddr,
1175 &fl6->saddr);
1176
1177 append:
1178 return skb_append_datato_frags(sk, skb, getfrag, from,
1179 (length - transhdrlen));
1180 }
1181
1182 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1183 gfp_t gfp)
1184 {
1185 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1186 }
1187
1188 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1189 gfp_t gfp)
1190 {
1191 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1192 }
1193
1194 static void ip6_append_data_mtu(unsigned int *mtu,
1195 int *maxfraglen,
1196 unsigned int fragheaderlen,
1197 struct sk_buff *skb,
1198 struct rt6_info *rt,
1199 unsigned int orig_mtu)
1200 {
1201 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1202 if (!skb) {
1203 /* first fragment, reserve header_len */
1204 *mtu = orig_mtu - rt->dst.header_len;
1205
1206 } else {
1207 /*
1208 * this fragment is not first, the headers
1209 * space is regarded as data space.
1210 */
1211 *mtu = orig_mtu;
1212 }
1213 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1214 + fragheaderlen - sizeof(struct frag_hdr);
1215 }
1216 }
1217
1218 static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1219 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1220 struct rt6_info *rt, struct flowi6 *fl6)
1221 {
1222 struct ipv6_pinfo *np = inet6_sk(sk);
1223 unsigned int mtu;
1224 struct ipv6_txoptions *opt = ipc6->opt;
1225
1226 /*
1227 * setup for corking
1228 */
1229 if (opt) {
1230 if (WARN_ON(v6_cork->opt))
1231 return -EINVAL;
1232
1233 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1234 if (unlikely(!v6_cork->opt))
1235 return -ENOBUFS;
1236
1237 v6_cork->opt->tot_len = opt->tot_len;
1238 v6_cork->opt->opt_flen = opt->opt_flen;
1239 v6_cork->opt->opt_nflen = opt->opt_nflen;
1240
1241 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1242 sk->sk_allocation);
1243 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1244 return -ENOBUFS;
1245
1246 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1247 sk->sk_allocation);
1248 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1249 return -ENOBUFS;
1250
1251 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1252 sk->sk_allocation);
1253 if (opt->hopopt && !v6_cork->opt->hopopt)
1254 return -ENOBUFS;
1255
1256 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1257 sk->sk_allocation);
1258 if (opt->srcrt && !v6_cork->opt->srcrt)
1259 return -ENOBUFS;
1260
1261 /* need source address above miyazawa*/
1262 }
1263 dst_hold(&rt->dst);
1264 cork->base.dst = &rt->dst;
1265 cork->fl.u.ip6 = *fl6;
1266 v6_cork->hop_limit = ipc6->hlimit;
1267 v6_cork->tclass = ipc6->tclass;
1268 if (rt->dst.flags & DST_XFRM_TUNNEL)
1269 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1270 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1271 else
1272 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1273 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1274 if (np->frag_size < mtu) {
1275 if (np->frag_size)
1276 mtu = np->frag_size;
1277 }
1278 cork->base.fragsize = mtu;
1279 if (dst_allfrag(rt->dst.path))
1280 cork->base.flags |= IPCORK_ALLFRAG;
1281 cork->base.length = 0;
1282
1283 return 0;
1284 }
1285
1286 static int __ip6_append_data(struct sock *sk,
1287 struct flowi6 *fl6,
1288 struct sk_buff_head *queue,
1289 struct inet_cork *cork,
1290 struct inet6_cork *v6_cork,
1291 struct page_frag *pfrag,
1292 int getfrag(void *from, char *to, int offset,
1293 int len, int odd, struct sk_buff *skb),
1294 void *from, int length, int transhdrlen,
1295 unsigned int flags, struct ipcm6_cookie *ipc6,
1296 const struct sockcm_cookie *sockc)
1297 {
1298 struct sk_buff *skb, *skb_prev = NULL;
1299 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1300 int exthdrlen = 0;
1301 int dst_exthdrlen = 0;
1302 int hh_len;
1303 int copy;
1304 int err;
1305 int offset = 0;
1306 __u8 tx_flags = 0;
1307 u32 tskey = 0;
1308 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1309 struct ipv6_txoptions *opt = v6_cork->opt;
1310 int csummode = CHECKSUM_NONE;
1311 unsigned int maxnonfragsize, headersize;
1312
1313 skb = skb_peek_tail(queue);
1314 if (!skb) {
1315 exthdrlen = opt ? opt->opt_flen : 0;
1316 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1317 }
1318
1319 mtu = cork->fragsize;
1320 orig_mtu = mtu;
1321
1322 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1323
1324 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1325 (opt ? opt->opt_nflen : 0);
1326 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1327 sizeof(struct frag_hdr);
1328
1329 headersize = sizeof(struct ipv6hdr) +
1330 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1331 (dst_allfrag(&rt->dst) ?
