Loading...
1/*
2 * IPv6 output functions
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/net/ipv4/ip_output.c
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 *
15 * Changes:
16 * A.N.Kuznetsov : airthmetics in fragmentation.
17 * extension headers are implemented.
18 * route changes now work.
19 * ip6_forward does not confuse sniffers.
20 * etc.
21 *
22 * H. von Brand : Added missing #include <linux/string.h>
23 * Imran Patel : frag id should be in NBO
24 * Kazunori MIYAZAWA @USAGI
25 * : add ip6_append_data and related functions
26 * for datagram xmit
27 */
28
29#include <linux/errno.h>
30#include <linux/kernel.h>
31#include <linux/string.h>
32#include <linux/socket.h>
33#include <linux/net.h>
34#include <linux/netdevice.h>
35#include <linux/if_arp.h>
36#include <linux/in6.h>
37#include <linux/tcp.h>
38#include <linux/route.h>
39#include <linux/module.h>
40#include <linux/slab.h>
41
42#include <linux/netfilter.h>
43#include <linux/netfilter_ipv6.h>
44
45#include <net/sock.h>
46#include <net/snmp.h>
47
48#include <net/ipv6.h>
49#include <net/ndisc.h>
50#include <net/protocol.h>
51#include <net/ip6_route.h>
52#include <net/addrconf.h>
53#include <net/rawv6.h>
54#include <net/icmp.h>
55#include <net/xfrm.h>
56#include <net/checksum.h>
57#include <linux/mroute6.h>
58
59int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
60
61int __ip6_local_out(struct sk_buff *skb)
62{
63 int len;
64
65 len = skb->len - sizeof(struct ipv6hdr);
66 if (len > IPV6_MAXPLEN)
67 len = 0;
68 ipv6_hdr(skb)->payload_len = htons(len);
69
70 return nf_hook(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
71 skb_dst(skb)->dev, dst_output);
72}
73
74int ip6_local_out(struct sk_buff *skb)
75{
76 int err;
77
78 err = __ip6_local_out(skb);
79 if (likely(err == 1))
80 err = dst_output(skb);
81
82 return err;
83}
84EXPORT_SYMBOL_GPL(ip6_local_out);
85
86/* dev_loopback_xmit for use with netfilter. */
87static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
88{
89 skb_reset_mac_header(newskb);
90 __skb_pull(newskb, skb_network_offset(newskb));
91 newskb->pkt_type = PACKET_LOOPBACK;
92 newskb->ip_summed = CHECKSUM_UNNECESSARY;
93 WARN_ON(!skb_dst(newskb));
94
95 netif_rx_ni(newskb);
96 return 0;
97}
98
99static int ip6_finish_output2(struct sk_buff *skb)
100{
101 struct dst_entry *dst = skb_dst(skb);
102 struct net_device *dev = dst->dev;
103 struct neighbour *neigh;
104
105 skb->protocol = htons(ETH_P_IPV6);
106 skb->dev = dev;
107
108 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
109 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
110
111 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
112 ((mroute6_socket(dev_net(dev), skb) &&
113 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
114 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
115 &ipv6_hdr(skb)->saddr))) {
116 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
117
118 /* Do not check for IFF_ALLMULTI; multicast routing
119 is not supported in any case.
120 */
121 if (newskb)
122 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
123 newskb, NULL, newskb->dev,
124 ip6_dev_loopback_xmit);
125
126 if (ipv6_hdr(skb)->hop_limit == 0) {
127 IP6_INC_STATS(dev_net(dev), idev,
128 IPSTATS_MIB_OUTDISCARDS);
129 kfree_skb(skb);
130 return 0;
131 }
132 }
133
134 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
135 skb->len);
136 }
137
138 rcu_read_lock();
139 neigh = dst_get_neighbour_noref(dst);
140 if (neigh) {
141 int res = neigh_output(neigh, skb);
142
143 rcu_read_unlock();
144 return res;
145 }
146 rcu_read_unlock();
147 IP6_INC_STATS_BH(dev_net(dst->dev),
148 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
149 kfree_skb(skb);
150 return -EINVAL;
151}
152
153static int ip6_finish_output(struct sk_buff *skb)
154{
155 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
156 dst_allfrag(skb_dst(skb)))
157 return ip6_fragment(skb, ip6_finish_output2);
158 else
159 return ip6_finish_output2(skb);
160}
161
162int ip6_output(struct sk_buff *skb)
163{
164 struct net_device *dev = skb_dst(skb)->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
166 if (unlikely(idev->cnf.disable_ipv6)) {
167 IP6_INC_STATS(dev_net(dev), idev,
168 IPSTATS_MIB_OUTDISCARDS);
169 kfree_skb(skb);
170 return 0;
171 }
172
173 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
174 ip6_finish_output,
175 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
176}
177
178/*
179 * xmit an sk_buff (used by TCP, SCTP and DCCP)
180 */
181
182int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
183 struct ipv6_txoptions *opt, int tclass)
184{
185 struct net *net = sock_net(sk);
186 struct ipv6_pinfo *np = inet6_sk(sk);
187 struct in6_addr *first_hop = &fl6->daddr;
188 struct dst_entry *dst = skb_dst(skb);
189 struct ipv6hdr *hdr;
190 u8 proto = fl6->flowi6_proto;
191 int seg_len = skb->len;
192 int hlimit = -1;
193 u32 mtu;
194
195 if (opt) {
196 unsigned int head_room;
197
198 /* First: exthdrs may take lots of space (~8K for now)
199 MAX_HEADER is not enough.
200 */
201 head_room = opt->opt_nflen + opt->opt_flen;
202 seg_len += head_room;
203 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
204
205 if (skb_headroom(skb) < head_room) {
206 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
207 if (skb2 == NULL) {
208 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
209 IPSTATS_MIB_OUTDISCARDS);
210 kfree_skb(skb);
211 return -ENOBUFS;
212 }
213 consume_skb(skb);
214 skb = skb2;
215 skb_set_owner_w(skb, sk);
216 }
217 if (opt->opt_flen)
218 ipv6_push_frag_opts(skb, opt, &proto);
219 if (opt->opt_nflen)
220 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
221 }
222
223 skb_push(skb, sizeof(struct ipv6hdr));
224 skb_reset_network_header(skb);
225 hdr = ipv6_hdr(skb);
226
227 /*
228 * Fill in the IPv6 header
229 */
230 if (np)
231 hlimit = np->hop_limit;
232 if (hlimit < 0)
233 hlimit = ip6_dst_hoplimit(dst);
234
235 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl6->flowlabel;
236
237 hdr->payload_len = htons(seg_len);
238 hdr->nexthdr = proto;
239 hdr->hop_limit = hlimit;
240
241 hdr->saddr = fl6->saddr;
242 hdr->daddr = *first_hop;
243
244 skb->priority = sk->sk_priority;
245 skb->mark = sk->sk_mark;
246
247 mtu = dst_mtu(dst);
248 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
249 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
250 IPSTATS_MIB_OUT, skb->len);
251 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
252 dst->dev, dst_output);
253 }
254
255 net_dbg_ratelimited("IPv6: sending pkt_too_big to self\n");
256 skb->dev = dst->dev;
257 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
258 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
259 kfree_skb(skb);
260 return -EMSGSIZE;
261}
262
263EXPORT_SYMBOL(ip6_xmit);
264
265/*
266 * To avoid extra problems ND packets are send through this
267 * routine. It's code duplication but I really want to avoid
268 * extra checks since ipv6_build_header is used by TCP (which
269 * is for us performance critical)
270 */
271
272int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
273 const struct in6_addr *saddr, const struct in6_addr *daddr,
274 int proto, int len)
275{
276 struct ipv6_pinfo *np = inet6_sk(sk);
277 struct ipv6hdr *hdr;
278
279 skb->protocol = htons(ETH_P_IPV6);
280 skb->dev = dev;
281
282 skb_reset_network_header(skb);
283 skb_put(skb, sizeof(struct ipv6hdr));
284 hdr = ipv6_hdr(skb);
285
286 *(__be32*)hdr = htonl(0x60000000);
287
288 hdr->payload_len = htons(len);
289 hdr->nexthdr = proto;
290 hdr->hop_limit = np->hop_limit;
291
292 hdr->saddr = *saddr;
293 hdr->daddr = *daddr;
294
295 return 0;
296}
297
298static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
299{
300 struct ip6_ra_chain *ra;
301 struct sock *last = NULL;
302
303 read_lock(&ip6_ra_lock);
304 for (ra = ip6_ra_chain; ra; ra = ra->next) {
305 struct sock *sk = ra->sk;
306 if (sk && ra->sel == sel &&
307 (!sk->sk_bound_dev_if ||
308 sk->sk_bound_dev_if == skb->dev->ifindex)) {
309 if (last) {
310 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
311 if (skb2)
312 rawv6_rcv(last, skb2);
313 }
314 last = sk;
315 }
316 }
317
318 if (last) {
319 rawv6_rcv(last, skb);
320 read_unlock(&ip6_ra_lock);
321 return 1;
322 }
323 read_unlock(&ip6_ra_lock);
324 return 0;
325}
326
327static int ip6_forward_proxy_check(struct sk_buff *skb)
328{
329 struct ipv6hdr *hdr = ipv6_hdr(skb);
330 u8 nexthdr = hdr->nexthdr;
331 __be16 frag_off;
332 int offset;
333
334 if (ipv6_ext_hdr(nexthdr)) {
335 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
336 if (offset < 0)
337 return 0;
338 } else
339 offset = sizeof(struct ipv6hdr);
340
341 if (nexthdr == IPPROTO_ICMPV6) {
342 struct icmp6hdr *icmp6;
343
344 if (!pskb_may_pull(skb, (skb_network_header(skb) +
345 offset + 1 - skb->data)))
346 return 0;
347
348 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
349
350 switch (icmp6->icmp6_type) {
351 case NDISC_ROUTER_SOLICITATION:
352 case NDISC_ROUTER_ADVERTISEMENT:
353 case NDISC_NEIGHBOUR_SOLICITATION:
354 case NDISC_NEIGHBOUR_ADVERTISEMENT:
355 case NDISC_REDIRECT:
356 /* For reaction involving unicast neighbor discovery
357 * message destined to the proxied address, pass it to
358 * input function.
359 */
360 return 1;
361 default:
362 break;
363 }
364 }
365
366 /*
367 * The proxying router can't forward traffic sent to a link-local
368 * address, so signal the sender and discard the packet. This
369 * behavior is clarified by the MIPv6 specification.
370 */
371 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
372 dst_link_failure(skb);
373 return -1;
374 }
375
376 return 0;
377}
378
379static inline int ip6_forward_finish(struct sk_buff *skb)
380{
381 return dst_output(skb);
382}
383
384int ip6_forward(struct sk_buff *skb)
385{
386 struct dst_entry *dst = skb_dst(skb);
387 struct ipv6hdr *hdr = ipv6_hdr(skb);
388 struct inet6_skb_parm *opt = IP6CB(skb);
389 struct net *net = dev_net(dst->dev);
390 u32 mtu;
391
392 if (net->ipv6.devconf_all->forwarding == 0)
393 goto error;
394
395 if (skb_warn_if_lro(skb))
396 goto drop;
397
398 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
399 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
400 goto drop;
401 }
402
403 if (skb->pkt_type != PACKET_HOST)
404 goto drop;
405
406 skb_forward_csum(skb);
407
408 /*
409 * We DO NOT make any processing on
410 * RA packets, pushing them to user level AS IS
411 * without ane WARRANTY that application will be able
412 * to interpret them. The reason is that we
413 * cannot make anything clever here.
414 *
415 * We are not end-node, so that if packet contains
416 * AH/ESP, we cannot make anything.
417 * Defragmentation also would be mistake, RA packets
418 * cannot be fragmented, because there is no warranty
419 * that different fragments will go along one path. --ANK
420 */
421 if (opt->ra) {
422 u8 *ptr = skb_network_header(skb) + opt->ra;
423 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
424 return 0;
425 }
426
427 /*
428 * check and decrement ttl
429 */
430 if (hdr->hop_limit <= 1) {
431 /* Force OUTPUT device used as source address */
432 skb->dev = dst->dev;
433 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
434 IP6_INC_STATS_BH(net,
435 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
436
437 kfree_skb(skb);
438 return -ETIMEDOUT;
439 }
440
441 /* XXX: idev->cnf.proxy_ndp? */
442 if (net->ipv6.devconf_all->proxy_ndp &&
443 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
444 int proxied = ip6_forward_proxy_check(skb);
445 if (proxied > 0)
446 return ip6_input(skb);
447 else if (proxied < 0) {
448 IP6_INC_STATS(net, ip6_dst_idev(dst),
449 IPSTATS_MIB_INDISCARDS);
450 goto drop;
451 }
452 }
453
454 if (!xfrm6_route_forward(skb)) {
455 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
456 goto drop;
457 }
458 dst = skb_dst(skb);
459
460 /* IPv6 specs say nothing about it, but it is clear that we cannot
461 send redirects to source routed frames.
462 We don't send redirects to frames decapsulated from IPsec.
463 */
464 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
465 struct in6_addr *target = NULL;
466 struct rt6_info *rt;
467
468 /*
469 * incoming and outgoing devices are the same
470 * send a redirect.
