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
59 int __ip6_local_out(struct sk_buff *skb)
60 {
61 int len;
62
63 len = skb->len - sizeof(struct ipv6hdr);
64 if (len > IPV6_MAXPLEN)
65 len = 0;
66 ipv6_hdr(skb)->payload_len = htons(len);
67
68 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
69 skb_dst(skb)->dev, dst_output);
70 }
71
72 int ip6_local_out(struct sk_buff *skb)
73 {
74 int err;
75
76 err = __ip6_local_out(skb);
77 if (likely(err == 1))
78 err = dst_output(skb);
79
80 return err;
81 }
82 EXPORT_SYMBOL_GPL(ip6_local_out);
83
84 static int ip6_finish_output2(struct sk_buff *skb)
85 {
86 struct dst_entry *dst = skb_dst(skb);
87 struct net_device *dev = dst->dev;
88 struct neighbour *neigh;
89 struct in6_addr *nexthop;
90 int ret;
91
92 skb->protocol = htons(ETH_P_IPV6);
93 skb->dev = dev;
94
95 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
96 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
97
98 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
99 ((mroute6_socket(dev_net(dev), skb) &&
100 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
101 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
102 &ipv6_hdr(skb)->saddr))) {
103 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
104
105 /* Do not check for IFF_ALLMULTI; multicast routing
106 is not supported in any case.
107 */
108 if (newskb)
109 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
110 newskb, NULL, newskb->dev,
111 dev_loopback_xmit);
112
113 if (ipv6_hdr(skb)->hop_limit == 0) {
114 IP6_INC_STATS(dev_net(dev), idev,
115 IPSTATS_MIB_OUTDISCARDS);
116 kfree_skb(skb);
117 return 0;
118 }
119 }
120
121 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
122 skb->len);
123
124 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
125 IPV6_ADDR_SCOPE_NODELOCAL &&
126 !(dev->flags & IFF_LOOPBACK)) {
127 kfree_skb(skb);
128 return 0;
129 }
130 }
131
132 rcu_read_lock_bh();
133 nexthop = rt6_nexthop((struct rt6_info *)dst);
134 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
135 if (unlikely(!neigh))
136 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
137 if (!IS_ERR(neigh)) {
138 ret = dst_neigh_output(dst, neigh, skb);
139 rcu_read_unlock_bh();
140 return ret;
141 }
142 rcu_read_unlock_bh();
143
144 IP6_INC_STATS(dev_net(dst->dev),
145 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
146 kfree_skb(skb);
147 return -EINVAL;
148 }
149
150 static int ip6_finish_output(struct sk_buff *skb)
151 {
152 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
153 dst_allfrag(skb_dst(skb)) ||
154 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
155 return ip6_fragment(skb, ip6_finish_output2);
156 else
157 return ip6_finish_output2(skb);
158 }
159
160 int ip6_output(struct sk_buff *skb)
161 {
162 struct net_device *dev = skb_dst(skb)->dev;
163 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
164 if (unlikely(idev->cnf.disable_ipv6)) {
165 IP6_INC_STATS(dev_net(dev), idev,
166 IPSTATS_MIB_OUTDISCARDS);
167 kfree_skb(skb);
168 return 0;
169 }
170
171 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
172 ip6_finish_output,
173 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
174 }
175
176 /*
177 * xmit an sk_buff (used by TCP, SCTP and DCCP)
178 */
179
180 int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
181 struct ipv6_txoptions *opt, int tclass)
182 {
183 struct net *net = sock_net(sk);
184 struct ipv6_pinfo *np = inet6_sk(sk);
185 struct in6_addr *first_hop = &fl6->daddr;
186 struct dst_entry *dst = skb_dst(skb);
187 struct ipv6hdr *hdr;
188 u8 proto = fl6->flowi6_proto;
189 int seg_len = skb->len;
190 int hlimit = -1;
191 u32 mtu;
192
193 if (opt) {
194 unsigned int head_room;
195
196 /* First: exthdrs may take lots of space (~8K for now)
197 MAX_HEADER is not enough.
198 */
199 head_room = opt->opt_nflen + opt->opt_flen;
200 seg_len += head_room;
201 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
202
203 if (skb_headroom(skb) < head_room) {
204 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
205 if (skb2 == NULL) {
206 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
207 IPSTATS_MIB_OUTDISCARDS);
208 kfree_skb(skb);
209 return -ENOBUFS;
210 }
211 consume_skb(skb);
212 skb = skb2;
213 skb_set_owner_w(skb, sk);
214 }
215 if (opt->opt_flen)
216 ipv6_push_frag_opts(skb, opt, &proto);
217 if (opt->opt_nflen)
218 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
219 }
220
221 skb_push(skb, sizeof(struct ipv6hdr));
222 skb_reset_network_header(skb);
223 hdr = ipv6_hdr(skb);
224
225 /*
226 * Fill in the IPv6 header
227 */
228 if (np)
229 hlimit = np->hop_limit;
230 if (hlimit < 0)
231 hlimit = ip6_dst_hoplimit(dst);
232
233 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
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->priority = sk->sk_priority;
243 skb->mark = sk->sk_mark;
244
245 mtu = dst_mtu(dst);
246 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
247 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
248 IPSTATS_MIB_OUT, skb->len);
249 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
250 dst->dev, dst_output);
251 }
252
253 skb->dev = dst->dev;