1332 sizeof(struct frag_hdr) : 0) +
1333 rt->rt6i_nfheader_len;
1334
1335 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1336 (sk->sk_protocol == IPPROTO_UDP ||
1337 sk->sk_protocol == IPPROTO_RAW)) {
1338 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1339 sizeof(struct ipv6hdr));
1340 goto emsgsize;
1341 }
1342
1343 if (ip6_sk_ignore_df(sk))
1344 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1345 else
1346 maxnonfragsize = mtu;
1347
1348 if (cork->length + length > maxnonfragsize - headersize) {
1349 emsgsize:
1350 ipv6_local_error(sk, EMSGSIZE, fl6,
1351 mtu - headersize +
1352 sizeof(struct ipv6hdr));
1353 return -EMSGSIZE;
1354 }
1355
1356 /* CHECKSUM_PARTIAL only with no extension headers and when
1357 * we are not going to fragment
1358 */
1359 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1360 headersize == sizeof(struct ipv6hdr) &&
1361 length <= mtu - headersize &&
1362 !(flags & MSG_MORE) &&
1363 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1364 csummode = CHECKSUM_PARTIAL;
1365
1366 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1367 sock_tx_timestamp(sk, sockc->tsflags, &tx_flags);
1368 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1369 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1370 tskey = sk->sk_tskey++;
1371 }
1372
1373 /*
1374 * Let's try using as much space as possible.
1375 * Use MTU if total length of the message fits into the MTU.
1376 * Otherwise, we need to reserve fragment header and
1377 * fragment alignment (= 8-15 octects, in total).
1378 *
1379 * Note that we may need to "move" the data from the tail of
1380 * of the buffer to the new fragment when we split
1381 * the message.
1382 *
1383 * FIXME: It may be fragmented into multiple chunks
1384 * at once if non-fragmentable extension headers
1385 * are too large.
1386 * --yoshfuji
1387 */
1388
1389 cork->length += length;
1390 if ((((length + fragheaderlen) > mtu) ||
1391 (skb && skb_is_gso(skb))) &&
1392 (sk->sk_protocol == IPPROTO_UDP) &&
1393 (rt->dst.dev->features & NETIF_F_UFO) && !dst_xfrm(&rt->dst) &&
1394 (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) {
1395 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1396 hh_len, fragheaderlen, exthdrlen,
1397 transhdrlen, mtu, flags, fl6);
1398 if (err)
1399 goto error;
1400 return 0;
1401 }
1402
1403 if (!skb)
1404 goto alloc_new_skb;
1405
1406 while (length > 0) {
1407 /* Check if the remaining data fits into current packet. */
1408 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1409 if (copy < length)
1410 copy = maxfraglen - skb->len;
1411
1412 if (copy <= 0) {
1413 char *data;
1414 unsigned int datalen;
1415 unsigned int fraglen;
1416 unsigned int fraggap;
1417 unsigned int alloclen;
1418 alloc_new_skb:
1419 /* There's no room in the current skb */
1420 if (skb)
1421 fraggap = skb->len - maxfraglen;
1422 else
1423 fraggap = 0;
1424 /* update mtu and maxfraglen if necessary */
1425 if (!skb || !skb_prev)
1426 ip6_append_data_mtu(&mtu, &maxfraglen,
1427 fragheaderlen, skb, rt,
1428 orig_mtu);
1429
1430 skb_prev = skb;
1431
1432 /*
1433 * If remaining data exceeds the mtu,
1434 * we know we need more fragment(s).