471 */
472
473 rt = (struct rt6_info *) dst;
474 if (rt->rt6i_flags & RTF_GATEWAY)
475 target = &rt->rt6i_gateway;
476 else
477 target = &hdr->daddr;
478
479 if (!rt->rt6i_peer)
480 rt6_bind_peer(rt, 1);
481
482 /* Limit redirects both by destination (here)
483 and by source (inside ndisc_send_redirect)
484 */
485 if (inet_peer_xrlim_allow(rt->rt6i_peer, 1*HZ))
486 ndisc_send_redirect(skb, target);
487 } else {
488 int addrtype = ipv6_addr_type(&hdr->saddr);
489
490 /* This check is security critical. */
491 if (addrtype == IPV6_ADDR_ANY ||
492 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
493 goto error;
494 if (addrtype & IPV6_ADDR_LINKLOCAL) {
495 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
496 ICMPV6_NOT_NEIGHBOUR, 0);
497 goto error;
498 }
499 }
500
501 mtu = dst_mtu(dst);
502 if (mtu < IPV6_MIN_MTU)
503 mtu = IPV6_MIN_MTU;
504
505 if (skb->len > mtu && !skb_is_gso(skb)) {
506 /* Again, force OUTPUT device used as source address */
507 skb->dev = dst->dev;
508 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
509 IP6_INC_STATS_BH(net,
510 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
511 IP6_INC_STATS_BH(net,
512 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
513 kfree_skb(skb);
514 return -EMSGSIZE;
515 }
516
517 if (skb_cow(skb, dst->dev->hard_header_len)) {
518 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
519 goto drop;
520 }
521
522 hdr = ipv6_hdr(skb);
523
524 /* Mangling hops number delayed to point after skb COW */
525
526 hdr->hop_limit--;
527
528 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
529 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
530 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
531 ip6_forward_finish);
532
533error:
534 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
535drop:
536 kfree_skb(skb);
537 return -EINVAL;
538}
539
540static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
541{
542 to->pkt_type = from->pkt_type;
543 to->priority = from->priority;
544 to->protocol = from->protocol;
545 skb_dst_drop(to);
546 skb_dst_set(to, dst_clone(skb_dst(from)));
547 to->dev = from->dev;
548 to->mark = from->mark;
549
550#ifdef CONFIG_NET_SCHED
551 to->tc_index = from->tc_index;
552#endif
553 nf_copy(to, from);
554#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
555 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
556 to->nf_trace = from->nf_trace;
557#endif
558 skb_copy_secmark(to, from);
559}
560
561int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
562{
563 u16 offset = sizeof(struct ipv6hdr);
564 struct ipv6_opt_hdr *exthdr =
565 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
566 unsigned int packet_len = skb->tail - skb->network_header;
567 int found_rhdr = 0;
568 *nexthdr = &ipv6_hdr(skb)->nexthdr;
569
570 while (offset + 1 <= packet_len) {
571
572 switch (**nexthdr) {
573
574 case NEXTHDR_HOP:
575 break;
576 case NEXTHDR_ROUTING:
577 found_rhdr = 1;
578 break;
579 case NEXTHDR_DEST:
580#if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
581 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
582 break;
583#endif
584 if (found_rhdr)
585 return offset;
586 break;
587 default :
588 return offset;
589 }
590
591 offset += ipv6_optlen(exthdr);
592 *nexthdr = &exthdr->nexthdr;
593 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
594 offset);
595 }
596
597 return offset;
598}
599
600void ipv6_select_ident(struct frag_hdr *fhdr, struct rt6_info *rt)
601{
602 static atomic_t ipv6_fragmentation_id;
603 int old, new;
604
605 if (rt && !(rt->dst.flags & DST_NOPEER)) {
606 struct inet_peer *peer;
607
608 if (!rt->rt6i_peer)
609 rt6_bind_peer(rt, 1);
610 peer = rt->rt6i_peer;
611 if (peer) {
612 fhdr->identification = htonl(inet_getid(peer, 0));
613 return;
614 }
615 }
616 do {
617 old = atomic_read(&ipv6_fragmentation_id);
618 new = old + 1;
619 if (!new)
620 new = 1;
621 } while (atomic_cmpxchg(&ipv6_fragmentation_id, old, new) != old);
622 fhdr->identification = htonl(new);
623}
624
625int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
626{
627 struct sk_buff *frag;
628 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
629 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
630 struct ipv6hdr *tmp_hdr;
631 struct frag_hdr *fh;
632 unsigned int mtu, hlen, left, len;
633 int hroom, troom;
634 __be32 frag_id = 0;
635 int ptr, offset = 0, err=0;
636 u8 *prevhdr, nexthdr = 0;
637 struct net *net = dev_net(skb_dst(skb)->dev);
638
639 hlen = ip6_find_1stfragopt(skb, &prevhdr);
640 nexthdr = *prevhdr;
641
642 mtu = ip6_skb_dst_mtu(skb);
643
644 /* We must not fragment if the socket is set to force MTU discovery
645 * or if the skb it not generated by a local socket.
646 */
647 if (unlikely(!skb->local_df && skb->len > mtu)) {
648 if (skb->sk && dst_allfrag(skb_dst(skb)))
649 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
650
651 skb->dev = skb_dst(skb)->dev;
652 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
653 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
654 IPSTATS_MIB_FRAGFAILS);
655 kfree_skb(skb);
656 return -EMSGSIZE;
657 }
658
659 if (np && np->frag_size < mtu) {
660 if (np->frag_size)
661 mtu = np->frag_size;
662 }
663 mtu -= hlen + sizeof(struct frag_hdr);
664
665 if (skb_has_frag_list(skb)) {
666 int first_len = skb_pagelen(skb);
667 struct sk_buff *frag2;
668
669 if (first_len - hlen > mtu ||
670 ((first_len - hlen) & 7) ||
671 skb_cloned(skb))
672 goto slow_path;
673
674 skb_walk_frags(skb, frag) {
675 /* Correct geometry. */
676 if (frag->len > mtu ||
677 ((frag->len & 7) && frag->next) ||
678 skb_headroom(frag) < hlen)
679 goto slow_path_clean;
680
681 /* Partially cloned skb? */
682 if (skb_shared(frag))
683 goto slow_path_clean;
684
685 BUG_ON(frag->sk);
686 if (skb->sk) {
687 frag->sk = skb->sk;
688 frag->destructor = sock_wfree;
689 }
690 skb->truesize -= frag->truesize;
691 }
692
693 err = 0;
694 offset = 0;
695 frag = skb_shinfo(skb)->frag_list;
696 skb_frag_list_init(skb);
697 /* BUILD HEADER */
698
699 *prevhdr = NEXTHDR_FRAGMENT;
700 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
701 if (!tmp_hdr) {
702 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
703 IPSTATS_MIB_FRAGFAILS);
704 return -ENOMEM;
705 }
706
707 __skb_pull(skb, hlen);
708 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
709 __skb_push(skb, hlen);
710 skb_reset_network_header(skb);
711 memcpy(skb_network_header(skb), tmp_hdr, hlen);
712
713 ipv6_select_ident(fh, rt);
714 fh->nexthdr = nexthdr;
715 fh->reserved = 0;
716 fh->frag_off = htons(IP6_MF);
717 frag_id = fh->identification;
718
719 first_len = skb_pagelen(skb);
720 skb->data_len = first_len - skb_headlen(skb);
721 skb->len = first_len;
722 ipv6_hdr(skb)->payload_len = htons(first_len -
723 sizeof(struct ipv6hdr));
724
725 dst_hold(&rt->dst);
726
727 for (;;) {
728 /* Prepare header of the next frame,
729 * before previous one went down. */
730 if (frag) {
731 frag->ip_summed = CHECKSUM_NONE;
732 skb_reset_transport_header(frag);
733 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
734 __skb_push(frag, hlen);
735 skb_reset_network_header(frag);
736 memcpy(skb_network_header(frag), tmp_hdr,
737 hlen);
738 offset += skb->len - hlen - sizeof(struct frag_hdr);
739 fh->nexthdr = nexthdr;
740 fh->reserved = 0;
741 fh->frag_off = htons(offset);
742 if (frag->next != NULL)
743 fh->frag_off |= htons(IP6_MF);
744 fh->identification = frag_id;
745 ipv6_hdr(frag)->payload_len =
746 htons(frag->len -
747 sizeof(struct ipv6hdr));
748 ip6_copy_metadata(frag, skb);
749 }
750
751 err = output(skb);
752 if(!err)
753 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
754 IPSTATS_MIB_FRAGCREATES);
755
756 if (err || !frag)
757 break;
758
759 skb = frag;
760 frag = skb->next;
761 skb->next = NULL;
762 }
763
764 kfree(tmp_hdr);
765
766 if (err == 0) {
767 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
768 IPSTATS_MIB_FRAGOKS);
769 dst_release(&rt->dst);
770 return 0;
771 }
772
773 while (frag) {
774 skb = frag->next;
775 kfree_skb(frag);
776 frag = skb;
777 }
778
779 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
780 IPSTATS_MIB_FRAGFAILS);
781 dst_release(&rt->dst);
782 return err;
783
784slow_path_clean:
785 skb_walk_frags(skb, frag2) {
786 if (frag2 == frag)
787 break;
788 frag2->sk = NULL;
789 frag2->destructor = NULL;
790 skb->truesize += frag2->truesize;
791 }
792 }
793
794slow_path:
795 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
796 skb_checksum_help(skb))
797 goto fail;
798
799 left = skb->len - hlen; /* Space per frame */
800 ptr = hlen; /* Where to start from */
801
802 /*
803 * Fragment the datagram.
804 */
805
806 *prevhdr = NEXTHDR_FRAGMENT;
807 hroom = LL_RESERVED_SPACE(rt->dst.dev);
808 troom = rt->dst.dev->needed_tailroom;
809
810 /*
811 * Keep copying data until we run out.
812 */
813 while(left > 0) {
814 len = left;
815 /* IF: it doesn't fit, use 'mtu' - the data space left */
816 if (len > mtu)
817 len = mtu;
818 /* IF: we are not sending up to and including the packet end
819 then align the next start on an eight byte boundary */
820 if (len < left) {
821 len &= ~7;
822 }
823 /*
824 * Allocate buffer.
825 */
826
827 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
828 hroom + troom, GFP_ATOMIC)) == NULL) {
829 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
830 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
831 IPSTATS_MIB_FRAGFAILS);
832 err = -ENOMEM;
833 goto fail;
834 }
835
836 /*
837 * Set up data on packet
838 */
839
840 ip6_copy_metadata(frag, skb);
841 skb_reserve(frag, hroom);
842 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
843 skb_reset_network_header(frag);
844 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
845 frag->transport_header = (frag->network_header + hlen +
846 sizeof(struct frag_hdr));
847
848 /*
849 * Charge the memory for the fragment to any owner
850 * it might possess
851 */
852 if (skb->sk)
853 skb_set_owner_w(frag, skb->sk);
854
855 /*
856 * Copy the packet header into the new buffer.
857 */
858 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
859
860 /*
861 * Build fragment header.
862 */
863 fh->nexthdr = nexthdr;
864 fh->reserved = 0;
865 if (!frag_id) {
866 ipv6_select_ident(fh, rt);
867 frag_id = fh->identification;
868 } else
869 fh->identification = frag_id;
870
871 /*
872 * Copy a block of the IP datagram.
873 */
874 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
875 BUG();
876 left -= len;
877
878 fh->frag_off = htons(offset);
879 if (left > 0)
880 fh->frag_off |= htons(IP6_MF);
881 ipv6_hdr(frag)->payload_len = htons(frag->len -
882 sizeof(struct ipv6hdr));
883
884 ptr += len;
885 offset += len;
886
887 /*
888 * Put this fragment into the sending queue.
889 */
890 err = output(frag);
891 if (err)
892 goto fail;
893
894 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
895 IPSTATS_MIB_FRAGCREATES);
896 }
897 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
898 IPSTATS_MIB_FRAGOKS);
899 consume_skb(skb);
900 return err;
901
902fail:
903 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
904 IPSTATS_MIB_FRAGFAILS);
905 kfree_skb(skb);
906 return err;
907}
908
909static inline int ip6_rt_check(const struct rt6key *rt_key,
910 const struct in6_addr *fl_addr,
911 const struct in6_addr *addr_cache)
912{
913 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
914 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
915}
916
917static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
918 struct dst_entry *dst,
919 const struct flowi6 *fl6)
920{
921 struct ipv6_pinfo *np = inet6_sk(sk);
922 struct rt6_info *rt = (struct rt6_info *)dst;
923
924 if (!dst)
925 goto out;
926
927 /* Yes, checking route validity in not connected
928 * case is not very simple. Take into account,
929 * that we do not support routing by source, TOS,
930 * and MSG_DONTROUTE --ANK (980726)
931 *
932 * 1. ip6_rt_check(): If route was host route,
933 * check that cached destination is current.
934 * If it is network route, we still may
935 * check its validity using saved pointer
936 * to the last used address: daddr_cache.
937 * We do not want to save whole address now,
938 * (because main consumer of this service
939 * is tcp, which has not this problem),
940 * so that the last trick works only on connected
941 * sockets.
942 * 2. oif also should be the same.