254 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
255 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
256 kfree_skb(skb);
257 return -EMSGSIZE;
258 }
259
260 EXPORT_SYMBOL(ip6_xmit);
261
262 static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
263 {
264 struct ip6_ra_chain *ra;
265 struct sock *last = NULL;
266
267 read_lock(&ip6_ra_lock);
268 for (ra = ip6_ra_chain; ra; ra = ra->next) {
269 struct sock *sk = ra->sk;
270 if (sk && ra->sel == sel &&
271 (!sk->sk_bound_dev_if ||
272 sk->sk_bound_dev_if == skb->dev->ifindex)) {
273 if (last) {
274 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
275 if (skb2)
276 rawv6_rcv(last, skb2);
277 }
278 last = sk;
279 }
280 }
281
282 if (last) {
283 rawv6_rcv(last, skb);
284 read_unlock(&ip6_ra_lock);
285 return 1;
286 }
287 read_unlock(&ip6_ra_lock);
288 return 0;
289 }
290
291 static int ip6_forward_proxy_check(struct sk_buff *skb)
292 {
293 struct ipv6hdr *hdr = ipv6_hdr(skb);
294 u8 nexthdr = hdr->nexthdr;
295 __be16 frag_off;
296 int offset;
297
298 if (ipv6_ext_hdr(nexthdr)) {
299 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
300 if (offset < 0)
301 return 0;
302 } else
303 offset = sizeof(struct ipv6hdr);
304
305 if (nexthdr == IPPROTO_ICMPV6) {
306 struct icmp6hdr *icmp6;
307
308 if (!pskb_may_pull(skb, (skb_network_header(skb) +
309 offset + 1 - skb->data)))
310 return 0;
311
312 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
313
314 switch (icmp6->icmp6_type) {
315 case NDISC_ROUTER_SOLICITATION:
316 case NDISC_ROUTER_ADVERTISEMENT:
317 case NDISC_NEIGHBOUR_SOLICITATION:
318 case NDISC_NEIGHBOUR_ADVERTISEMENT:
319 case NDISC_REDIRECT:
320 /* For reaction involving unicast neighbor discovery
321 * message destined to the proxied address, pass it to
322 * input function.
323 */
324 return 1;
325 default:
326 break;
327 }
328 }
329
330 /*
331 * The proxying router can't forward traffic sent to a link-local
332 * address, so signal the sender and discard the packet. This
333 * behavior is clarified by the MIPv6 specification.
334 */
335 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
336 dst_link_failure(skb);
337 return -1;
338 }
339
340 return 0;
341 }
342
343 static inline int ip6_forward_finish(struct sk_buff *skb)
344 {
345 return dst_output(skb);
346 }
347
348 static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
349 {
350 if (skb->len <= mtu)
351 return false;
352
353 /* ipv6 conntrack defrag sets max_frag_size + local_df */
354 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
355 return true;
356
357 if (skb->local_df)
358 return false;
359
360 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
361 return false;
362
363 return true;
364 }
365
366 int ip6_forward(struct sk_buff *skb)
367 {
368 struct dst_entry *dst = skb_dst(skb);
369 struct ipv6hdr *hdr = ipv6_hdr(skb);
370 struct inet6_skb_parm *opt = IP6CB(skb);
371 struct net *net = dev_net(dst->dev);
372 u32 mtu;
373
374 if (net->ipv6.devconf_all->forwarding == 0)
375 goto error;
376
377 if (skb_warn_if_lro(skb))
378 goto drop;
379
380 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
382 goto drop;
383 }
384
385 if (skb->pkt_type != PACKET_HOST)
386 goto drop;
387
388 skb_forward_csum(skb);
389
390 /*
391 * We DO NOT make any processing on
392 * RA packets, pushing them to user level AS IS
393 * without ane WARRANTY that application will be able
394 * to interpret them. The reason is that we
395 * cannot make anything clever here.
396 *
397 * We are not end-node, so that if packet contains
398 * AH/ESP, we cannot make anything.
399 * Defragmentation also would be mistake, RA packets
400 * cannot be fragmented, because there is no warranty
401 * that different fragments will go along one path. --ANK
402 */
403 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
404 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
405 return 0;
406 }
407
408 /*
409 * check and decrement ttl
410 */
411 if (hdr->hop_limit <= 1) {
412 /* Force OUTPUT device used as source address */
413 skb->dev = dst->dev;
414 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
415 IP6_INC_STATS_BH(net,
416 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
417
418 kfree_skb(skb);
419 return -ETIMEDOUT;
420 }
421
422 /* XXX: idev->cnf.proxy_ndp? */
423 if (net->ipv6.devconf_all->proxy_ndp &&
424 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
425 int proxied = ip6_forward_proxy_check(skb);
426 if (proxied > 0)
427 return ip6_input(skb);
428 else if (proxied < 0) {
429 IP6_INC_STATS(net, ip6_dst_idev(dst),
430 IPSTATS_MIB_INDISCARDS);
431 goto drop;
432 }
433 }
434
435 if (!xfrm6_route_forward(skb)) {
436 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
437 goto drop;
438 }
439 dst = skb_dst(skb);
440
441 /* IPv6 specs say nothing about it, but it is clear that we cannot
442 send redirects to source routed frames.
443 We don't send redirects to frames decapsulated from IPsec.
444 */
445 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
446 struct in6_addr *target = NULL;
447 struct inet_peer *peer;
448 struct rt6_info *rt;
449
450 /*
451 * incoming and outgoing devices are the same
452 * send a redirect.
