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 struct inet6_dev *idev = ip6_dst_idev(dst);
64 unsigned int hh_len = LL_RESERVED_SPACE(dev);
65 const struct in6_addr *daddr, *nexthop;
66 struct ipv6hdr *hdr;
67 struct neighbour *neigh;
68 int ret;
69
70 /* Be paranoid, rather than too clever. */
71 if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
72 skb = skb_expand_head(skb, hh_len);
73 if (!skb) {
74 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
75 return -ENOMEM;
76 }
77 }
78
79 hdr = ipv6_hdr(skb);
80 daddr = &hdr->daddr;
81 if (ipv6_addr_is_multicast(daddr)) {
82 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
83 ((mroute6_is_socket(net, skb) &&
84 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
85 ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
86 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
87
88 /* Do not check for IFF_ALLMULTI; multicast routing
89 is not supported in any case.
90 */
91 if (newskb)
92 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
93 net, sk, newskb, NULL, newskb->dev,
94 dev_loopback_xmit);
95
96 if (hdr->hop_limit == 0) {
97 IP6_INC_STATS(net, idev,
98 IPSTATS_MIB_OUTDISCARDS);
99 kfree_skb(skb);
100 return 0;
101 }
102 }
103
104 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
105 if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
106 !(dev->flags & IFF_LOOPBACK)) {
107 kfree_skb(skb);
108 return 0;
109 }
110 }
111
112 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
113 int res = lwtunnel_xmit(skb);
114
115 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
116 return res;
117 }
118
119 rcu_read_lock_bh();
120 nexthop = rt6_nexthop((struct rt6_info *)dst, daddr);
121 neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
122
123 if (unlikely(IS_ERR_OR_NULL(neigh))) {
124 if (unlikely(!neigh))
125 neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
126 if (IS_ERR(neigh)) {
127 rcu_read_unlock_bh();
128 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
129 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
130 return -EINVAL;
131 }
132 }
133 sock_confirm_neigh(skb, neigh);
134 ret = neigh_output(neigh, skb, false);
135 rcu_read_unlock_bh();
136 return ret;
137}
138
139static int
140ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
141 struct sk_buff *skb, unsigned int mtu)
142{
143 struct sk_buff *segs, *nskb;
144 netdev_features_t features;
145 int ret = 0;
146
147 /* Please see corresponding comment in ip_finish_output_gso
148 * describing the cases where GSO segment length exceeds the
149 * egress MTU.
150 */
151 features = netif_skb_features(skb);
152 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
153 if (IS_ERR_OR_NULL(segs)) {
154 kfree_skb(skb);
155 return -ENOMEM;
156 }
157
158 consume_skb(skb);
159
160 skb_list_walk_safe(segs, segs, nskb) {
161 int err;
162
163 skb_mark_not_on_list(segs);
164 err = ip6_fragment(net, sk, segs, ip6_finish_output2);
165 if (err && ret == 0)
166 ret = err;
167 }
168
169 return ret;
170}
171
172static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
173{
174 unsigned int mtu;
175
176#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
177 /* Policy lookup after SNAT yielded a new policy */
178 if (skb_dst(skb)->xfrm) {
179 IP6CB(skb)->flags |= IP6SKB_REROUTED;
180 return dst_output(net, sk, skb);
181 }
182#endif
183
184 mtu = ip6_skb_dst_mtu(skb);
185 if (skb_is_gso(skb) &&
186 !(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
187 !skb_gso_validate_network_len(skb, mtu))
188 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
189
190 if ((skb->len > mtu && !skb_is_gso(skb)) ||
191 dst_allfrag(skb_dst(skb)) ||
192 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
193 return ip6_fragment(net, sk, skb, ip6_finish_output2);
194 else
195 return ip6_finish_output2(net, sk, skb);
196}
197
198static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
199{
200 int ret;
201
202 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
203 switch (ret) {
204 case NET_XMIT_SUCCESS:
205 case NET_XMIT_CN:
206 return __ip6_finish_output(net, sk, skb) ? : ret;
207 default:
208 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
209 return ret;
210 }
211}
212
213int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
214{
215 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
216 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
217
218 skb->protocol = htons(ETH_P_IPV6);
219 skb->dev = dev;
220
221 if (unlikely(idev->cnf.disable_ipv6)) {
222 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
223 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
224 return 0;
225 }
226
227 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
228 net, sk, skb, indev, dev,
229 ip6_finish_output,
230 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
231}
232EXPORT_SYMBOL(ip6_output);
233
234bool ip6_autoflowlabel(struct net *net, const struct ipv6_pinfo *np)
235{
236 if (!np->autoflowlabel_set)
237 return ip6_default_np_autolabel(net);
238 else
239 return np->autoflowlabel;
240}
241
242/*
243 * xmit an sk_buff (used by TCP, SCTP and DCCP)
244 * Note : socket lock is not held for SYNACK packets, but might be modified
245 * by calls to skb_set_owner_w() and ipv6_local_error(),
246 * which are using proper atomic operations or spinlocks.