943 */
944 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
945#ifdef CONFIG_IPV6_SUBTREES
946 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
947#endif
948 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
949 dst_release(dst);
950 dst = NULL;
951 }
952
953out:
954 return dst;
955}
956
957static int ip6_dst_lookup_tail(struct sock *sk,
958 struct dst_entry **dst, struct flowi6 *fl6)
959{
960 struct net *net = sock_net(sk);
961#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
962 struct neighbour *n;
963#endif
964 int err;
965
966 if (*dst == NULL)
967 *dst = ip6_route_output(net, sk, fl6);
968
969 if ((err = (*dst)->error))
970 goto out_err_release;
971
972 if (ipv6_addr_any(&fl6->saddr)) {
973 struct rt6_info *rt = (struct rt6_info *) *dst;
974 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
975 sk ? inet6_sk(sk)->srcprefs : 0,
976 &fl6->saddr);
977 if (err)
978 goto out_err_release;
979 }
980
981#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
982 /*
983 * Here if the dst entry we've looked up
984 * has a neighbour entry that is in the INCOMPLETE
985 * state and the src address from the flow is
986 * marked as OPTIMISTIC, we release the found
987 * dst entry and replace it instead with the
988 * dst entry of the nexthop router
989 */
990 rcu_read_lock();
991 n = dst_get_neighbour_noref(*dst);
992 if (n && !(n->nud_state & NUD_VALID)) {
993 struct inet6_ifaddr *ifp;
994 struct flowi6 fl_gw6;
995 int redirect;
996
997 rcu_read_unlock();
998 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
999 (*dst)->dev, 1);
1000
1001 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1002 if (ifp)
1003 in6_ifa_put(ifp);
1004
1005 if (redirect) {
1006 /*
1007 * We need to get the dst entry for the
1008 * default router instead
1009 */
1010 dst_release(*dst);
1011 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1012 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1013 *dst = ip6_route_output(net, sk, &fl_gw6);
1014 if ((err = (*dst)->error))
1015 goto out_err_release;
1016 }
1017 } else {
1018 rcu_read_unlock();
1019 }
1020#endif
1021
1022 return 0;
1023
1024out_err_release:
1025 if (err == -ENETUNREACH)
1026 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1027 dst_release(*dst);
1028 *dst = NULL;
1029 return err;
1030}
1031
1032/**
1033 * ip6_dst_lookup - perform route lookup on flow
1034 * @sk: socket which provides route info
1035 * @dst: pointer to dst_entry * for result
1036 * @fl6: flow to lookup
1037 *
1038 * This function performs a route lookup on the given flow.
1039 *
1040 * It returns zero on success, or a standard errno code on error.
1041 */
1042int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
1043{
1044 *dst = NULL;
1045 return ip6_dst_lookup_tail(sk, dst, fl6);
1046}
1047EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1048
1049/**
1050 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1051 * @sk: socket which provides route info
1052 * @fl6: flow to lookup
1053 * @final_dst: final destination address for ipsec lookup
1054 * @can_sleep: we are in a sleepable context
1055 *
1056 * This function performs a route lookup on the given flow.
1057 *
1058 * It returns a valid dst pointer on success, or a pointer encoded
1059 * error code.
1060 */
1061struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1062 const struct in6_addr *final_dst,
1063 bool can_sleep)
1064{
1065 struct dst_entry *dst = NULL;
1066 int err;
1067
1068 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1069 if (err)
1070 return ERR_PTR(err);
1071 if (final_dst)
1072 fl6->daddr = *final_dst;
1073 if (can_sleep)
1074 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1075
1076 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1077}
1078EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1079
1080/**
1081 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1082 * @sk: socket which provides the dst cache and route info
1083 * @fl6: flow to lookup
1084 * @final_dst: final destination address for ipsec lookup
1085 * @can_sleep: we are in a sleepable context
1086 *
1087 * This function performs a route lookup on the given flow with the
1088 * possibility of using the cached route in the socket if it is valid.
1089 * It will take the socket dst lock when operating on the dst cache.
1090 * As a result, this function can only be used in process context.
1091 *
1092 * It returns a valid dst pointer on success, or a pointer encoded
1093 * error code.
1094 */
1095struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1096 const struct in6_addr *final_dst,
1097 bool can_sleep)
1098{
1099 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1100 int err;
1101
1102 dst = ip6_sk_dst_check(sk, dst, fl6);
1103
1104 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1105 if (err)
1106 return ERR_PTR(err);
1107 if (final_dst)
1108 fl6->daddr = *final_dst;
1109 if (can_sleep)
1110 fl6->flowi6_flags |= FLOWI_FLAG_CAN_SLEEP;
1111
1112 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1113}
1114EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1115
1116static inline int ip6_ufo_append_data(struct sock *sk,
1117 int getfrag(void *from, char *to, int offset, int len,
1118 int odd, struct sk_buff *skb),
1119 void *from, int length, int hh_len, int fragheaderlen,
1120 int transhdrlen, int mtu,unsigned int flags,
1121 struct rt6_info *rt)
1122
1123{
1124 struct sk_buff *skb;
1125 int err;
1126
1127 /* There is support for UDP large send offload by network
1128 * device, so create one single skb packet containing complete
1129 * udp datagram
1130 */
1131 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1132 skb = sock_alloc_send_skb(sk,
1133 hh_len + fragheaderlen + transhdrlen + 20,
1134 (flags & MSG_DONTWAIT), &err);
1135 if (skb == NULL)
1136 return err;
1137
1138 /* reserve space for Hardware header */
1139 skb_reserve(skb, hh_len);
1140
1141 /* create space for UDP/IP header */
1142 skb_put(skb,fragheaderlen + transhdrlen);
1143
1144 /* initialize network header pointer */
1145 skb_reset_network_header(skb);
1146
1147 /* initialize protocol header pointer */
1148 skb->transport_header = skb->network_header + fragheaderlen;
1149
1150 skb->ip_summed = CHECKSUM_PARTIAL;
1151 skb->csum = 0;
1152 }
1153
1154 err = skb_append_datato_frags(sk,skb, getfrag, from,
1155 (length - transhdrlen));
1156 if (!err) {
1157 struct frag_hdr fhdr;
1158
1159 /* Specify the length of each IPv6 datagram fragment.
1160 * It has to be a multiple of 8.
1161 */
1162 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1163 sizeof(struct frag_hdr)) & ~7;
1164 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1165 ipv6_select_ident(&fhdr, rt);
1166 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1167 __skb_queue_tail(&sk->sk_write_queue, skb);
1168
1169 return 0;
1170 }
1171 /* There is not enough support do UPD LSO,
1172 * so follow normal path
1173 */
1174 kfree_skb(skb);
1175
1176 return err;
1177}
1178
1179static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1180 gfp_t gfp)
1181{
1182 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1183}
1184
1185static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1186 gfp_t gfp)
1187{
1188 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1189}
1190
1191static void ip6_append_data_mtu(int *mtu,
1192 int *maxfraglen,
1193 unsigned int fragheaderlen,
1194 struct sk_buff *skb,
1195 struct rt6_info *rt)
1196{
1197 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1198 if (skb == NULL) {
1199 /* first fragment, reserve header_len */
1200 *mtu = *mtu - rt->dst.header_len;
1201
1202 } else {
1203 /*
1204 * this fragment is not first, the headers
1205 * space is regarded as data space.
1206 */
1207 *mtu = dst_mtu(rt->dst.path);
1208 }
1209 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1210 + fragheaderlen - sizeof(struct frag_hdr);
1211 }
1212}
1213
1214int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1215 int offset, int len, int odd, struct sk_buff *skb),
1216 void *from, int length, int transhdrlen,
1217 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1218 struct rt6_info *rt, unsigned int flags, int dontfrag)
1219{
1220 struct inet_sock *inet = inet_sk(sk);
1221 struct ipv6_pinfo *np = inet6_sk(sk);
1222 struct inet_cork *cork;
1223 struct sk_buff *skb, *skb_prev = NULL;
1224 unsigned int maxfraglen, fragheaderlen;
1225 int exthdrlen;
1226 int dst_exthdrlen;
1227 int hh_len;
1228 int mtu;
1229 int copy;
1230 int err;
1231 int offset = 0;
1232 __u8 tx_flags = 0;
1233
1234 if (flags&MSG_PROBE)
1235 return 0;
1236 cork = &inet->cork.base;
1237 if (skb_queue_empty(&sk->sk_write_queue)) {
1238 /*
1239 * setup for corking
1240 */
1241 if (opt) {
1242 if (WARN_ON(np->cork.opt))
1243 return -EINVAL;
1244
1245 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1246 if (unlikely(np->cork.opt == NULL))
1247 return -ENOBUFS;
1248
1249 np->cork.opt->tot_len = opt->tot_len;
1250 np->cork.opt->opt_flen = opt->opt_flen;
1251 np->cork.opt->opt_nflen = opt->opt_nflen;
1252
1253 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1254 sk->sk_allocation);
1255 if (opt->dst0opt && !np->cork.opt->dst0opt)
1256 return -ENOBUFS;
1257
1258 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1259 sk->sk_allocation);
1260 if (opt->dst1opt && !np->cork.opt->dst1opt)
1261 return -ENOBUFS;
1262
1263 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1264 sk->sk_allocation);
1265 if (opt->hopopt && !np->cork.opt->hopopt)
1266 return -ENOBUFS;
1267
1268 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1269 sk->sk_allocation);
1270 if (opt->srcrt && !np->cork.opt->srcrt)
1271 return -ENOBUFS;
1272
1273 /* need source address above miyazawa*/
1274 }
1275 dst_hold(&rt->dst);
1276 cork->dst = &rt->dst;
1277 inet->cork.fl.u.ip6 = *fl6;
1278 np->cork.hop_limit = hlimit;
1279 np->cork.tclass = tclass;
1280 if (rt->dst.flags & DST_XFRM_TUNNEL)
1281 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1282 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1283 else
1284 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1285 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1286 if (np->frag_size < mtu) {
1287 if (np->frag_size)
1288 mtu = np->frag_size;
1289 }
1290 cork->fragsize = mtu;
1291 if (dst_allfrag(rt->dst.path))
1292 cork->flags |= IPCORK_ALLFRAG;
1293 cork->length = 0;
1294 sk->sk_sndmsg_page = NULL;
1295 sk->sk_sndmsg_off = 0;
1296 exthdrlen = (opt ? opt->opt_flen : 0) - rt->rt6i_nfheader_len;
1297 length += exthdrlen;
1298 transhdrlen += exthdrlen;
1299 dst_exthdrlen = rt->dst.header_len;
1300 } else {
1301 rt = (struct rt6_info *)cork->dst;
1302 fl6 = &inet->cork.fl.u.ip6;
1303 opt = np->cork.opt;
1304 transhdrlen = 0;
1305 exthdrlen = 0;
1306 dst_exthdrlen = 0;
1307 mtu = cork->fragsize;
1308 }
1309
1310 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1311
1312 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1313 (opt ? opt->opt_nflen : 0);
1314 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1315
1316 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1317 if (cork->length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1318 ipv6_local_error(sk, EMSGSIZE, fl6, mtu-exthdrlen);
1319 return -EMSGSIZE;
1320 }
1321 }
1322
1323 /* For UDP, check if TX timestamp is enabled */
1324 if (sk->sk_type == SOCK_DGRAM) {
1325 err = sock_tx_timestamp(sk, &tx_flags);
1326 if (err)
1327 goto error;
1328 }
1329
1330 /*
1331 * Let's try using as much space as possible.
1332 * Use MTU if total length of the message fits into the MTU.
1333 * Otherwise, we need to reserve fragment header and
1334 * fragment alignment (= 8-15 octects, in total).
1335 *
1336 * Note that we may need to "move" the data from the tail of
1337 * of the buffer to the new fragment when we split
1338 * the message.
1339 *
1340 * FIXME: It may be fragmented into multiple chunks
1341 * at once if non-fragmentable extension headers
1342 * are too large.
1343 * --yoshfuji
1344 */
1345
1346 cork->length += length;
1347 if (length > mtu) {
1348 int proto = sk->sk_protocol;
1349 if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
1350 ipv6_local_rxpmtu(sk, fl6, mtu-exthdrlen);
1351 return -EMSGSIZE;
1352 }
1353
1354 if (proto == IPPROTO_UDP &&
1355 (rt->dst.dev->features & NETIF_F_UFO)) {
1356
1357 err = ip6_ufo_append_data(sk, getfrag, from, length,
1358 hh_len, fragheaderlen,
1359 transhdrlen, mtu, flags, rt);
1360 if (err)
1361 goto error;
1362 return 0;
1363 }
1364 }
1365
1366 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1367 goto alloc_new_skb;
1368
1369 while (length > 0) {
1370 /* Check if the remaining data fits into current packet. */
1371 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1372 if (copy < length)
1373 copy = maxfraglen - skb->len;
1374
1375 if (copy <= 0) {
1376 char *data;
1377 unsigned int datalen;
1378 unsigned int fraglen;
1379 unsigned int fraggap;
1380 unsigned int alloclen;
1381alloc_new_skb:
1382 /* There's no room in the current skb */
1383 if (skb)
1384 fraggap = skb->len - maxfraglen;
1385 else
1386 fraggap = 0;
1387 /* update mtu and maxfraglen if necessary */
1388 if (skb == NULL || skb_prev == NULL)
1389 ip6_append_data_mtu(&mtu, &maxfraglen,
1390 fragheaderlen, skb, rt);
1391
1392 skb_prev = skb;
1393
1394 /*
1395 * If remaining data exceeds the mtu,
1396 * we know we need more fragment(s).