453 */
454
455 rt = (struct rt6_info *) dst;
456 if (rt->rt6i_flags & RTF_GATEWAY)
457 target = &rt->rt6i_gateway;
458 else
459 target = &hdr->daddr;
460
461 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
462
463 /* Limit redirects both by destination (here)
464 and by source (inside ndisc_send_redirect)
465 */
466 if (inet_peer_xrlim_allow(peer, 1*HZ))
467 ndisc_send_redirect(skb, target);
468 if (peer)
469 inet_putpeer(peer);
470 } else {
471 int addrtype = ipv6_addr_type(&hdr->saddr);
472
473 /* This check is security critical. */
474 if (addrtype == IPV6_ADDR_ANY ||
475 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
476 goto error;
477 if (addrtype & IPV6_ADDR_LINKLOCAL) {
478 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
479 ICMPV6_NOT_NEIGHBOUR, 0);
480 goto error;
481 }
482 }
483
484 mtu = dst_mtu(dst);
485 if (mtu < IPV6_MIN_MTU)
486 mtu = IPV6_MIN_MTU;
487
488 if (ip6_pkt_too_big(skb, mtu)) {
489 /* Again, force OUTPUT device used as source address */
490 skb->dev = dst->dev;
491 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
492 IP6_INC_STATS_BH(net,
493 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
494 IP6_INC_STATS_BH(net,
495 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
496 kfree_skb(skb);
497 return -EMSGSIZE;
498 }
499
500 if (skb_cow(skb, dst->dev->hard_header_len)) {
501 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
502 goto drop;
503 }
504
505 hdr = ipv6_hdr(skb);
506
507 /* Mangling hops number delayed to point after skb COW */
508
509 hdr->hop_limit--;
510
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
514 ip6_forward_finish);
515
516 error:
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
518 drop:
519 kfree_skb(skb);
520 return -EINVAL;
521 }
522
523 static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
524 {
525 to->pkt_type = from->pkt_type;
526 to->priority = from->priority;
527 to->protocol = from->protocol;
528 skb_dst_drop(to);
529 skb_dst_set(to, dst_clone(skb_dst(from)));
530 to->dev = from->dev;
531 to->mark = from->mark;
532
533 #ifdef CONFIG_NET_SCHED
534 to->tc_index = from->tc_index;
535 #endif
536 nf_copy(to, from);
537 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
538 to->nf_trace = from->nf_trace;
539 #endif
540 skb_copy_secmark(to, from);
541 }
542
543 static void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
544 {
545 static u32 ip6_idents_hashrnd __read_mostly;
546 static bool hashrnd_initialized = false;
547 u32 hash, id;
548
549 if (unlikely(!hashrnd_initialized)) {
550 hashrnd_initialized = true;
551 get_random_bytes(&ip6_idents_hashrnd, sizeof(ip6_idents_hashrnd));
552 }
553 hash = __ipv6_addr_jhash(&rt->rt6i_dst.addr, ip6_idents_hashrnd);
554 hash = __ipv6_addr_jhash(&rt->rt6i_src.addr, hash);
555
556 id = ip_idents_reserve(hash, 1);
557 fhdr->identification = htonl(id);
558 }
559
560 int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
561 {
562 struct sk_buff *frag;
563 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
564 struct ipv6_pinfo *np = skb->sk ? 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 = 0;
570 int ptr, offset = 0, err=0;
571 u8 *prevhdr, nexthdr = 0;
572 struct net *net = dev_net(skb_dst(skb)->dev);
573
574 hlen = ip6_find_1stfragopt(skb, &prevhdr);
575 nexthdr = *prevhdr;
576
577 mtu = ip6_skb_dst_mtu(skb);
578
579 /* We must not fragment if the socket is set to force MTU discovery
580 * or if the skb it not generated by a local socket.
581 */
582 if (unlikely(!skb->local_df && skb->len > mtu) ||
583 (IP6CB(skb)->frag_max_size &&
584 IP6CB(skb)->frag_max_size > mtu)) {
585 if (skb->sk && dst_allfrag(skb_dst(skb)))
586 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
587
588 skb->dev = skb_dst(skb)->dev;
589 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
590 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
591 IPSTATS_MIB_FRAGFAILS);
592 kfree_skb(skb);
593 return -EMSGSIZE;
594 }
595
596 if (np && np->frag_size < mtu) {
597 if (np->frag_size)
598 mtu = np->frag_size;
599 }
600 mtu -= hlen + sizeof(struct frag_hdr);
601
602 if (skb_has_frag_list(skb)) {
603 int first_len = skb_pagelen(skb);
604 struct sk_buff *frag2;
605
606 if (first_len - hlen > mtu ||
607 ((first_len - hlen) & 7) ||
608 skb_cloned(skb))
609 goto slow_path;
610
611 skb_walk_frags(skb, frag) {
612 /* Correct geometry. */
613 if (frag->len > mtu ||
614 ((frag->len & 7) && frag->next) ||
615 skb_headroom(frag) < hlen)
616 goto slow_path_clean;
617
618 /* Partially cloned skb? */
619 if (skb_shared(frag))
620 goto slow_path_clean;
621
622 BUG_ON(frag->sk);
623 if (skb->sk) {
624 frag->sk = skb->sk;
625 frag->destructor = sock_wfree;
626 }
627 skb->truesize -= frag->truesize;
628 }
629
630 err = 0;
631 offset = 0;
632 frag = skb_shinfo(skb)->frag_list;
633 skb_frag_list_init(skb);
634 /* BUILD HEADER */
635
636 *prevhdr = NEXTHDR_FRAGMENT;
637 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
638 if (!tmp_hdr) {
639 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
640 IPSTATS_MIB_FRAGFAILS);
641 return -ENOMEM;
642 }
643
644 __skb_pull(skb, hlen);
645 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
646 __skb_push(skb, hlen);
647 skb_reset_network_header(skb);
648 memcpy(skb_network_header(skb), tmp_hdr, hlen);
649
650 ipv6_select_ident(fh, rt);
651 fh->nexthdr = nexthdr;
652 fh->reserved = 0;
653 fh->frag_off = htons(IP6_MF);
654 frag_id = fh->identification;
655
656 first_len = skb_pagelen(skb);
657 skb->data_len = first_len - skb_headlen(skb);
658 skb->len = first_len;