1435 */
1436 datalen = length + fraggap;
1437
1438 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1439 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1440 if ((flags & MSG_MORE) &&
1441 !(rt->dst.dev->features&NETIF_F_SG))
1442 alloclen = mtu;
1443 else
1444 alloclen = datalen + fragheaderlen;
1445
1446 alloclen += dst_exthdrlen;
1447
1448 if (datalen != length + fraggap) {
1449 /*
1450 * this is not the last fragment, the trailer
1451 * space is regarded as data space.
1452 */
1453 datalen += rt->dst.trailer_len;
1454 }
1455
1456 alloclen += rt->dst.trailer_len;
1457 fraglen = datalen + fragheaderlen;
1458
1459 /*
1460 * We just reserve space for fragment header.
1461 * Note: this may be overallocation if the message
1462 * (without MSG_MORE) fits into the MTU.
1463 */
1464 alloclen += sizeof(struct frag_hdr);
1465
1466 if (transhdrlen) {
1467 skb = sock_alloc_send_skb(sk,
1468 alloclen + hh_len,
1469 (flags & MSG_DONTWAIT), &err);
1470 } else {
1471 skb = NULL;
1472 if (atomic_read(&sk->sk_wmem_alloc) <=
1473 2 * sk->sk_sndbuf)
1474 skb = sock_wmalloc(sk,
1475 alloclen + hh_len, 1,
1476 sk->sk_allocation);
1477 if (unlikely(!skb))
1478 err = -ENOBUFS;
1479 }
1480 if (!skb)
1481 goto error;
1482 /*
1483 * Fill in the control structures
1484 */
1485 skb->protocol = htons(ETH_P_IPV6);
1486 skb->ip_summed = csummode;
1487 skb->csum = 0;
1488 /* reserve for fragmentation and ipsec header */
1489 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1490 dst_exthdrlen);
1491
1492 /* Only the initial fragment is time stamped */
1493 skb_shinfo(skb)->tx_flags = tx_flags;
1494 tx_flags = 0;
1495 skb_shinfo(skb)->tskey = tskey;
1496 tskey = 0;
1497
1498 /*
1499 * Find where to start putting bytes
1500 */
1501 data = skb_put(skb, fraglen);
1502 skb_set_network_header(skb, exthdrlen);
1503 data += fragheaderlen;
1504 skb->transport_header = (skb->network_header +
1505 fragheaderlen);
1506 if (fraggap) {
1507 skb->csum = skb_copy_and_csum_bits(
1508 skb_prev, maxfraglen,
1509 data + transhdrlen, fraggap, 0);
1510 skb_prev->csum = csum_sub(skb_prev->csum,
1511 skb->csum);
1512 data += fraggap;
1513 pskb_trim_unique(skb_prev, maxfraglen);
1514 }
1515 copy = datalen - transhdrlen - fraggap;
1516
1517 if (copy < 0) {
1518 err = -EINVAL;
1519 kfree_skb(skb);
1520 goto error;
1521 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1522 err = -EFAULT;
1523 kfree_skb(skb);
1524 goto error;
1525 }
1526
1527 offset += copy;
1528 length -= datalen - fraggap;
1529 transhdrlen = 0;
1530 exthdrlen = 0;
1531 dst_exthdrlen = 0;
1532
1533 if ((flags & MSG_CONFIRM) && !skb_prev)
1534 skb_set_dst_pending_confirm(skb, 1);
1535
1536 /*
1537 * Put the packet on the pending queue
1538 */
1539 __skb_queue_tail(queue, skb);
1540 continue;
1541 }
1542
1543 if (copy > length)
1544 copy = length;
1545
1546 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1547 unsigned int off;
1548
1549 off = skb->len;
1550 if (getfrag(from, skb_put(skb, copy),
1551 offset, copy, off, skb) < 0) {
1552 __skb_trim(skb, off);
1553 err = -EFAULT;
1554 goto error;
1555 }
1556 } else {
1557 int i = skb_shinfo(skb)->nr_frags;
1558
1559 err = -ENOMEM;
1560 if (!sk_page_frag_refill(sk, pfrag))
1561 goto error;
1562
1563 if (!skb_can_coalesce(skb, i, pfrag->page,
1564 pfrag->offset)) {
1565 err = -EMSGSIZE;
1566 if (i == MAX_SKB_FRAGS)
1567 goto error;
1568
1569 __skb_fill_page_desc(skb, i, pfrag->page,
1570 pfrag->offset, 0);
1571 skb_shinfo(skb)->nr_frags = ++i;
1572 get_page(pfrag->page);
1573 }
1574 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1575 if (getfrag(from,