1397 */
1398 datalen = length + fraggap;
1399
1400 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1401 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1402 if ((flags & MSG_MORE) &&
1403 !(rt->dst.dev->features&NETIF_F_SG))
1404 alloclen = mtu;
1405 else
1406 alloclen = datalen + fragheaderlen;
1407
1408 alloclen += dst_exthdrlen;
1409
1410 if (datalen != length + fraggap) {
1411 /*
1412 * this is not the last fragment, the trailer
1413 * space is regarded as data space.
1414 */
1415 datalen += rt->dst.trailer_len;
1416 }
1417
1418 alloclen += rt->dst.trailer_len;
1419 fraglen = datalen + fragheaderlen;
1420
1421 /*
1422 * We just reserve space for fragment header.
1423 * Note: this may be overallocation if the message
1424 * (without MSG_MORE) fits into the MTU.
1425 */
1426 alloclen += sizeof(struct frag_hdr);
1427
1428 if (transhdrlen) {
1429 skb = sock_alloc_send_skb(sk,
1430 alloclen + hh_len,
1431 (flags & MSG_DONTWAIT), &err);
1432 } else {
1433 skb = NULL;
1434 if (atomic_read(&sk->sk_wmem_alloc) <=
1435 2 * sk->sk_sndbuf)
1436 skb = sock_wmalloc(sk,
1437 alloclen + hh_len, 1,
1438 sk->sk_allocation);
1439 if (unlikely(skb == NULL))
1440 err = -ENOBUFS;
1441 else {
1442 /* Only the initial fragment
1443 * is time stamped.
1444 */
1445 tx_flags = 0;
1446 }
1447 }
1448 if (skb == NULL)
1449 goto error;
1450 /*
1451 * Fill in the control structures
1452 */
1453 skb->ip_summed = CHECKSUM_NONE;
1454 skb->csum = 0;
1455 /* reserve for fragmentation and ipsec header */
1456 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1457 dst_exthdrlen);
1458
1459 if (sk->sk_type == SOCK_DGRAM)
1460 skb_shinfo(skb)->tx_flags = tx_flags;
1461
1462 /*
1463 * Find where to start putting bytes
1464 */
1465 data = skb_put(skb, fraglen);
1466 skb_set_network_header(skb, exthdrlen);
1467 data += fragheaderlen;
1468 skb->transport_header = (skb->network_header +
1469 fragheaderlen);
1470 if (fraggap) {
1471 skb->csum = skb_copy_and_csum_bits(
1472 skb_prev, maxfraglen,
1473 data + transhdrlen, fraggap, 0);
1474 skb_prev->csum = csum_sub(skb_prev->csum,
1475 skb->csum);
1476 data += fraggap;
1477 pskb_trim_unique(skb_prev, maxfraglen);
1478 }
1479 copy = datalen - transhdrlen - fraggap;
1480
1481 if (copy < 0) {
1482 err = -EINVAL;
1483 kfree_skb(skb);
1484 goto error;
1485 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1486 err = -EFAULT;
1487 kfree_skb(skb);
1488 goto error;
1489 }
1490
1491 offset += copy;
1492 length -= datalen - fraggap;
1493 transhdrlen = 0;
1494 exthdrlen = 0;
1495 dst_exthdrlen = 0;
1496
1497 /*
1498 * Put the packet on the pending queue
1499 */
1500 __skb_queue_tail(&sk->sk_write_queue, skb);
1501 continue;
1502 }
1503
1504 if (copy > length)
1505 copy = length;
1506
1507 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1508 unsigned int off;
1509
1510 off = skb->len;
1511 if (getfrag(from, skb_put(skb, copy),
1512 offset, copy, off, skb) < 0) {
1513 __skb_trim(skb, off);
1514 err = -EFAULT;
1515 goto error;
1516 }
1517 } else {
1518 int i = skb_shinfo(skb)->nr_frags;
1519 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1520 struct page *page = sk->sk_sndmsg_page;
1521 int off = sk->sk_sndmsg_off;
1522 unsigned int left;
1523
1524 if (page && (left = PAGE_SIZE - off) > 0) {
1525 if (copy >= left)
1526 copy = left;
1527 if (page != skb_frag_page(frag)) {
1528 if (i == MAX_SKB_FRAGS) {
1529 err = -EMSGSIZE;
1530 goto error;
1531 }
1532 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1533 skb_frag_ref(skb, i);
1534 frag = &skb_shinfo(skb)->frags[i];
1535 }
1536 } else if(i < MAX_SKB_FRAGS) {
1537 if (copy > PAGE_SIZE)
1538 copy = PAGE_SIZE;
1539 page = alloc_pages(sk->sk_allocation, 0);
1540 if (page == NULL) {
1541 err = -ENOMEM;
1542 goto error;
1543 }
1544 sk->sk_sndmsg_page = page;
1545 sk->sk_sndmsg_off = 0;
1546
1547 skb_fill_page_desc(skb, i, page, 0, 0);
1548 frag = &skb_shinfo(skb)->frags[i];
1549 } else {
1550 err = -EMSGSIZE;
1551 goto error;
1552 }
1553 if (getfrag(from,
1554 skb_frag_address(frag) + skb_frag_size(frag),
1555 offset, copy, skb->len, skb) < 0) {
1556 err = -EFAULT;
1557 goto error;
1558 }
1559 sk->sk_sndmsg_off += copy;
1560 skb_frag_size_add(frag, copy);
1561 skb->len += copy;
1562 skb->data_len += copy;
1563 skb->truesize += copy;
1564 atomic_add(copy, &sk->sk_wmem_alloc);
1565 }
1566 offset += copy;
1567 length -= copy;
1568 }
1569 return 0;
1570error:
1571 cork->length -= length;
1572 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1573 return err;
1574}
1575EXPORT_SYMBOL_GPL(ip6_append_data);
1576
1577static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1578{
1579 if (np->cork.opt) {
1580 kfree(np->cork.opt->dst0opt);
1581 kfree(np->cork.opt->dst1opt);
1582 kfree(np->cork.opt->hopopt);
1583 kfree(np->cork.opt->srcrt);
1584 kfree(np->cork.opt);
1585 np->cork.opt = NULL;
1586 }
1587
1588 if (inet->cork.base.dst) {
1589 dst_release(inet->cork.base.dst);
1590 inet->cork.base.dst = NULL;
1591 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1592 }
1593 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1594}
1595
1596int ip6_push_pending_frames(struct sock *sk)
1597{
1598 struct sk_buff *skb, *tmp_skb;
1599 struct sk_buff **tail_skb;
1600 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1601 struct inet_sock *inet = inet_sk(sk);
1602 struct ipv6_pinfo *np = inet6_sk(sk);
1603 struct net *net = sock_net(sk);
1604 struct ipv6hdr *hdr;
1605 struct ipv6_txoptions *opt = np->cork.opt;
1606 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1607 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1608 unsigned char proto = fl6->flowi6_proto;
1609 int err = 0;
1610
1611 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1612 goto out;
1613 tail_skb = &(skb_shinfo(skb)->frag_list);
1614
1615 /* move skb->data to ip header from ext header */
1616 if (skb->data < skb_network_header(skb))
1617 __skb_pull(skb, skb_network_offset(skb));
1618 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1619 __skb_pull(tmp_skb, skb_network_header_len(skb));
1620 *tail_skb = tmp_skb;
1621 tail_skb = &(tmp_skb->next);
1622 skb->len += tmp_skb->len;
1623 skb->data_len += tmp_skb->len;
1624 skb->truesize += tmp_skb->truesize;
1625 tmp_skb->destructor = NULL;
1626 tmp_skb->sk = NULL;
1627 }
1628
1629 /* Allow local fragmentation. */
1630 if (np->pmtudisc < IPV6_PMTUDISC_DO)
1631 skb->local_df = 1;
1632
1633 *final_dst = fl6->daddr;
1634 __skb_pull(skb, skb_network_header_len(skb));
1635 if (opt && opt->opt_flen)
1636 ipv6_push_frag_opts(skb, opt, &proto);
1637 if (opt && opt->opt_nflen)
1638 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1639
1640 skb_push(skb, sizeof(struct ipv6hdr));
1641 skb_reset_network_header(skb);
1642 hdr = ipv6_hdr(skb);
1643
1644 *(__be32*)hdr = fl6->flowlabel |
1645 htonl(0x60000000 | ((int)np->cork.tclass << 20));
1646
1647 hdr->hop_limit = np->cork.hop_limit;
1648 hdr->nexthdr = proto;
1649 hdr->saddr = fl6->saddr;
1650 hdr->daddr = *final_dst;
1651
1652 skb->priority = sk->sk_priority;
1653 skb->mark = sk->sk_mark;
1654
1655 skb_dst_set(skb, dst_clone(&rt->dst));
1656 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1657 if (proto == IPPROTO_ICMPV6) {
1658 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1659
1660 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
1661 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
1662 }
1663
1664 err = ip6_local_out(skb);
1665 if (err) {
1666 if (err > 0)
1667 err = net_xmit_errno(err);
1668 if (err)
1669 goto error;
1670 }
1671
1672out:
1673 ip6_cork_release(inet, np);
1674 return err;
1675error:
1676 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1677 goto out;
1678}
1679EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1680
1681void ip6_flush_pending_frames(struct sock *sk)
1682{
1683 struct sk_buff *skb;
1684
1685 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1686 if (skb_dst(skb))
1687 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1688 IPSTATS_MIB_OUTDISCARDS);
1689 kfree_skb(skb);
1690 }
1691
1692 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1693}
1694EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * IPv6 output functions
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on linux/net/ipv4/ip_output.c
10 *
11 * Changes:
12 * A.N.Kuznetsov : airthmetics in fragmentation.
13 * extension headers are implemented.
14 * route changes now work.
15 * ip6_forward does not confuse sniffers.
16 * etc.
17 *
18 * H. von Brand : Added missing #include <linux/string.h>
19 * Imran Patel : frag id should be in NBO
20 * Kazunori MIYAZAWA @USAGI
21 * : add ip6_append_data and related functions
22 * for datagram xmit
23 */
24
25#include <linux/errno.h>
26#include <linux/kernel.h>
27#include <linux/string.h>
28#include <linux/socket.h>
29#include <linux/net.h>
30#include <linux/netdevice.h>
31#include <linux/if_arp.h>
32#include <linux/in6.h>
33#include <linux/tcp.h>
34#include <linux/route.h>
35#include <linux/module.h>
36#include <linux/slab.h>
37
38#include <linux/bpf-cgroup.h>
39#include <linux/netfilter.h>
40#include <linux/netfilter_ipv6.h>
41
42#include <net/sock.h>
43#include <net/snmp.h>
44
45#include <net/ipv6.h>
46#include <net/ndisc.h>
47#include <net/protocol.h>
48#include <net/ip6_route.h>
49#include <net/addrconf.h>
50#include <net/rawv6.h>
51#include <net/icmp.h>
52#include <net/xfrm.h>
53#include <net/checksum.h>
54#include <linux/mroute6.h>
55#include <net/l3mdev.h>
56#include <net/lwtunnel.h>
57#include <net/ip_tunnels.h>
58
59static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
60{
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 const struct in6_addr *nexthop;
64 struct neighbour *neigh;
65 int ret;
66
67 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
68 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
69
70 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
71 ((mroute6_is_socket(net, skb) &&
72 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
73 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
74 &ipv6_hdr(skb)->saddr))) {
75 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
76
77 /* Do not check for IFF_ALLMULTI; multicast routing
78 is not supported in any case.
79 */
80 if (newskb)
81 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
82 net, sk, newskb, NULL, newskb->dev,
83 dev_loopback_xmit);
84
85 if (ipv6_hdr(skb)->hop_limit == 0) {
86 IP6_INC_STATS(net, idev,
87 IPSTATS_MIB_OUTDISCARDS);
88 kfree_skb(skb);
89 return 0;
90 }
91 }
92
93 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
94
95 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
96 IPV6_ADDR_SCOPE_NODELOCAL &&
97 !(dev->flags & IFF_LOOPBACK)) {
98 kfree_skb(skb);
99 return 0;
100 }
101 }
102
103 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
104 int res = lwtunnel_xmit(skb);
105
106 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
107 return res;
108 }
109
110 rcu_read_lock_bh();
111 nexthop = rt6_nexthop((struct rt6_info *)dst, &ipv6_hdr(skb)->daddr);
112 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
113 if (unlikely(!neigh))
114 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
115 if (!IS_ERR(neigh)) {
116 sock_confirm_neigh(skb, neigh);
117 ret = neigh_output(neigh, skb, false);
118 rcu_read_unlock_bh();
119 return ret;
120 }
121 rcu_read_unlock_bh();
122
123 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
124 kfree_skb(skb);
125 return -EINVAL;
126}
127
128static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
129{
130#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
131 /* Policy lookup after SNAT yielded a new policy */
132 if (skb_dst(skb)->xfrm) {
133 IPCB(skb)->flags |= IPSKB_REROUTED;
134 return dst_output(net, sk, skb);
135 }
136#endif
137
138 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
139 dst_allfrag(skb_dst(skb)) ||
140 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
141 return ip6_fragment(net, sk, skb, ip6_finish_output2);
142 else
143 return ip6_finish_output2(net, sk, skb);
144}
145
146static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
147{
148 int ret;
149
150 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
151 switch (ret) {
152 case NET_XMIT_SUCCESS:
153 return __ip6_finish_output(net, sk, skb);
154 case NET_XMIT_CN:
155 return __ip6_finish_output(net, sk, skb) ? : ret;
156 default:
157 kfree_skb(skb);
158 return ret;
159 }
160}
161
162int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
163{
164 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
165 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
166
167 skb->protocol = htons(ETH_P_IPV6);
168 skb->dev = dev;
169
170 if (unlikely(idev->cnf.disable_ipv6)) {
171 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
172 kfree_skb(skb);
173 return 0;
174 }
175
176 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
177 net, sk, skb, indev, dev,
178 ip6_finish_output,
179 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
180}
181
182bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
183{
184 if (!np->autoflowlabel_set)
185 return ip6_default_np_autolabel(net);
186 else
187 return np->autoflowlabel;
188}
189
190/*
191 * xmit an sk_buff (used by TCP, SCTP and DCCP)
192 * Note : socket lock is not held for SYNACK packets, but might be modified
193 * by calls to skb_set_owner_w() and ipv6_local_error(),
194 * which are using proper atomic operations or spinlocks.