659 ipv6_hdr(skb)->payload_len = htons(first_len -
660 sizeof(struct ipv6hdr));
661
662 dst_hold(&rt->dst);
663
664 for (;;) {
665 /* Prepare header of the next frame,
666 * before previous one went down. */
667 if (frag) {
668 frag->ip_summed = CHECKSUM_NONE;
669 skb_reset_transport_header(frag);
670 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
671 __skb_push(frag, hlen);
672 skb_reset_network_header(frag);
673 memcpy(skb_network_header(frag), tmp_hdr,
674 hlen);
675 offset += skb->len - hlen - sizeof(struct frag_hdr);
676 fh->nexthdr = nexthdr;
677 fh->reserved = 0;
678 fh->frag_off = htons(offset);
679 if (frag->next != NULL)
680 fh->frag_off |= htons(IP6_MF);
681 fh->identification = frag_id;
682 ipv6_hdr(frag)->payload_len =
683 htons(frag->len -
684 sizeof(struct ipv6hdr));
685 ip6_copy_metadata(frag, skb);
686 }
687
688 err = output(skb);
689 if(!err)
690 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
691 IPSTATS_MIB_FRAGCREATES);
692
693 if (err || !frag)
694 break;
695
696 skb = frag;
697 frag = skb->next;
698 skb->next = NULL;
699 }
700
701 kfree(tmp_hdr);
702
703 if (err == 0) {
704 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
705 IPSTATS_MIB_FRAGOKS);
706 ip6_rt_put(rt);
707 return 0;
708 }
709
710 while (frag) {
711 skb = frag->next;
712 kfree_skb(frag);
713 frag = skb;
714 }
715
716 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
717 IPSTATS_MIB_FRAGFAILS);
718 ip6_rt_put(rt);
719 return err;
720
721 slow_path_clean:
722 skb_walk_frags(skb, frag2) {
723 if (frag2 == frag)
724 break;
725 frag2->sk = NULL;
726 frag2->destructor = NULL;
727 skb->truesize += frag2->truesize;
728 }
729 }
730
731 slow_path:
732 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
733 skb_checksum_help(skb))
734 goto fail;
735
736 left = skb->len - hlen; /* Space per frame */
737 ptr = hlen; /* Where to start from */
738
739 /*
740 * Fragment the datagram.
741 */
742
743 hroom = LL_RESERVED_SPACE(rt->dst.dev);
744 troom = rt->dst.dev->needed_tailroom;
745
746 /*
747 * Keep copying data until we run out.
748 */
749 while(left > 0) {
750 u8 *fragnexthdr_offset;
751
752 len = left;
753 /* IF: it doesn't fit, use 'mtu' - the data space left */
754 if (len > mtu)
755 len = mtu;
756 /* IF: we are not sending up to and including the packet end
757 then align the next start on an eight byte boundary */
758 if (len < left) {
759 len &= ~7;
760 }
761 /*
762 * Allocate buffer.
763 */
764
765 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
766 hroom + troom, GFP_ATOMIC)) == NULL) {
767 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
768 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
769 IPSTATS_MIB_FRAGFAILS);
770 err = -ENOMEM;
771 goto fail;
772 }
773
774 /*
775 * Set up data on packet
776 */
777
778 ip6_copy_metadata(frag, skb);
779 skb_reserve(frag, hroom);
780 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
781 skb_reset_network_header(frag);
782 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
783 frag->transport_header = (frag->network_header + hlen +
784 sizeof(struct frag_hdr));
785
786 /*
787 * Charge the memory for the fragment to any owner
788 * it might possess
789 */
790 if (skb->sk)
791 skb_set_owner_w(frag, skb->sk);
792
793 /*
794 * Copy the packet header into the new buffer.
795 */
796 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
797
798 fragnexthdr_offset = skb_network_header(frag);
799 fragnexthdr_offset += prevhdr - skb_network_header(skb);
800 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
801
802 /*
803 * Build fragment header.
804 */
805 fh->nexthdr = nexthdr;
806 fh->reserved = 0;
807 if (!frag_id) {
808 ipv6_select_ident(fh, rt);
809 frag_id = fh->identification;
810 } else
811 fh->identification = frag_id;
812
813 /*
814 * Copy a block of the IP datagram.
815 */
816 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
817 BUG();
818 left -= len;
819
820 fh->frag_off = htons(offset);
821 if (left > 0)
822 fh->frag_off |= htons(IP6_MF);
823 ipv6_hdr(frag)->payload_len = htons(frag->len -
824 sizeof(struct ipv6hdr));
825
826 ptr += len;
827 offset += len;
828
829 /*
830 * Put this fragment into the sending queue.
831 */
832 err = output(frag);
833 if (err)
834 goto fail;
835
836 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
837 IPSTATS_MIB_FRAGCREATES);
838 }
839 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
840 IPSTATS_MIB_FRAGOKS);
841 consume_skb(skb);
842 return err;
843
844 fail:
845 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
846 IPSTATS_MIB_FRAGFAILS);
847 kfree_skb(skb);
848 return err;
849 }
850
851 static inline int ip6_rt_check(const struct rt6key *rt_key,
852 const struct in6_addr *fl_addr,
853 const struct in6_addr *addr_cache)
854 {
855 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
856 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
857 }
858
859 static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
860 struct dst_entry *dst,
861 const struct flowi6 *fl6)
862 {
863 struct ipv6_pinfo *np = inet6_sk(sk);
864 struct rt6_info *rt;
865
866 if (!dst)
867 goto out;
868
869 if (dst->ops->family != AF_INET6) {
870 dst_release(dst);
871 return NULL;
872 }
873
874 rt = (struct rt6_info *)dst;
875 /* Yes, checking route validity in not connected
876 * case is not very simple. Take into account,
877 * that we do not support routing by source, TOS,
878 * and MSG_DONTROUTE --ANK (980726)
879 *
880 * 1. ip6_rt_check(): If route was host route,
881 * check that cached destination is current.