1576 page_address(pfrag->page) + pfrag->offset,
1577 offset, copy, skb->len, skb) < 0)
1578 goto error_efault;
1579
1580 pfrag->offset += copy;
1581 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1582 skb->len += copy;
1583 skb->data_len += copy;
1584 skb->truesize += copy;
1585 atomic_add(copy, &sk->sk_wmem_alloc);
1586 }
1587 offset += copy;
1588 length -= copy;
1589 }
1590
1591 return 0;
1592
1593 error_efault:
1594 err = -EFAULT;
1595 error:
1596 cork->length -= length;
1597 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1598 return err;
1599 }
1600
1601 int ip6_append_data(struct sock *sk,
1602 int getfrag(void *from, char *to, int offset, int len,
1603 int odd, struct sk_buff *skb),
1604 void *from, int length, int transhdrlen,
1605 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1606 struct rt6_info *rt, unsigned int flags,
1607 const struct sockcm_cookie *sockc)
1608 {
1609 struct inet_sock *inet = inet_sk(sk);
1610 struct ipv6_pinfo *np = inet6_sk(sk);
1611 int exthdrlen;
1612 int err;
1613
1614 if (flags&MSG_PROBE)
1615 return 0;
1616 if (skb_queue_empty(&sk->sk_write_queue)) {
1617 /*
1618 * setup for corking
1619 */
1620 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1621 ipc6, rt, fl6);
1622 if (err)
1623 return err;
1624
1625 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1626 length += exthdrlen;
1627 transhdrlen += exthdrlen;
1628 } else {
1629 fl6 = &inet->cork.fl.u.ip6;
1630 transhdrlen = 0;
1631 }
1632
1633 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1634 &np->cork, sk_page_frag(sk), getfrag,
1635 from, length, transhdrlen, flags, ipc6, sockc);
1636 }
1637 EXPORT_SYMBOL_GPL(ip6_append_data);
1638
1639 static void ip6_cork_release(struct inet_cork_full *cork,
1640 struct inet6_cork *v6_cork)
1641 {
1642 if (v6_cork->opt) {
1643 kfree(v6_cork->opt->dst0opt);
1644 kfree(v6_cork->opt->dst1opt);
1645 kfree(v6_cork->opt->hopopt);
1646 kfree(v6_cork->opt->srcrt);
1647 kfree(v6_cork->opt);
1648 v6_cork->opt = NULL;
1649 }
1650
1651 if (cork->base.dst) {
1652 dst_release(cork->base.dst);
1653 cork->base.dst = NULL;
1654 cork->base.flags &= ~IPCORK_ALLFRAG;
1655 }
1656 memset(&cork->fl, 0, sizeof(cork->fl));
1657 }
1658
1659 struct sk_buff *__ip6_make_skb(struct sock *sk,
1660 struct sk_buff_head *queue,
1661 struct inet_cork_full *cork,
1662 struct inet6_cork *v6_cork)
1663 {
1664 struct sk_buff *skb, *tmp_skb;
1665 struct sk_buff **tail_skb;
1666 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1667 struct ipv6_pinfo *np = inet6_sk(sk);
1668 struct net *net = sock_net(sk);
1669 struct ipv6hdr *hdr;
1670 struct ipv6_txoptions *opt = v6_cork->opt;
1671 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1672 struct flowi6 *fl6 = &cork->fl.u.ip6;
1673 unsigned char proto = fl6->flowi6_proto;
1674
1675 skb = __skb_dequeue(queue);
1676 if (!skb)
1677 goto out;
1678 tail_skb = &(skb_shinfo(skb)->frag_list);
1679
1680 /* move skb->data to ip header from ext header */
1681 if (skb->data < skb_network_header(skb))
1682 __skb_pull(skb, skb_network_offset(skb));
1683 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1684 __skb_pull(tmp_skb, skb_network_header_len(skb));
1685 *tail_skb = tmp_skb;
1686 tail_skb = &(tmp_skb->next);
1687 skb->len += tmp_skb->len;
1688 skb->data_len += tmp_skb->len;
1689 skb->truesize += tmp_skb->truesize;
1690 tmp_skb->destructor = NULL;
1691 tmp_skb->sk = NULL;
1692 }
1693
1694 /* Allow local fragmentation. */
1695 skb->ignore_df = ip6_sk_ignore_df(sk);
1696
1697 *final_dst = fl6->daddr;
1698 __skb_pull(skb, skb_network_header_len(skb));
1699 if (opt && opt->opt_flen)
1700 ipv6_push_frag_opts(skb, opt, &proto);