195 */
196int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
197 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
198{
199 struct net *net = sock_net(sk);
200 const struct ipv6_pinfo *np = inet6_sk(sk);
201 struct in6_addr *first_hop = &fl6->daddr;
202 struct dst_entry *dst = skb_dst(skb);
203 unsigned int head_room;
204 struct ipv6hdr *hdr;
205 u8 proto = fl6->flowi6_proto;
206 int seg_len = skb->len;
207 int hlimit = -1;
208 u32 mtu;
209
210 head_room = sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
211 if (opt)
212 head_room += opt->opt_nflen + opt->opt_flen;
213
214 if (unlikely(skb_headroom(skb) < head_room)) {
215 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
216 if (!skb2) {
217 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
218 IPSTATS_MIB_OUTDISCARDS);
219 kfree_skb(skb);
220 return -ENOBUFS;
221 }
222 if (skb->sk)
223 skb_set_owner_w(skb2, skb->sk);
224 consume_skb(skb);
225 skb = skb2;
226 }
227
228 if (opt) {
229 seg_len += opt->opt_nflen + opt->opt_flen;
230
231 if (opt->opt_flen)
232 ipv6_push_frag_opts(skb, opt, &proto);
233
234 if (opt->opt_nflen)
235 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
236 &fl6->saddr);
237 }
238
239 skb_push(skb, sizeof(struct ipv6hdr));
240 skb_reset_network_header(skb);
241 hdr = ipv6_hdr(skb);
242
243 /*
244 * Fill in the IPv6 header
245 */
246 if (np)
247 hlimit = np->hop_limit;
248 if (hlimit < 0)
249 hlimit = ip6_dst_hoplimit(dst);
250
251 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
252 ip6_autoflowlabel(net, np), fl6));
253
254 hdr->payload_len = htons(seg_len);
255 hdr->nexthdr = proto;
256 hdr->hop_limit = hlimit;
257
258 hdr->saddr = fl6->saddr;
259 hdr->daddr = *first_hop;
260
261 skb->protocol = htons(ETH_P_IPV6);
262 skb->priority = priority;
263 skb->mark = mark;
264
265 mtu = dst_mtu(dst);
266 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
267 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
268 IPSTATS_MIB_OUT, skb->len);
269
270 /* if egress device is enslaved to an L3 master device pass the
271 * skb to its handler for processing
272 */
273 skb = l3mdev_ip6_out((struct sock *)sk, skb);
274 if (unlikely(!skb))
275 return 0;
276
277 /* hooks should never assume socket lock is held.
278 * we promote our socket to non const
279 */
280 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
281 net, (struct sock *)sk, skb, NULL, dst->dev,
282 dst_output);
283 }
284
285 skb->dev = dst->dev;
286 /* ipv6_local_error() does not require socket lock,
287 * we promote our socket to non const
288 */
289 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
290
291 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
292 kfree_skb(skb);
293 return -EMSGSIZE;
294}
295EXPORT_SYMBOL(ip6_xmit);
296
297static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
298{
299 struct ip6_ra_chain *ra;
300 struct sock *last = NULL;
301
302 read_lock(&ip6_ra_lock);
303 for (ra = ip6_ra_chain; ra; ra = ra->next) {
304 struct sock *sk = ra->sk;
305 if (sk && ra->sel == sel &&
306 (!sk->sk_bound_dev_if ||
307 sk->sk_bound_dev_if == skb->dev->ifindex)) {
308 struct ipv6_pinfo *np = inet6_sk(sk);
309
310 if (np && np->rtalert_isolate &&
311 !net_eq(sock_net(sk), dev_net(skb->dev))) {
312 continue;
313 }
314 if (last) {
315 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
316 if (skb2)
317 rawv6_rcv(last, skb2);
318 }
319 last = sk;
320 }
321 }
322
323 if (last) {
324 rawv6_rcv(last, skb);
325 read_unlock(&ip6_ra_lock);
326 return 1;
327 }
328 read_unlock(&ip6_ra_lock);
329 return 0;
330}
331
332static int ip6_forward_proxy_check(struct sk_buff *skb)
333{
334 struct ipv6hdr *hdr = ipv6_hdr(skb);
335 u8 nexthdr = hdr->nexthdr;
336 __be16 frag_off;
337 int offset;
338
339 if (ipv6_ext_hdr(nexthdr)) {
340 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
341 if (offset < 0)
342 return 0;
343 } else
344 offset = sizeof(struct ipv6hdr);
345
346 if (nexthdr == IPPROTO_ICMPV6) {
347 struct icmp6hdr *icmp6;
348
349 if (!pskb_may_pull(skb, (skb_network_header(skb) +
350 offset + 1 - skb->data)))
351 return 0;
352
353 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
354
355 switch (icmp6->icmp6_type) {
356 case NDISC_ROUTER_SOLICITATION:
357 case NDISC_ROUTER_ADVERTISEMENT:
358 case NDISC_NEIGHBOUR_SOLICITATION:
359 case NDISC_NEIGHBOUR_ADVERTISEMENT:
360 case NDISC_REDIRECT:
361 /* For reaction involving unicast neighbor discovery
362 * message destined to the proxied address, pass it to
363 * input function.
364 */
365 return 1;
366 default:
367 break;
368 }
369 }
370
371 /*
372 * The proxying router can't forward traffic sent to a link-local
373 * address, so signal the sender and discard the packet. This
374 * behavior is clarified by the MIPv6 specification.
375 */
376 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
377 dst_link_failure(skb);
378 return -1;
379 }
380
381 return 0;
382}
383
384static inline int ip6_forward_finish(struct net *net, struct sock *sk,
385 struct sk_buff *skb)
386{
387 struct dst_entry *dst = skb_dst(skb);
388
389 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
390 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
391
392#ifdef CONFIG_NET_SWITCHDEV
393 if (skb->offload_l3_fwd_mark) {
394 consume_skb(skb);
395 return 0;
396 }
397#endif
398
399 skb->tstamp = 0;
400 return dst_output(net, sk, skb);
401}
402
403static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
404{
405 if (skb->len <= mtu)
406 return false;
407
408 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
409 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
410 return true;
411
412 if (skb->ignore_df)
413 return false;
414
415 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
416 return false;
417
418 return true;
419}
420
421int ip6_forward(struct sk_buff *skb)
422{
423 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
424 struct dst_entry *dst = skb_dst(skb);
425 struct ipv6hdr *hdr = ipv6_hdr(skb);
426 struct inet6_skb_parm *opt = IP6CB(skb);
427 struct net *net = dev_net(dst->dev);
428 u32 mtu;
429
430 if (net->ipv6.devconf_all->forwarding == 0)
431 goto error;
432
433 if (skb->pkt_type != PACKET_HOST)
434 goto drop;
435
436 if (unlikely(skb->sk))
437 goto drop;
438
439 if (skb_warn_if_lro(skb))
440 goto drop;
441
442 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
443 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
444 goto drop;
445 }
446
447 skb_forward_csum(skb);
448
449 /*
450 * We DO NOT make any processing on
451 * RA packets, pushing them to user level AS IS
452 * without ane WARRANTY that application will be able
453 * to interpret them. The reason is that we
454 * cannot make anything clever here.
455 *
456 * We are not end-node, so that if packet contains
457 * AH/ESP, we cannot make anything.
458 * Defragmentation also would be mistake, RA packets
459 * cannot be fragmented, because there is no warranty
460 * that different fragments will go along one path. --ANK
461 */
462 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
463 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
464 return 0;
465 }
466
467 /*
468 * check and decrement ttl
469 */
470 if (hdr->hop_limit <= 1) {
471 /* Force OUTPUT device used as source address */
472 skb->dev = dst->dev;
473 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
474 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
475
476 kfree_skb(skb);
477 return -ETIMEDOUT;
478 }
479
480 /* XXX: idev->cnf.proxy_ndp? */
481 if (net->ipv6.devconf_all->proxy_ndp &&
482 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
483 int proxied = ip6_forward_proxy_check(skb);
484 if (proxied > 0)
485 return ip6_input(skb);
486 else if (proxied < 0) {
487 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
488 goto drop;
489 }
490 }
491
492 if (!xfrm6_route_forward(skb)) {
493 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
494 goto drop;
495 }
496 dst = skb_dst(skb);
497
498 /* IPv6 specs say nothing about it, but it is clear that we cannot
499 send redirects to source routed frames.
500 We don't send redirects to frames decapsulated from IPsec.
501 */
502 if (IP6CB(skb)->iif == dst->dev->ifindex &&
503 opt->srcrt == 0 && !skb_sec_path(skb)) {
504 struct in6_addr *target = NULL;
505 struct inet_peer *peer;
506 struct rt6_info *rt;
507
508 /*
509 * incoming and outgoing devices are the same
510 * send a redirect.