882 * If it is network route, we still may
883 * check its validity using saved pointer
884 * to the last used address: daddr_cache.
885 * We do not want to save whole address now,
886 * (because main consumer of this service
887 * is tcp, which has not this problem),
888 * so that the last trick works only on connected
889 * sockets.
890 * 2. oif also should be the same.
891 */
892 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
893 #ifdef CONFIG_IPV6_SUBTREES
894 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
895 #endif
896 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
897 dst_release(dst);
898 dst = NULL;
899 }
900
901 out:
902 return dst;
903 }
904
905 static int ip6_dst_lookup_tail(struct sock *sk,
906 struct dst_entry **dst, struct flowi6 *fl6)
907 {
908 struct net *net = sock_net(sk);
909 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
910 struct neighbour *n;
911 struct rt6_info *rt;
912 #endif
913 int err;
914
915 if (*dst == NULL)
916 *dst = ip6_route_output(net, sk, fl6);
917
918 if ((err = (*dst)->error))
919 goto out_err_release;
920
921 if (ipv6_addr_any(&fl6->saddr)) {
922 struct rt6_info *rt = (struct rt6_info *) *dst;
923 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
924 sk ? inet6_sk(sk)->srcprefs : 0,
925 &fl6->saddr);
926 if (err)
927 goto out_err_release;
928 }
929
930 #ifdef CONFIG_IPV6_OPTIMISTIC_DAD
931 /*
932 * Here if the dst entry we've looked up
933 * has a neighbour entry that is in the INCOMPLETE
934 * state and the src address from the flow is
935 * marked as OPTIMISTIC, we release the found
936 * dst entry and replace it instead with the
937 * dst entry of the nexthop router
938 */
939 rt = (struct rt6_info *) *dst;
940 rcu_read_lock_bh();
941 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
942 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
943 rcu_read_unlock_bh();
944
945 if (err) {
946 struct inet6_ifaddr *ifp;
947 struct flowi6 fl_gw6;
948 int redirect;
949
950 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
951 (*dst)->dev, 1);
952
953 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
954 if (ifp)
955 in6_ifa_put(ifp);
956
957 if (redirect) {
958 /*
959 * We need to get the dst entry for the
960 * default router instead
961 */
962 dst_release(*dst);
963 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
964 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
965 *dst = ip6_route_output(net, sk, &fl_gw6);
966 if ((err = (*dst)->error))
967 goto out_err_release;
968 }
969 }
970 #endif
971
972 return 0;
973
974 out_err_release:
975 if (err == -ENETUNREACH)
976 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
977 dst_release(*dst);
978 *dst = NULL;
979 return err;
980 }
981
982 /**
983 * ip6_dst_lookup - perform route lookup on flow
984 * @sk: socket which provides route info
985 * @dst: pointer to dst_entry * for result
986 * @fl6: flow to lookup
987 *
988 * This function performs a route lookup on the given flow.
989 *
990 * It returns zero on success, or a standard errno code on error.
991 */
992 int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
993 {
994 *dst = NULL;
995 return ip6_dst_lookup_tail(sk, dst, fl6);
996 }
997 EXPORT_SYMBOL_GPL(ip6_dst_lookup);
998
999 /**
1000 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1001 * @sk: socket which provides route info
1002 * @fl6: flow to lookup
1003 * @final_dst: final destination address for ipsec lookup
1004 * @can_sleep: we are in a sleepable context
1005 *
1006 * This function performs a route lookup on the given flow.
1007 *
1008 * It returns a valid dst pointer on success, or a pointer encoded
1009 * error code.
1010 */
1011 struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1012 const struct in6_addr *final_dst,
1013 bool can_sleep)
1014 {
1015 struct dst_entry *dst = NULL;
1016 int err;
1017
1018 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1019 if (err)
1020 return ERR_PTR(err);
1021 if (final_dst)
1022 fl6->daddr = *final_dst;
1023 if (can_sleep)
1024 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1025
1026 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1027 }
1028 EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1029
1030 /**
1031 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1032 * @sk: socket which provides the dst cache and route info
1033 * @fl6: flow to lookup
1034 * @final_dst: final destination address for ipsec lookup
1035 * @can_sleep: we are in a sleepable context
1036 *
1037 * This function performs a route lookup on the given flow with the
1038 * possibility of using the cached route in the socket if it is valid.
1039 * It will take the socket dst lock when operating on the dst cache.
1040 * As a result, this function can only be used in process context.
1041 *
1042 * It returns a valid dst pointer on success, or a pointer encoded
1043 * error code.
1044 */
1045 struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1046 const struct in6_addr *final_dst,
1047 bool can_sleep)
1048 {
1049 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1050 int err;
1051
1052 dst = ip6_sk_dst_check(sk, dst, fl6);
1053
1054 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1055 if (err)
1056 return ERR_PTR(err);
1057 if (final_dst)
1058 fl6->daddr = *final_dst;
1059 if (can_sleep)
1060 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1061
1062 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1063 }
1064 EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1065
1066 static inline int ip6_ufo_append_data(struct sock *sk,
1067 int getfrag(void *from, char *to, int offset, int len,
1068 int odd, struct sk_buff *skb),
1069 void *from, int length, int hh_len, int fragheaderlen,
1070 int transhdrlen, int mtu,unsigned int flags,
1071 struct rt6_info *rt)
1072
1073 {
1074 struct sk_buff *skb;
1075 int err;
1076
1077 /* There is support for UDP large send offload by network
1078 * device, so create one single skb packet containing complete
1079 * udp datagram
1080 */
1081 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1082 struct frag_hdr fhdr;
1083
1084 skb = sock_alloc_send_skb(sk,
1085 hh_len + fragheaderlen + transhdrlen + 20,
1086 (flags & MSG_DONTWAIT), &err);
1087 if (skb == NULL)
1088 return err;
1089
1090 /* reserve space for Hardware header */
1091 skb_reserve(skb, hh_len);
1092
1093 /* create space for UDP/IP header */
1094 skb_put(skb,fragheaderlen + transhdrlen);
1095
1096 /* initialize network header pointer */
1097 skb_reset_network_header(skb);
1098
1099 /* initialize protocol header pointer */
1100 skb->transport_header = skb->network_header + fragheaderlen;
1101
1102 skb->ip_summed = CHECKSUM_PARTIAL;
1103 skb->csum = 0;
1104
1105 /* Specify the length of each IPv6 datagram fragment.