1701 if (opt && opt->opt_nflen)
1702 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1703
1704 skb_push(skb, sizeof(struct ipv6hdr));
1705 skb_reset_network_header(skb);
1706 hdr = ipv6_hdr(skb);
1707
1708 ip6_flow_hdr(hdr, v6_cork->tclass,
1709 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1710 np->autoflowlabel, fl6));
1711 hdr->hop_limit = v6_cork->hop_limit;
1712 hdr->nexthdr = proto;
1713 hdr->saddr = fl6->saddr;
1714 hdr->daddr = *final_dst;
1715
1716 skb->priority = sk->sk_priority;
1717 skb->mark = sk->sk_mark;
1718
1719 skb_dst_set(skb, dst_clone(&rt->dst));
1720 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1721 if (proto == IPPROTO_ICMPV6) {
1722 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1723
1724 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1725 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1726 }
1727
1728 ip6_cork_release(cork, v6_cork);
1729 out:
1730 return skb;
1731 }
1732
1733 int ip6_send_skb(struct sk_buff *skb)
1734 {
1735 struct net *net = sock_net(skb->sk);
1736 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1737 int err;
1738
1739 err = ip6_local_out(net, skb->sk, skb);
1740 if (err) {
1741 if (err > 0)
1742 err = net_xmit_errno(err);
1743 if (err)
1744 IP6_INC_STATS(net, rt->rt6i_idev,
1745 IPSTATS_MIB_OUTDISCARDS);
1746 }
1747
1748 return err;
1749 }
1750
1751 int ip6_push_pending_frames(struct sock *sk)
1752 {
1753 struct sk_buff *skb;
1754
1755 skb = ip6_finish_skb(sk);
1756 if (!skb)
1757 return 0;
1758
1759 return ip6_send_skb(skb);
1760 }
1761 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1762
1763 static void __ip6_flush_pending_frames(struct sock *sk,
1764 struct sk_buff_head *queue,
1765 struct inet_cork_full *cork,
1766 struct inet6_cork *v6_cork)
1767 {
1768 struct sk_buff *skb;
1769
1770 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1771 if (skb_dst(skb))
1772 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1773 IPSTATS_MIB_OUTDISCARDS);
1774 kfree_skb(skb);
1775 }
1776
1777 ip6_cork_release(cork, v6_cork);
1778 }
1779
1780 void ip6_flush_pending_frames(struct sock *sk)
1781 {
1782 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1783 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1784 }
1785 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1786
1787 struct sk_buff *ip6_make_skb(struct sock *sk,
1788 int getfrag(void *from, char *to, int offset,
1789 int len, int odd, struct sk_buff *skb),
1790 void *from, int length, int transhdrlen,
1791 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1792 struct rt6_info *rt, unsigned int flags,
1793 const struct sockcm_cookie *sockc)
1794 {
1795 struct inet_cork_full cork;
1796 struct inet6_cork v6_cork;
1797 struct sk_buff_head queue;
1798 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1799 int err;
1800
1801 if (flags & MSG_PROBE)
1802 return NULL;
1803
1804 __skb_queue_head_init(&queue);
1805
1806 cork.base.flags = 0;
1807 cork.base.addr = 0;
1808 cork.base.opt = NULL;
1809 v6_cork.opt = NULL;
1810 err = ip6_setup_cork(sk, &cork, &v6_cork, ipc6, rt, fl6);
1811 if (err)
1812 return ERR_PTR(err);
1813
1814 if (ipc6->dontfrag < 0)
1815 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1816
1817 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1818 &current->task_frag, getfrag, from,
1819 length + exthdrlen, transhdrlen + exthdrlen,
1820 flags, ipc6, sockc);
1821 if (err) {
1822 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1823 return ERR_PTR(err);
1824 }
1825
1826 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1827 }
Linux with added FPC-III support