511 */
512
513 rt = (struct rt6_info *) dst;
514 if (rt->rt6i_flags & RTF_GATEWAY)
515 target = &rt->rt6i_gateway;
516 else
517 target = &hdr->daddr;
518
519 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
520
521 /* Limit redirects both by destination (here)
522 and by source (inside ndisc_send_redirect)
523 */
524 if (inet_peer_xrlim_allow(peer, 1*HZ))
525 ndisc_send_redirect(skb, target);
526 if (peer)
527 inet_putpeer(peer);
528 } else {
529 int addrtype = ipv6_addr_type(&hdr->saddr);
530
531 /* This check is security critical. */
532 if (addrtype == IPV6_ADDR_ANY ||
533 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
534 goto error;
535 if (addrtype & IPV6_ADDR_LINKLOCAL) {
536 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
537 ICMPV6_NOT_NEIGHBOUR, 0);
538 goto error;
539 }
540 }
541
542 mtu = ip6_dst_mtu_forward(dst);
543 if (mtu < IPV6_MIN_MTU)
544 mtu = IPV6_MIN_MTU;
545
546 if (ip6_pkt_too_big(skb, mtu)) {
547 /* Again, force OUTPUT device used as source address */
548 skb->dev = dst->dev;
549 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
550 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
551 __IP6_INC_STATS(net, ip6_dst_idev(dst),
552 IPSTATS_MIB_FRAGFAILS);
553 kfree_skb(skb);
554 return -EMSGSIZE;
555 }
556
557 if (skb_cow(skb, dst->dev->hard_header_len)) {
558 __IP6_INC_STATS(net, ip6_dst_idev(dst),
559 IPSTATS_MIB_OUTDISCARDS);
560 goto drop;
561 }
562
563 hdr = ipv6_hdr(skb);
564
565 /* Mangling hops number delayed to point after skb COW */
566
567 hdr->hop_limit--;
568
569 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
570 net, NULL, skb, skb->dev, dst->dev,
571 ip6_forward_finish);
572
573error:
574 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
575drop:
576 kfree_skb(skb);
577 return -EINVAL;
578}
579
580static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
581{
582 to->pkt_type = from->pkt_type;
583 to->priority = from->priority;
584 to->protocol = from->protocol;
585 skb_dst_drop(to);
586 skb_dst_set(to, dst_clone(skb_dst(from)));
587 to->dev = from->dev;
588 to->mark = from->mark;
589
590 skb_copy_hash(to, from);
591
592#ifdef CONFIG_NET_SCHED
593 to->tc_index = from->tc_index;
594#endif
595 nf_copy(to, from);
596 skb_ext_copy(to, from);
597 skb_copy_secmark(to, from);
598}
599
600int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
601 u8 nexthdr, __be32 frag_id,
602 struct ip6_fraglist_iter *iter)
603{
604 unsigned int first_len;
605 struct frag_hdr *fh;
606
607 /* BUILD HEADER */
608 *prevhdr = NEXTHDR_FRAGMENT;
609 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
610 if (!iter->tmp_hdr)
611 return -ENOMEM;
612
613 iter->frag = skb_shinfo(skb)->frag_list;
614 skb_frag_list_init(skb);
615
616 iter->offset = 0;
617 iter->hlen = hlen;
618 iter->frag_id = frag_id;
619 iter->nexthdr = nexthdr;
620
621 __skb_pull(skb, hlen);
622 fh = __skb_push(skb, sizeof(struct frag_hdr));
623 __skb_push(skb, hlen);
624 skb_reset_network_header(skb);
625 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
626
627 fh->nexthdr = nexthdr;
628 fh->reserved = 0;
629 fh->frag_off = htons(IP6_MF);
630 fh->identification = frag_id;
631
632 first_len = skb_pagelen(skb);
633 skb->data_len = first_len - skb_headlen(skb);
634 skb->len = first_len;
635 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
636
637 return 0;
638}
639EXPORT_SYMBOL(ip6_fraglist_init);
640
641void ip6_fraglist_prepare(struct sk_buff *skb,
642 struct ip6_fraglist_iter *iter)
643{
644 struct sk_buff *frag = iter->frag;
645 unsigned int hlen = iter->hlen;
646 struct frag_hdr *fh;
647
648 frag->ip_summed = CHECKSUM_NONE;
649 skb_reset_transport_header(frag);
650 fh = __skb_push(frag, sizeof(struct frag_hdr));
651 __skb_push(frag, hlen);
652 skb_reset_network_header(frag);
653 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
654 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
655 fh->nexthdr = iter->nexthdr;
656 fh->reserved = 0;
657 fh->frag_off = htons(iter->offset);
658 if (frag->next)
659 fh->frag_off |= htons(IP6_MF);
660 fh->identification = iter->frag_id;
661 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
662 ip6_copy_metadata(frag, skb);
663}
664EXPORT_SYMBOL(ip6_fraglist_prepare);
665
666void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
667 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
668 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
669{
670 state->prevhdr = prevhdr;
671 state->nexthdr = nexthdr;
672 state->frag_id = frag_id;
673
674 state->hlen = hlen;
675 state->mtu = mtu;
676
677 state->left = skb->len - hlen; /* Space per frame */
678 state->ptr = hlen; /* Where to start from */
679
680 state->hroom = hdr_room;
681 state->troom = needed_tailroom;
682
683 state->offset = 0;
684}
685EXPORT_SYMBOL(ip6_frag_init);
686
687struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
688{
689 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
690 struct sk_buff *frag;
691 struct frag_hdr *fh;
692 unsigned int len;
693
694 len = state->left;
695 /* IF: it doesn't fit, use 'mtu' - the data space left */
696 if (len > state->mtu)
697 len = state->mtu;
698 /* IF: we are not sending up to and including the packet end
699 then align the next start on an eight byte boundary */
700 if (len < state->left)
701 len &= ~7;
702
703 /* Allocate buffer */
704 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
705 state->hroom + state->troom, GFP_ATOMIC);
706 if (!frag)
707 return ERR_PTR(-ENOMEM);
708
709 /*
710 * Set up data on packet
711 */
712
713 ip6_copy_metadata(frag, skb);
714 skb_reserve(frag, state->hroom);
715 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
716 skb_reset_network_header(frag);
717 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
718 frag->transport_header = (frag->network_header + state->hlen +
719 sizeof(struct frag_hdr));
720
721 /*
722 * Charge the memory for the fragment to any owner
723 * it might possess
724 */
725 if (skb->sk)
726 skb_set_owner_w(frag, skb->sk);
727
728 /*
729 * Copy the packet header into the new buffer.
730 */
731 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
732
733 fragnexthdr_offset = skb_network_header(frag);
734 fragnexthdr_offset += prevhdr - skb_network_header(skb);
735 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
736
737 /*
738 * Build fragment header.
739 */
740 fh->nexthdr = state->nexthdr;
741 fh->reserved = 0;
742 fh->identification = state->frag_id;
743
744 /*
745 * Copy a block of the IP datagram.
746 */
747 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
748 len));
749 state->left -= len;
750
751 fh->frag_off = htons(state->offset);
752 if (state->left > 0)
753 fh->frag_off |= htons(IP6_MF);
754 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
755
756 state->ptr += len;
757 state->offset += len;
758
759 return frag;
760}
761EXPORT_SYMBOL(ip6_frag_next);
762
763int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
764 int (*output)(struct net *, struct sock *, struct sk_buff *))
765{
766 struct sk_buff *frag;
767 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
768 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
769 inet6_sk(skb->sk) : NULL;
770 struct ip6_frag_state state;
771 unsigned int mtu, hlen, nexthdr_offset;
772 ktime_t tstamp = skb->tstamp;
773 int hroom, err = 0;
774 __be32 frag_id;
775 u8 *prevhdr, nexthdr = 0;
776
777 err = ip6_find_1stfragopt(skb, &prevhdr);
778 if (err < 0)
779 goto fail;
780 hlen = err;
781 nexthdr = *prevhdr;
782 nexthdr_offset = prevhdr - skb_network_header(skb);
783
784 mtu = ip6_skb_dst_mtu(skb);
785
786 /* We must not fragment if the socket is set to force MTU discovery
787 * or if the skb it not generated by a local socket.
788 */
789 if (unlikely(!skb->ignore_df && skb->len > mtu))
790 goto fail_toobig;
791
792 if (IP6CB(skb)->frag_max_size) {
793 if (IP6CB(skb)->frag_max_size > mtu)
794 goto fail_toobig;
795
796 /* don't send fragments larger than what we received */
797 mtu = IP6CB(skb)->frag_max_size;
798 if (mtu < IPV6_MIN_MTU)
799 mtu = IPV6_MIN_MTU;
800 }
801
802 if (np && np->frag_size < mtu) {
803 if (np->frag_size)
804 mtu = np->frag_size;
805 }
806 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
807 goto fail_toobig;
808 mtu -= hlen + sizeof(struct frag_hdr);
809
810 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
811 &ipv6_hdr(skb)->saddr);
812
813 if (skb->ip_summed == CHECKSUM_PARTIAL &&
814 (err = skb_checksum_help(skb)))
815 goto fail;
816
817 prevhdr = skb_network_header(skb) + nexthdr_offset;
818 hroom = LL_RESERVED_SPACE(rt->dst.dev);
819 if (skb_has_frag_list(skb)) {
820 unsigned int first_len = skb_pagelen(skb);
821 struct ip6_fraglist_iter iter;
822 struct sk_buff *frag2;
823
824 if (first_len - hlen > mtu ||
825 ((first_len - hlen) & 7) ||
826 skb_cloned(skb) ||
827 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
828 goto slow_path;
829
830 skb_walk_frags(skb, frag) {
831 /* Correct geometry. */
832 if (frag->len > mtu ||
833 ((frag->len & 7) && frag->next) ||
834 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
835 goto slow_path_clean;
836
837 /* Partially cloned skb? */
838 if (skb_shared(frag))
839 goto slow_path_clean;
840
841 BUG_ON(frag->sk);
842 if (skb->sk) {
843 frag->sk = skb->sk;
844 frag->destructor = sock_wfree;
845 }
846 skb->truesize -= frag->truesize;
847 }
848
849 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
850 &iter);
851 if (err < 0)
852 goto fail;
853
854 for (;;) {
855 /* Prepare header of the next frame,
856 * before previous one went down. */
857 if (iter.frag)
858 ip6_fraglist_prepare(skb, &iter);
859
860 skb->tstamp = tstamp;
861 err = output(net, sk, skb);
862 if (!err)
863 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
864 IPSTATS_MIB_FRAGCREATES);
865
866 if (err || !iter.frag)
867 break;
868
869 skb = ip6_fraglist_next(&iter);
870 }
871
872 kfree(iter.tmp_hdr);
873
874 if (err == 0) {
875 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
876 IPSTATS_MIB_FRAGOKS);
877 return 0;
878 }
879
880 kfree_skb_list(iter.frag);
881
882 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
883 IPSTATS_MIB_FRAGFAILS);
884 return err;
885
886slow_path_clean:
887 skb_walk_frags(skb, frag2) {
888 if (frag2 == frag)
889 break;
890 frag2->sk = NULL;
891 frag2->destructor = NULL;
892 skb->truesize += frag2->truesize;
893 }
894 }
895
896slow_path:
897 /*
898 * Fragment the datagram.
899 */
900
901 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
902 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
903 &state);
904
905 /*
906 * Keep copying data until we run out.
907 */
908
909 while (state.left > 0) {
910 frag = ip6_frag_next(skb, &state);
911 if (IS_ERR(frag)) {
912 err = PTR_ERR(frag);
913 goto fail;
914 }
915
916 /*
917 * Put this fragment into the sending queue.
918 */
919 frag->tstamp = tstamp;
920 err = output(net, sk, frag);
921 if (err)
922 goto fail;
923
924 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
925 IPSTATS_MIB_FRAGCREATES);
926 }
927 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
928 IPSTATS_MIB_FRAGOKS);
929 consume_skb(skb);
930 return err;
931
932fail_toobig:
933 if (skb->sk && dst_allfrag(skb_dst(skb)))
934 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
935
936 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
937 err = -EMSGSIZE;
938
939fail:
940 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
941 IPSTATS_MIB_FRAGFAILS);
942 kfree_skb(skb);
943 return err;
944}
945
946static inline int ip6_rt_check(const struct rt6key *rt_key,
947 const struct in6_addr *fl_addr,
948 const struct in6_addr *addr_cache)
949{
950 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
951 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
952}
953
954static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
955 struct dst_entry *dst,
956 const struct flowi6 *fl6)
957{
958 struct ipv6_pinfo *np = inet6_sk(sk);
959 struct rt6_info *rt;
960
961 if (!dst)
962 goto out;
963
964 if (dst->ops->family != AF_INET6) {
965 dst_release(dst);
966 return NULL;
967 }
968
969 rt = (struct rt6_info *)dst;
970 /* Yes, checking route validity in not connected
971 * case is not very simple. Take into account,
972 * that we do not support routing by source, TOS,
973 * and MSG_DONTROUTE --ANK (980726)
974 *
975 * 1. ip6_rt_check(): If route was host route,
976 * check that cached destination is current.
977 * If it is network route, we still may
978 * check its validity using saved pointer
979 * to the last used address: daddr_cache.
980 * We do not want to save whole address now,
981 * (because main consumer of this service
982 * is tcp, which has not this problem),
983 * so that the last trick works only on connected
984 * sockets.
985 * 2. oif also should be the same.
986 */
987 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
988#ifdef CONFIG_IPV6_SUBTREES
989 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
990#endif
991 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
992 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
993 dst_release(dst);
994 dst = NULL;
995 }
996
997out:
998 return dst;
999}
1000
1001static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1002 struct dst_entry **dst, struct flowi6 *fl6)
1003{
1004#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1005 struct neighbour *n;
1006 struct rt6_info *rt;
1007#endif
1008 int err;
1009 int flags = 0;
1010
1011 /* The correct way to handle this would be to do
1012 * ip6_route_get_saddr, and then ip6_route_output; however,
1013 * the route-specific preferred source forces the
1014 * ip6_route_output call _before_ ip6_route_get_saddr.
1015 *
1016 * In source specific routing (no src=any default route),
1017 * ip6_route_output will fail given src=any saddr, though, so
1018 * that's why we try it again later.
1019 */
1020 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
1021 struct fib6_info *from;
1022 struct rt6_info *rt;
1023 bool had_dst = *dst != NULL;
1024
1025 if (!had_dst)
1026 *dst = ip6_route_output(net, sk, fl6);
1027 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1028
1029 rcu_read_lock();
1030 from = rt ? rcu_dereference(rt->from) : NULL;
1031 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1032 sk ? inet6_sk(sk)->srcprefs : 0,
1033 &fl6->saddr);
1034 rcu_read_unlock();
1035
1036 if (err)
1037 goto out_err_release;
1038
1039 /* If we had an erroneous initial result, pretend it
1040 * never existed and let the SA-enabled version take
1041 * over.
1042 */
1043 if (!had_dst && (*dst)->error) {
1044 dst_release(*dst);
1045 *dst = NULL;
1046 }
1047
1048 if (fl6->flowi6_oif)
1049 flags |= RT6_LOOKUP_F_IFACE;
1050 }
1051
1052 if (!*dst)
1053 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1054
1055 err = (*dst)->error;
1056 if (err)
1057 goto out_err_release;
1058
1059#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1060 /*
1061 * Here if the dst entry we've looked up
1062 * has a neighbour entry that is in the INCOMPLETE
1063 * state and the src address from the flow is
1064 * marked as OPTIMISTIC, we release the found
1065 * dst entry and replace it instead with the
1066 * dst entry of the nexthop router
1067 */
1068 rt = (struct rt6_info *) *dst;
1069 rcu_read_lock_bh();
1070 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1071 rt6_nexthop(rt, &fl6->daddr));
1072 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1073 rcu_read_unlock_bh();
1074
1075 if (err) {
1076 struct inet6_ifaddr *ifp;
1077 struct flowi6 fl_gw6;
1078 int redirect;
1079
1080 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1081 (*dst)->dev, 1);
1082
1083 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1084 if (ifp)
1085 in6_ifa_put(ifp);
1086
1087 if (redirect) {
1088 /*
1089 * We need to get the dst entry for the
1090 * default router instead
1091 */
1092 dst_release(*dst);
1093 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1094 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1095 *dst = ip6_route_output(net, sk, &fl_gw6);
1096 err = (*dst)->error;
1097 if (err)
1098 goto out_err_release;
1099 }
1100 }
1101#endif
1102 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1103 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1104 err = -EAFNOSUPPORT;
1105 goto out_err_release;
1106 }
1107
1108 return 0;
1109
1110out_err_release:
1111 dst_release(*dst);
1112 *dst = NULL;
1113
1114 if (err == -ENETUNREACH)
1115 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1116 return err;
1117}
1118
1119/**
1120 * ip6_dst_lookup - perform route lookup on flow
1121 * @net: Network namespace to perform lookup in
1122 * @sk: socket which provides route info
1123 * @dst: pointer to dst_entry * for result
1124 * @fl6: flow to lookup
1125 *
1126 * This function performs a route lookup on the given flow.