1106 * It has to be a multiple of 8.
1107 */
1108 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1109 sizeof(struct frag_hdr)) & ~7;
1110 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1111 ipv6_select_ident(&fhdr, rt);
1112 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1113 __skb_queue_tail(&sk->sk_write_queue, skb);
1114 }
1115
1116 return skb_append_datato_frags(sk, skb, getfrag, from,
1117 (length - transhdrlen));
1118 }
1119
1120 static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1121 gfp_t gfp)
1122 {
1123 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1124 }
1125
1126 static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1127 gfp_t gfp)
1128 {
1129 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1130 }
1131
1132 static void ip6_append_data_mtu(unsigned int *mtu,
1133 int *maxfraglen,
1134 unsigned int fragheaderlen,
1135 struct sk_buff *skb,
1136 struct rt6_info *rt,
1137 unsigned int orig_mtu)
1138 {
1139 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1140 if (skb == NULL) {
1141 /* first fragment, reserve header_len */
1142 *mtu = orig_mtu - rt->dst.header_len;
1143
1144 } else {
1145 /*
1146 * this fragment is not first, the headers
1147 * space is regarded as data space.
1148 */
1149 *mtu = orig_mtu;
1150 }
1151 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1152 + fragheaderlen - sizeof(struct frag_hdr);
1153 }
1154 }
1155
1156 int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1157 int offset, int len, int odd, struct sk_buff *skb),
1158 void *from, int length, int transhdrlen,
1159 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1160 struct rt6_info *rt, unsigned int flags, int dontfrag)
1161 {
1162 struct inet_sock *inet = inet_sk(sk);
1163 struct ipv6_pinfo *np = inet6_sk(sk);
1164 struct inet_cork *cork;
1165 struct sk_buff *skb, *skb_prev = NULL;
1166 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1167 int exthdrlen;
1168 int dst_exthdrlen;
1169 int hh_len;
1170 int copy;
1171 int err;
1172 int offset = 0;
1173 __u8 tx_flags = 0;
1174
1175 if (flags&MSG_PROBE)
1176 return 0;
1177 cork = &inet->cork.base;
1178 if (skb_queue_empty(&sk->sk_write_queue)) {
1179 /*
1180 * setup for corking
1181 */
1182 if (opt) {
1183 if (WARN_ON(np->cork.opt))
1184 return -EINVAL;
1185
1186 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1187 if (unlikely(np->cork.opt == NULL))
1188 return -ENOBUFS;
1189
1190 np->cork.opt->tot_len = opt->tot_len;
1191 np->cork.opt->opt_flen = opt->opt_flen;
1192 np->cork.opt->opt_nflen = opt->opt_nflen;
1193
1194 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1195 sk->sk_allocation);
1196 if (opt->dst0opt && !np->cork.opt->dst0opt)
1197 return -ENOBUFS;
1198
1199 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1200 sk->sk_allocation);
1201 if (opt->dst1opt && !np->cork.opt->dst1opt)
1202 return -ENOBUFS;
1203
1204 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1205 sk->sk_allocation);
1206 if (opt->hopopt && !np->cork.opt->hopopt)
1207 return -ENOBUFS;
1208
1209 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1210 sk->sk_allocation);
1211 if (opt->srcrt && !np->cork.opt->srcrt)
1212 return -ENOBUFS;
1213
1214 /* need source address above miyazawa*/
1215 }
1216 dst_hold(&rt->dst);
1217 cork->dst = &rt->dst;
1218 inet->cork.fl.u.ip6 = *fl6;
1219 np->cork.hop_limit = hlimit;
1220 np->cork.tclass = tclass;
1221 if (rt->dst.flags & DST_XFRM_TUNNEL)
1222 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1223 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1224 else
1225 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1226 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1227 if (np->frag_size < mtu) {
1228 if (np->frag_size)
1229 mtu = np->frag_size;
1230 }
1231 cork->fragsize = mtu;
1232 if (dst_allfrag(rt->dst.path))
1233 cork->flags |= IPCORK_ALLFRAG;
1234 cork->length = 0;
1235 exthdrlen = (opt ? opt->opt_flen : 0);
1236 length += exthdrlen;
1237 transhdrlen += exthdrlen;
1238 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1239 } else {
1240 rt = (struct rt6_info *)cork->dst;
1241 fl6 = &inet->cork.fl.u.ip6;
1242 opt = np->cork.opt;
1243 transhdrlen = 0;
1244 exthdrlen = 0;
1245 dst_exthdrlen = 0;
1246 mtu = cork->fragsize;
1247 }
1248 orig_mtu = mtu;
1249
1250 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1251
1252 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1253 (opt ? opt->opt_nflen : 0);
1254 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1255
1256 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1257 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1258 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1259 return -EMSGSIZE;
1260 }
1261 }
1262
1263 /* For UDP, check if TX timestamp is enabled */
1264 if (sk->sk_type == SOCK_DGRAM)
1265 sock_tx_timestamp(sk, &tx_flags);
1266
1267 /*
1268 * Let's try using as much space as possible.
1269 * Use MTU if total length of the message fits into the MTU.
1270 * Otherwise, we need to reserve fragment header and
1271 * fragment alignment (= 8-15 octects, in total).
1272 *
1273 * Note that we may need to "move" the data from the tail of
1274 * of the buffer to the new fragment when we split
1275 * the message.
1276 *
1277 * FIXME: It may be fragmented into multiple chunks
1278 * at once if non-fragmentable extension headers
1279 * are too large.
1280 * --yoshfuji
1281 */
1282
1283 if ((length > mtu) && dontfrag && (sk->sk_protocol == IPPROTO_UDP ||
1284 sk->sk_protocol == IPPROTO_RAW)) {
1285 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1286 return -EMSGSIZE;
1287 }
1288
1289 skb = skb_peek_tail(&sk->sk_write_queue);
1290 cork->length += length;
1291 if ((skb && skb_has_frags(skb)) ||
1292 (((length + fragheaderlen) > mtu) &&
1293 (skb_queue_len(&sk->sk_write_queue) <= 1) &&
1294 (sk->sk_protocol == IPPROTO_UDP) &&
1295 (rt->dst.dev->features & NETIF_F_UFO) &&
1296 (sk->sk_type == SOCK_DGRAM))) {
1297 err = ip6_ufo_append_data(sk, getfrag, from, length,
1298 hh_len, fragheaderlen,
1299 transhdrlen, mtu, flags, rt);
1300 if (err)
1301 goto error;
1302 return 0;
1303 }
1304
1305 if (!skb)
1306 goto alloc_new_skb;
1307
1308 while (length > 0) {
1309 /* Check if the remaining data fits into current packet. */
1310 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1311 if (copy < length)
1312 copy = maxfraglen - skb->len;
1313
1314 if (copy <= 0) {
1315 char *data;
1316 unsigned int datalen;
1317 unsigned int fraglen;
1318 unsigned int fraggap;
1319 unsigned int alloclen;
1320 alloc_new_skb:
1321 /* There's no room in the current skb */
1322 if (skb)
1323 fraggap = skb->len - maxfraglen;
1324 else
1325 fraggap = 0;
1326 /* update mtu and maxfraglen if necessary */
1327 if (skb == NULL || skb_prev == NULL)
1328 ip6_append_data_mtu(&mtu, &maxfraglen,
1329 fragheaderlen, skb, rt,
1330 orig_mtu);
1331
1332 skb_prev = skb;
1333
1334 /*
1335 * If remaining data exceeds the mtu,
1336 * we know we need more fragment(s).
1337 */
1338 datalen = length + fraggap;
1339
1340 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1341 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1342 if ((flags & MSG_MORE) &&
1343 !(rt->dst.dev->features&NETIF_F_SG))
1344 alloclen = mtu;
1345 else
1346 alloclen = datalen + fragheaderlen;
1347
1348 alloclen += dst_exthdrlen;
1349
1350 if (datalen != length + fraggap) {
1351 /*
1352 * this is not the last fragment, the trailer
1353 * space is regarded as data space.
1354 */
1355 datalen += rt->dst.trailer_len;
1356 }
1357
1358 alloclen += rt->dst.trailer_len;
1359 fraglen = datalen + fragheaderlen;
1360
1361 /*
1362 * We just reserve space for fragment header.
1363 * Note: this may be overallocation if the message
1364 * (without MSG_MORE) fits into the MTU.
1365 */
1366 alloclen += sizeof(struct frag_hdr);
1367
1368 if (transhdrlen) {
1369 skb = sock_alloc_send_skb(sk,
1370 alloclen + hh_len,
1371 (flags & MSG_DONTWAIT), &err);
1372 } else {
1373 skb = NULL;
1374 if (atomic_read(&sk->sk_wmem_alloc) <=
1375 2 * sk->sk_sndbuf)
1376 skb = sock_wmalloc(sk,
1377 alloclen + hh_len, 1,
1378 sk->sk_allocation);
1379 if (unlikely(skb == NULL))
1380 err = -ENOBUFS;
1381 else {
1382 /* Only the initial fragment
1383 * is time stamped.