1127 *
1128 * It returns zero on success, or a standard errno code on error.
1129 */
1130int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1131 struct flowi6 *fl6)
1132{
1133 *dst = NULL;
1134 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1135}
1136EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1137
1138/**
1139 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1140 * @net: Network namespace to perform lookup in
1141 * @sk: socket which provides route info
1142 * @fl6: flow to lookup
1143 * @final_dst: final destination address for ipsec lookup
1144 *
1145 * This function performs a route lookup on the given flow.
1146 *
1147 * It returns a valid dst pointer on success, or a pointer encoded
1148 * error code.
1149 */
1150struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1151 const struct in6_addr *final_dst)
1152{
1153 struct dst_entry *dst = NULL;
1154 int err;
1155
1156 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1157 if (err)
1158 return ERR_PTR(err);
1159 if (final_dst)
1160 fl6->daddr = *final_dst;
1161
1162 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1163}
1164EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1165
1166/**
1167 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1168 * @sk: socket which provides the dst cache and route info
1169 * @fl6: flow to lookup
1170 * @final_dst: final destination address for ipsec lookup
1171 * @connected: whether @sk is connected or not
1172 *
1173 * This function performs a route lookup on the given flow with the
1174 * possibility of using the cached route in the socket if it is valid.
1175 * It will take the socket dst lock when operating on the dst cache.
1176 * As a result, this function can only be used in process context.
1177 *
1178 * In addition, for a connected socket, cache the dst in the socket
1179 * if the current cache is not valid.
1180 *
1181 * It returns a valid dst pointer on success, or a pointer encoded
1182 * error code.
1183 */
1184struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1185 const struct in6_addr *final_dst,
1186 bool connected)
1187{
1188 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1189
1190 dst = ip6_sk_dst_check(sk, dst, fl6);
1191 if (dst)
1192 return dst;
1193
1194 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1195 if (connected && !IS_ERR(dst))
1196 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1197
1198 return dst;
1199}
1200EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1201
1202/**
1203 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1204 * @skb: Packet for which lookup is done
1205 * @dev: Tunnel device
1206 * @net: Network namespace of tunnel device
1207 * @sock: Socket which provides route info
1208 * @saddr: Memory to store the src ip address
1209 * @info: Tunnel information
1210 * @protocol: IP protocol
1211 * @use_cache: Flag to enable cache usage
1212 * This function performs a route lookup on a tunnel
1213 *
1214 * It returns a valid dst pointer and stores src address to be used in
1215 * tunnel in param saddr on success, else a pointer encoded error code.
1216 */
1217
1218struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1219 struct net_device *dev,
1220 struct net *net,
1221 struct socket *sock,
1222 struct in6_addr *saddr,
1223 const struct ip_tunnel_info *info,
1224 u8 protocol,
1225 bool use_cache)
1226{
1227 struct dst_entry *dst = NULL;
1228#ifdef CONFIG_DST_CACHE
1229 struct dst_cache *dst_cache;
1230#endif
1231 struct flowi6 fl6;
1232 __u8 prio;
1233
1234#ifdef CONFIG_DST_CACHE
1235 dst_cache = (struct dst_cache *)&info->dst_cache;
1236 if (use_cache) {
1237 dst = dst_cache_get_ip6(dst_cache, saddr);
1238 if (dst)
1239 return dst;
1240 }
1241#endif
1242 memset(&fl6, 0, sizeof(fl6));
1243 fl6.flowi6_mark = skb->mark;
1244 fl6.flowi6_proto = protocol;
1245 fl6.daddr = info->key.u.ipv6.dst;
1246 fl6.saddr = info->key.u.ipv6.src;
1247 prio = info->key.tos;
1248 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(prio),
1249 info->key.label);
1250
1251 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1252 NULL);
1253 if (IS_ERR(dst)) {
1254 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1255 return ERR_PTR(-ENETUNREACH);
1256 }
1257 if (dst->dev == dev) { /* is this necessary? */
1258 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1259 dst_release(dst);
1260 return ERR_PTR(-ELOOP);
1261 }
1262#ifdef CONFIG_DST_CACHE
1263 if (use_cache)
1264 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1265#endif
1266 *saddr = fl6.saddr;
1267 return dst;
1268}
1269EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1270
1271static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1272 gfp_t gfp)
1273{
1274 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1275}
1276
1277static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1278 gfp_t gfp)
1279{
1280 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1281}
1282
1283static void ip6_append_data_mtu(unsigned int *mtu,
1284 int *maxfraglen,
1285 unsigned int fragheaderlen,
1286 struct sk_buff *skb,
1287 struct rt6_info *rt,
1288 unsigned int orig_mtu)
1289{
1290 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1291 if (!skb) {
1292 /* first fragment, reserve header_len */
1293 *mtu = orig_mtu - rt->dst.header_len;
1294
1295 } else {
1296 /*
1297 * this fragment is not first, the headers
1298 * space is regarded as data space.
1299 */
1300 *mtu = orig_mtu;
1301 }
1302 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1303 + fragheaderlen - sizeof(struct frag_hdr);
1304 }
1305}
1306
1307static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1308 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1309 struct rt6_info *rt, struct flowi6 *fl6)
1310{
1311 struct ipv6_pinfo *np = inet6_sk(sk);
1312 unsigned int mtu;
1313 struct ipv6_txoptions *opt = ipc6->opt;
1314
1315 /*
1316 * setup for corking
1317 */
1318 if (opt) {
1319 if (WARN_ON(v6_cork->opt))
1320 return -EINVAL;
1321
1322 v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1323 if (unlikely(!v6_cork->opt))
1324 return -ENOBUFS;
1325
1326 v6_cork->opt->tot_len = sizeof(*opt);
1327 v6_cork->opt->opt_flen = opt->opt_flen;
1328 v6_cork->opt->opt_nflen = opt->opt_nflen;
1329
1330 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1331 sk->sk_allocation);
1332 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1333 return -ENOBUFS;
1334
1335 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1336 sk->sk_allocation);
1337 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1338 return -ENOBUFS;
1339
1340 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1341 sk->sk_allocation);
1342 if (opt->hopopt && !v6_cork->opt->hopopt)
1343 return -ENOBUFS;
1344
1345 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1346 sk->sk_allocation);
1347 if (opt->srcrt && !v6_cork->opt->srcrt)
1348 return -ENOBUFS;
1349
1350 /* need source address above miyazawa*/
1351 }
1352 dst_hold(&rt->dst);
1353 cork->base.dst = &rt->dst;
1354 cork->fl.u.ip6 = *fl6;
1355 v6_cork->hop_limit = ipc6->hlimit;
1356 v6_cork->tclass = ipc6->tclass;
1357 if (rt->dst.flags & DST_XFRM_TUNNEL)
1358 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1359 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1360 else
1361 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1362 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1363 if (np->frag_size < mtu) {
1364 if (np->frag_size)
1365 mtu = np->frag_size;
1366 }
1367 if (mtu < IPV6_MIN_MTU)
1368 return -EINVAL;
1369 cork->base.fragsize = mtu;
1370 cork->base.gso_size = ipc6->gso_size;
1371 cork->base.tx_flags = 0;
1372 cork->base.mark = ipc6->sockc.mark;
1373 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1374
1375 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1376 cork->base.flags |= IPCORK_ALLFRAG;
1377 cork->base.length = 0;
1378
1379 cork->base.transmit_time = ipc6->sockc.transmit_time;
1380
1381 return 0;
1382}
1383
1384static int __ip6_append_data(struct sock *sk,
1385 struct flowi6 *fl6,
1386 struct sk_buff_head *queue,
1387 struct inet_cork *cork,
1388 struct inet6_cork *v6_cork,
1389 struct page_frag *pfrag,
1390 int getfrag(void *from, char *to, int offset,
1391 int len, int odd, struct sk_buff *skb),
1392 void *from, int length, int transhdrlen,
1393 unsigned int flags, struct ipcm6_cookie *ipc6)
1394{
1395 struct sk_buff *skb, *skb_prev = NULL;
1396 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1397 struct ubuf_info *uarg = NULL;
1398 int exthdrlen = 0;
1399 int dst_exthdrlen = 0;
1400 int hh_len;
1401 int copy;
1402 int err;
1403 int offset = 0;
1404 u32 tskey = 0;
1405 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1406 struct ipv6_txoptions *opt = v6_cork->opt;
1407 int csummode = CHECKSUM_NONE;
1408 unsigned int maxnonfragsize, headersize;
1409 unsigned int wmem_alloc_delta = 0;
1410 bool paged, extra_uref = false;
1411
1412 skb = skb_peek_tail(queue);
1413 if (!skb) {
1414 exthdrlen = opt ? opt->opt_flen : 0;
1415 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1416 }
1417
1418 paged = !!cork->gso_size;
1419 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1420 orig_mtu = mtu;
1421
1422 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1423 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1424 tskey = sk->sk_tskey++;
1425
1426 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1427
1428 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1429 (opt ? opt->opt_nflen : 0);
1430 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1431 sizeof(struct frag_hdr);
1432
1433 headersize = sizeof(struct ipv6hdr) +
1434 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1435 (dst_allfrag(&rt->dst) ?
1436 sizeof(struct frag_hdr) : 0) +
1437 rt->rt6i_nfheader_len;
1438
1439 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1440 * the first fragment
1441 */
1442 if (headersize + transhdrlen > mtu)
1443 goto emsgsize;
1444
1445 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1446 (sk->sk_protocol == IPPROTO_UDP ||
1447 sk->sk_protocol == IPPROTO_RAW)) {
1448 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1449 sizeof(struct ipv6hdr));
1450 goto emsgsize;
1451 }
1452
1453 if (ip6_sk_ignore_df(sk))
1454 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1455 else
1456 maxnonfragsize = mtu;
1457
1458 if (cork->length + length > maxnonfragsize - headersize) {
1459emsgsize:
1460 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1461 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1462 return -EMSGSIZE;
1463 }
1464
1465 /* CHECKSUM_PARTIAL only with no extension headers and when
1466 * we are not going to fragment
1467 */
1468 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1469 headersize == sizeof(struct ipv6hdr) &&
1470 length <= mtu - headersize &&
1471 (!(flags & MSG_MORE) || cork->gso_size) &&
1472 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1473 csummode = CHECKSUM_PARTIAL;
1474
1475 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1476 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
1477 if (!uarg)
1478 return -ENOBUFS;
1479 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1480 if (rt->dst.dev->features & NETIF_F_SG &&
1481 csummode == CHECKSUM_PARTIAL) {
1482 paged = true;
1483 } else {
1484 uarg->zerocopy = 0;
1485 skb_zcopy_set(skb, uarg, &extra_uref);
1486 }
1487 }
1488
1489 /*
1490 * Let's try using as much space as possible.
1491 * Use MTU if total length of the message fits into the MTU.
1492 * Otherwise, we need to reserve fragment header and
1493 * fragment alignment (= 8-15 octects, in total).
1494 *
1495 * Note that we may need to "move" the data from the tail of
1496 * of the buffer to the new fragment when we split
1497 * the message.
1498 *
1499 * FIXME: It may be fragmented into multiple chunks
1500 * at once if non-fragmentable extension headers
1501 * are too large.
1502 * --yoshfuji
1503 */
1504
1505 cork->length += length;
1506 if (!skb)
1507 goto alloc_new_skb;
1508
1509 while (length > 0) {
1510 /* Check if the remaining data fits into current packet. */
1511 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1512 if (copy < length)
1513 copy = maxfraglen - skb->len;
1514
1515 if (copy <= 0) {
1516 char *data;
1517 unsigned int datalen;
1518 unsigned int fraglen;
1519 unsigned int fraggap;
1520 unsigned int alloclen;
1521 unsigned int pagedlen;
1522alloc_new_skb:
1523 /* There's no room in the current skb */
1524 if (skb)
1525 fraggap = skb->len - maxfraglen;
1526 else
1527 fraggap = 0;
1528 /* update mtu and maxfraglen if necessary */
1529 if (!skb || !skb_prev)
1530 ip6_append_data_mtu(&mtu, &maxfraglen,
1531 fragheaderlen, skb, rt,
1532 orig_mtu);
1533
1534 skb_prev = skb;
1535
1536 /*
1537 * If remaining data exceeds the mtu,
1538 * we know we need more fragment(s).