1384 */
1385 tx_flags = 0;
1386 }
1387 }
1388 if (skb == NULL)
1389 goto error;
1390 /*
1391 * Fill in the control structures
1392 */
1393 skb->ip_summed = CHECKSUM_NONE;
1394 skb->csum = 0;
1395 /* reserve for fragmentation and ipsec header */
1396 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1397 dst_exthdrlen);
1398
1399 if (sk->sk_type == SOCK_DGRAM)
1400 skb_shinfo(skb)->tx_flags = tx_flags;
1401
1402 /*
1403 * Find where to start putting bytes
1404 */
1405 data = skb_put(skb, fraglen);
1406 skb_set_network_header(skb, exthdrlen);
1407 data += fragheaderlen;
1408 skb->transport_header = (skb->network_header +
1409 fragheaderlen);
1410 if (fraggap) {
1411 skb->csum = skb_copy_and_csum_bits(
1412 skb_prev, maxfraglen,
1413 data + transhdrlen, fraggap, 0);
1414 skb_prev->csum = csum_sub(skb_prev->csum,
1415 skb->csum);
1416 data += fraggap;
1417 pskb_trim_unique(skb_prev, maxfraglen);
1418 }
1419 copy = datalen - transhdrlen - fraggap;
1420
1421 if (copy < 0) {
1422 err = -EINVAL;
1423 kfree_skb(skb);
1424 goto error;
1425 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1426 err = -EFAULT;
1427 kfree_skb(skb);
1428 goto error;
1429 }
1430
1431 offset += copy;
1432 length -= datalen - fraggap;
1433 transhdrlen = 0;
1434 exthdrlen = 0;
1435 dst_exthdrlen = 0;
1436
1437 /*
1438 * Put the packet on the pending queue
1439 */
1440 __skb_queue_tail(&sk->sk_write_queue, skb);
1441 continue;
1442 }
1443
1444 if (copy > length)
1445 copy = length;
1446
1447 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1448 unsigned int off;
1449
1450 off = skb->len;
1451 if (getfrag(from, skb_put(skb, copy),
1452 offset, copy, off, skb) < 0) {
1453 __skb_trim(skb, off);
1454 err = -EFAULT;
1455 goto error;
1456 }
1457 } else {
1458 int i = skb_shinfo(skb)->nr_frags;
1459 struct page_frag *pfrag = sk_page_frag(sk);
1460
1461 err = -ENOMEM;
1462 if (!sk_page_frag_refill(sk, pfrag))
1463 goto error;
1464
1465 if (!skb_can_coalesce(skb, i, pfrag->page,
1466 pfrag->offset)) {
1467 err = -EMSGSIZE;
1468 if (i == MAX_SKB_FRAGS)
1469 goto error;
1470
1471 __skb_fill_page_desc(skb, i, pfrag->page,
1472 pfrag->offset, 0);
1473 skb_shinfo(skb)->nr_frags = ++i;
1474 get_page(pfrag->page);
1475 }
1476 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1477 if (getfrag(from,
1478 page_address(pfrag->page) + pfrag->offset,
1479 offset, copy, skb->len, skb) < 0)
1480 goto error_efault;
1481
1482 pfrag->offset += copy;
1483 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1484 skb->len += copy;
1485 skb->data_len += copy;
1486 skb->truesize += copy;
1487 atomic_add(copy, &sk->sk_wmem_alloc);
1488 }
1489 offset += copy;
1490 length -= copy;
1491 }
1492
1493 return 0;
1494
1495 error_efault:
1496 err = -EFAULT;
1497 error:
1498 cork->length -= length;
1499 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1500 return err;
1501 }
1502 EXPORT_SYMBOL_GPL(ip6_append_data);
1503
1504 static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1505 {
1506 if (np->cork.opt) {
1507 kfree(np->cork.opt->dst0opt);
1508 kfree(np->cork.opt->dst1opt);
1509 kfree(np->cork.opt->hopopt);
1510 kfree(np->cork.opt->srcrt);
1511 kfree(np->cork.opt);
1512 np->cork.opt = NULL;
1513 }
1514
1515 if (inet->cork.base.dst) {
1516 dst_release(inet->cork.base.dst);
1517 inet->cork.base.dst = NULL;
1518 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1519 }
1520 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1521 }
1522
1523 int ip6_push_pending_frames(struct sock *sk)
1524 {
1525 struct sk_buff *skb, *tmp_skb;
1526 struct sk_buff **tail_skb;
1527 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1528 struct inet_sock *inet = inet_sk(sk);
1529 struct ipv6_pinfo *np = inet6_sk(sk);
1530 struct net *net = sock_net(sk);
1531 struct ipv6hdr *hdr;
1532 struct ipv6_txoptions *opt = np->cork.opt;
1533 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1534 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1535 unsigned char proto = fl6->flowi6_proto;
1536 int err = 0;
1537
1538 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1539 goto out;
1540 tail_skb = &(skb_shinfo(skb)->frag_list);
1541
1542 /* move skb->data to ip header from ext header */
1543 if (skb->data < skb_network_header(skb))
1544 __skb_pull(skb, skb_network_offset(skb));
1545 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1546 __skb_pull(tmp_skb, skb_network_header_len(skb));
1547 *tail_skb = tmp_skb;
1548 tail_skb = &(tmp_skb->next);
1549 skb->len += tmp_skb->len;
1550 skb->data_len += tmp_skb->len;
1551 skb->truesize += tmp_skb->truesize;
1552 tmp_skb->destructor = NULL;
1553 tmp_skb->sk = NULL;
1554 }
1555
1556 /* Allow local fragmentation. */
1557 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1558 skb->local_df = 1;
1559
1560 *final_dst = fl6->daddr;
1561 __skb_pull(skb, skb_network_header_len(skb));
1562 if (opt && opt->opt_flen)
1563 ipv6_push_frag_opts(skb, opt, &proto);
1564 if (opt && opt->opt_nflen)
1565 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1566
1567 skb_push(skb, sizeof(struct ipv6hdr));
1568 skb_reset_network_header(skb);
1569 hdr = ipv6_hdr(skb);
1570
1571 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1572 hdr->hop_limit = np->cork.hop_limit;
1573 hdr->nexthdr = proto;
1574 hdr->saddr = fl6->saddr;
1575 hdr->daddr = *final_dst;
1576
1577 skb->priority = sk->sk_priority;
1578 skb->mark = sk->sk_mark;
1579
1580 skb_dst_set(skb, dst_clone(&rt->dst));
1581 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1582 if (proto == IPPROTO_ICMPV6) {
1583 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1584
1585 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1586 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1587 }
1588
1589 err = ip6_local_out(skb);
1590 if (err) {
1591 if (err > 0)
1592 err = net_xmit_errno(err);
1593 if (err)
1594 goto error;
1595 }
1596
1597 out:
1598 ip6_cork_release(inet, np);
1599 return err;
1600 error:
1601 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1602 goto out;
1603 }
1604 EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1605
1606 void ip6_flush_pending_frames(struct sock *sk)
1607 {
1608 struct sk_buff *skb;
1609
1610 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1611 if (skb_dst(skb))
1612 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1613 IPSTATS_MIB_OUTDISCARDS);
1614 kfree_skb(skb);
1615 }
1616
1617 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1618 }
1619 EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1620
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.