1539 */
1540 datalen = length + fraggap;
1541
1542 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1543 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1544 fraglen = datalen + fragheaderlen;
1545 pagedlen = 0;
1546
1547 if ((flags & MSG_MORE) &&
1548 !(rt->dst.dev->features&NETIF_F_SG))
1549 alloclen = mtu;
1550 else if (!paged)
1551 alloclen = fraglen;
1552 else {
1553 alloclen = min_t(int, fraglen, MAX_HEADER);
1554 pagedlen = fraglen - alloclen;
1555 }
1556
1557 alloclen += dst_exthdrlen;
1558
1559 if (datalen != length + fraggap) {
1560 /*
1561 * this is not the last fragment, the trailer
1562 * space is regarded as data space.
1563 */
1564 datalen += rt->dst.trailer_len;
1565 }
1566
1567 alloclen += rt->dst.trailer_len;
1568 fraglen = datalen + fragheaderlen;
1569
1570 /*
1571 * We just reserve space for fragment header.
1572 * Note: this may be overallocation if the message
1573 * (without MSG_MORE) fits into the MTU.
1574 */
1575 alloclen += sizeof(struct frag_hdr);
1576
1577 copy = datalen - transhdrlen - fraggap - pagedlen;
1578 if (copy < 0) {
1579 err = -EINVAL;
1580 goto error;
1581 }
1582 if (transhdrlen) {
1583 skb = sock_alloc_send_skb(sk,
1584 alloclen + hh_len,
1585 (flags & MSG_DONTWAIT), &err);
1586 } else {
1587 skb = NULL;
1588 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1589 2 * sk->sk_sndbuf)
1590 skb = alloc_skb(alloclen + hh_len,
1591 sk->sk_allocation);
1592 if (unlikely(!skb))
1593 err = -ENOBUFS;
1594 }
1595 if (!skb)
1596 goto error;
1597 /*
1598 * Fill in the control structures
1599 */
1600 skb->protocol = htons(ETH_P_IPV6);
1601 skb->ip_summed = csummode;
1602 skb->csum = 0;
1603 /* reserve for fragmentation and ipsec header */
1604 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1605 dst_exthdrlen);
1606
1607 /*
1608 * Find where to start putting bytes
1609 */
1610 data = skb_put(skb, fraglen - pagedlen);
1611 skb_set_network_header(skb, exthdrlen);
1612 data += fragheaderlen;
1613 skb->transport_header = (skb->network_header +
1614 fragheaderlen);
1615 if (fraggap) {
1616 skb->csum = skb_copy_and_csum_bits(
1617 skb_prev, maxfraglen,
1618 data + transhdrlen, fraggap, 0);
1619 skb_prev->csum = csum_sub(skb_prev->csum,
1620 skb->csum);
1621 data += fraggap;
1622 pskb_trim_unique(skb_prev, maxfraglen);
1623 }
1624 if (copy > 0 &&
1625 getfrag(from, data + transhdrlen, offset,
1626 copy, fraggap, skb) < 0) {
1627 err = -EFAULT;
1628 kfree_skb(skb);
1629 goto error;
1630 }
1631
1632 offset += copy;
1633 length -= copy + transhdrlen;
1634 transhdrlen = 0;
1635 exthdrlen = 0;
1636 dst_exthdrlen = 0;
1637
1638 /* Only the initial fragment is time stamped */
1639 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1640 cork->tx_flags = 0;
1641 skb_shinfo(skb)->tskey = tskey;
1642 tskey = 0;
1643 skb_zcopy_set(skb, uarg, &extra_uref);
1644
1645 if ((flags & MSG_CONFIRM) && !skb_prev)
1646 skb_set_dst_pending_confirm(skb, 1);
1647
1648 /*
1649 * Put the packet on the pending queue
1650 */
1651 if (!skb->destructor) {
1652 skb->destructor = sock_wfree;
1653 skb->sk = sk;
1654 wmem_alloc_delta += skb->truesize;
1655 }
1656 __skb_queue_tail(queue, skb);
1657 continue;
1658 }
1659
1660 if (copy > length)
1661 copy = length;
1662
1663 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1664 skb_tailroom(skb) >= copy) {
1665 unsigned int off;
1666
1667 off = skb->len;
1668 if (getfrag(from, skb_put(skb, copy),
1669 offset, copy, off, skb) < 0) {
1670 __skb_trim(skb, off);
1671 err = -EFAULT;
1672 goto error;
1673 }
1674 } else if (!uarg || !uarg->zerocopy) {
1675 int i = skb_shinfo(skb)->nr_frags;
1676
1677 err = -ENOMEM;
1678 if (!sk_page_frag_refill(sk, pfrag))
1679 goto error;
1680
1681 if (!skb_can_coalesce(skb, i, pfrag->page,
1682 pfrag->offset)) {
1683 err = -EMSGSIZE;
1684 if (i == MAX_SKB_FRAGS)
1685 goto error;
1686
1687 __skb_fill_page_desc(skb, i, pfrag->page,
1688 pfrag->offset, 0);
1689 skb_shinfo(skb)->nr_frags = ++i;
1690 get_page(pfrag->page);
1691 }
1692 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1693 if (getfrag(from,
1694 page_address(pfrag->page) + pfrag->offset,
1695 offset, copy, skb->len, skb) < 0)
1696 goto error_efault;
1697
1698 pfrag->offset += copy;
1699 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1700 skb->len += copy;
1701 skb->data_len += copy;
1702 skb->truesize += copy;
1703 wmem_alloc_delta += copy;
1704 } else {
1705 err = skb_zerocopy_iter_dgram(skb, from, copy);
1706 if (err < 0)
1707 goto error;
1708 }
1709 offset += copy;
1710 length -= copy;
1711 }
1712
1713 if (wmem_alloc_delta)
1714 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1715 return 0;
1716
1717error_efault:
1718 err = -EFAULT;
1719error:
1720 if (uarg)
1721 sock_zerocopy_put_abort(uarg, extra_uref);
1722 cork->length -= length;
1723 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1724 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1725 return err;
1726}
1727
1728int ip6_append_data(struct sock *sk,
1729 int getfrag(void *from, char *to, int offset, int len,
1730 int odd, struct sk_buff *skb),
1731 void *from, int length, int transhdrlen,
1732 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1733 struct rt6_info *rt, unsigned int flags)
1734{
1735 struct inet_sock *inet = inet_sk(sk);
1736 struct ipv6_pinfo *np = inet6_sk(sk);
1737 int exthdrlen;
1738 int err;
1739
1740 if (flags&MSG_PROBE)
1741 return 0;
1742 if (skb_queue_empty(&sk->sk_write_queue)) {
1743 /*
1744 * setup for corking
1745 */
1746 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1747 ipc6, rt, fl6);
1748 if (err)
1749 return err;
1750
1751 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1752 length += exthdrlen;
1753 transhdrlen += exthdrlen;
1754 } else {
1755 fl6 = &inet->cork.fl.u.ip6;
1756 transhdrlen = 0;
1757 }
1758
1759 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1760 &np->cork, sk_page_frag(sk), getfrag,
1761 from, length, transhdrlen, flags, ipc6);
1762}
1763EXPORT_SYMBOL_GPL(ip6_append_data);
1764
1765static void ip6_cork_release(struct inet_cork_full *cork,
1766 struct inet6_cork *v6_cork)
1767{
1768 if (v6_cork->opt) {
1769 kfree(v6_cork->opt->dst0opt);
1770 kfree(v6_cork->opt->dst1opt);
1771 kfree(v6_cork->opt->hopopt);
1772 kfree(v6_cork->opt->srcrt);
1773 kfree(v6_cork->opt);
1774 v6_cork->opt = NULL;
1775 }
1776
1777 if (cork->base.dst) {
1778 dst_release(cork->base.dst);
1779 cork->base.dst = NULL;
1780 cork->base.flags &= ~IPCORK_ALLFRAG;
1781 }
1782 memset(&cork->fl, 0, sizeof(cork->fl));
1783}
1784
1785struct sk_buff *__ip6_make_skb(struct sock *sk,
1786 struct sk_buff_head *queue,
1787 struct inet_cork_full *cork,
1788 struct inet6_cork *v6_cork)
1789{
1790 struct sk_buff *skb, *tmp_skb;
1791 struct sk_buff **tail_skb;
1792 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1793 struct ipv6_pinfo *np = inet6_sk(sk);
1794 struct net *net = sock_net(sk);
1795 struct ipv6hdr *hdr;
1796 struct ipv6_txoptions *opt = v6_cork->opt;
1797 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1798 struct flowi6 *fl6 = &cork->fl.u.ip6;
1799 unsigned char proto = fl6->flowi6_proto;
1800
1801 skb = __skb_dequeue(queue);
1802 if (!skb)
1803 goto out;
1804 tail_skb = &(skb_shinfo(skb)->frag_list);
1805
1806 /* move skb->data to ip header from ext header */
1807 if (skb->data < skb_network_header(skb))
1808 __skb_pull(skb, skb_network_offset(skb));
1809 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1810 __skb_pull(tmp_skb, skb_network_header_len(skb));
1811 *tail_skb = tmp_skb;
1812 tail_skb = &(tmp_skb->next);
1813 skb->len += tmp_skb->len;
1814 skb->data_len += tmp_skb->len;
1815 skb->truesize += tmp_skb->truesize;
1816 tmp_skb->destructor = NULL;
1817 tmp_skb->sk = NULL;
1818 }
1819
1820 /* Allow local fragmentation. */
1821 skb->ignore_df = ip6_sk_ignore_df(sk);
1822
1823 *final_dst = fl6->daddr;
1824 __skb_pull(skb, skb_network_header_len(skb));
1825 if (opt && opt->opt_flen)
1826 ipv6_push_frag_opts(skb, opt, &proto);
1827 if (opt && opt->opt_nflen)
1828 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1829
1830 skb_push(skb, sizeof(struct ipv6hdr));
1831 skb_reset_network_header(skb);
1832 hdr = ipv6_hdr(skb);
1833
1834 ip6_flow_hdr(hdr, v6_cork->tclass,
1835 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1836 ip6_autoflowlabel(net, np), fl6));
1837 hdr->hop_limit = v6_cork->hop_limit;
1838 hdr->nexthdr = proto;
1839 hdr->saddr = fl6->saddr;
1840 hdr->daddr = *final_dst;
1841
1842 skb->priority = sk->sk_priority;
1843 skb->mark = cork->base.mark;
1844
1845 skb->tstamp = cork->base.transmit_time;
1846
1847 skb_dst_set(skb, dst_clone(&rt->dst));
1848 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1849 if (proto == IPPROTO_ICMPV6) {
1850 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1851
1852 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1853 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1854 }
1855
1856 ip6_cork_release(cork, v6_cork);
1857out:
1858 return skb;
1859}
1860
1861int ip6_send_skb(struct sk_buff *skb)
1862{
1863 struct net *net = sock_net(skb->sk);
1864 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1865 int err;
1866
1867 err = ip6_local_out(net, skb->sk, skb);
1868 if (err) {
1869 if (err > 0)
1870 err = net_xmit_errno(err);
1871 if (err)
1872 IP6_INC_STATS(net, rt->rt6i_idev,
1873 IPSTATS_MIB_OUTDISCARDS);
1874 }
1875
1876 return err;
1877}
1878
1879int ip6_push_pending_frames(struct sock *sk)
1880{
1881 struct sk_buff *skb;
1882
1883 skb = ip6_finish_skb(sk);
1884 if (!skb)
1885 return 0;
1886
1887 return ip6_send_skb(skb);
1888}
1889EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1890
1891static void __ip6_flush_pending_frames(struct sock *sk,
1892 struct sk_buff_head *queue,
1893 struct inet_cork_full *cork,
1894 struct inet6_cork *v6_cork)
1895{
1896 struct sk_buff *skb;
1897
1898 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1899 if (skb_dst(skb))
1900 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1901 IPSTATS_MIB_OUTDISCARDS);
1902 kfree_skb(skb);
1903 }
1904
1905 ip6_cork_release(cork, v6_cork);
1906}
1907
1908void ip6_flush_pending_frames(struct sock *sk)
1909{
1910 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1911 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1912}
1913EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1914
1915struct sk_buff *ip6_make_skb(struct sock *sk,
1916 int getfrag(void *from, char *to, int offset,
1917 int len, int odd, struct sk_buff *skb),
1918 void *from, int length, int transhdrlen,
1919 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1920 struct rt6_info *rt, unsigned int flags,
1921 struct inet_cork_full *cork)
1922{
1923 struct inet6_cork v6_cork;
1924 struct sk_buff_head queue;
1925 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1926 int err;
1927
1928 if (flags & MSG_PROBE)
1929 return NULL;
1930
1931 __skb_queue_head_init(&queue);
1932
1933 cork->base.flags = 0;
1934 cork->base.addr = 0;
1935 cork->base.opt = NULL;
1936 cork->base.dst = NULL;
1937 v6_cork.opt = NULL;
1938 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt, fl6);
1939 if (err) {
1940 ip6_cork_release(cork, &v6_cork);
1941 return ERR_PTR(err);
1942 }
1943 if (ipc6->dontfrag < 0)
1944 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
1945
1946 err = __ip6_append_data(sk, fl6, &queue, &cork->base, &v6_cork,
1947 ¤t->task_frag, getfrag, from,
1948 length + exthdrlen, transhdrlen + exthdrlen,
1949 flags, ipc6);
1950 if (err) {
1951 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
1952 return ERR_PTR(err);
1953 }
1954
1955 return __ip6_make_skb(sk, &queue, cork, &v6_cork);
1956}