247 */
248int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
249 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
250{
251 struct net *net = sock_net(sk);
252 const struct ipv6_pinfo *np = inet6_sk(sk);
253 struct in6_addr *first_hop = &fl6->daddr;
254 struct dst_entry *dst = skb_dst(skb);
255 struct net_device *dev = dst->dev;
256 struct inet6_dev *idev = ip6_dst_idev(dst);
257 struct hop_jumbo_hdr *hop_jumbo;
258 int hoplen = sizeof(*hop_jumbo);
259 unsigned int head_room;
260 struct ipv6hdr *hdr;
261 u8 proto = fl6->flowi6_proto;
262 int seg_len = skb->len;
263 int hlimit = -1;
264 u32 mtu;
265
266 head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
267 if (opt)
268 head_room += opt->opt_nflen + opt->opt_flen;
269
270 if (unlikely(head_room > skb_headroom(skb))) {
271 skb = skb_expand_head(skb, head_room);
272 if (!skb) {
273 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
274 return -ENOBUFS;
275 }
276 }
277
278 if (opt) {
279 seg_len += opt->opt_nflen + opt->opt_flen;
280
281 if (opt->opt_flen)
282 ipv6_push_frag_opts(skb, opt, &proto);
283
284 if (opt->opt_nflen)
285 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
286 &fl6->saddr);
287 }
288
289 if (unlikely(seg_len > IPV6_MAXPLEN)) {
290 hop_jumbo = skb_push(skb, hoplen);
291
292 hop_jumbo->nexthdr = proto;
293 hop_jumbo->hdrlen = 0;
294 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
295 hop_jumbo->tlv_len = 4;
296 hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
297
298 proto = IPPROTO_HOPOPTS;
299 seg_len = 0;
300 IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
301 }
302
303 skb_push(skb, sizeof(struct ipv6hdr));
304 skb_reset_network_header(skb);
305 hdr = ipv6_hdr(skb);
306
307 /*
308 * Fill in the IPv6 header
309 */
310 if (np)
311 hlimit = np->hop_limit;
312 if (hlimit < 0)
313 hlimit = ip6_dst_hoplimit(dst);
314
315 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
316 ip6_autoflowlabel(net, np), fl6));
317
318 hdr->payload_len = htons(seg_len);
319 hdr->nexthdr = proto;
320 hdr->hop_limit = hlimit;
321
322 hdr->saddr = fl6->saddr;
323 hdr->daddr = *first_hop;
324
325 skb->protocol = htons(ETH_P_IPV6);
326 skb->priority = priority;
327 skb->mark = mark;
328
329 mtu = dst_mtu(dst);
330 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
331 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
332
333 /* if egress device is enslaved to an L3 master device pass the
334 * skb to its handler for processing
335 */
336 skb = l3mdev_ip6_out((struct sock *)sk, skb);
337 if (unlikely(!skb))
338 return 0;
339
340 /* hooks should never assume socket lock is held.
341 * we promote our socket to non const
342 */
343 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
344 net, (struct sock *)sk, skb, NULL, dev,
345 dst_output);
346 }
347
348 skb->dev = dev;
349 /* ipv6_local_error() does not require socket lock,
350 * we promote our socket to non const
351 */
352 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
353
354 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
355 kfree_skb(skb);
356 return -EMSGSIZE;
357}
358EXPORT_SYMBOL(ip6_xmit);
359
360static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
361{
362 struct ip6_ra_chain *ra;
363 struct sock *last = NULL;
364
365 read_lock(&ip6_ra_lock);
366 for (ra = ip6_ra_chain; ra; ra = ra->next) {
367 struct sock *sk = ra->sk;
368 if (sk && ra->sel == sel &&
369 (!sk->sk_bound_dev_if ||
370 sk->sk_bound_dev_if == skb->dev->ifindex)) {
371 struct ipv6_pinfo *np = inet6_sk(sk);
372
373 if (np && np->rtalert_isolate &&
374 !net_eq(sock_net(sk), dev_net(skb->dev))) {
375 continue;
376 }
377 if (last) {
378 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
379 if (skb2)
380 rawv6_rcv(last, skb2);
381 }
382 last = sk;
383 }
384 }
385
386 if (last) {
387 rawv6_rcv(last, skb);
388 read_unlock(&ip6_ra_lock);
389 return 1;
390 }
391 read_unlock(&ip6_ra_lock);
392 return 0;
393}
394
395static int ip6_forward_proxy_check(struct sk_buff *skb)
396{
397 struct ipv6hdr *hdr = ipv6_hdr(skb);
398 u8 nexthdr = hdr->nexthdr;
399 __be16 frag_off;
400 int offset;
401
402 if (ipv6_ext_hdr(nexthdr)) {
403 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
404 if (offset < 0)
405 return 0;
406 } else
407 offset = sizeof(struct ipv6hdr);
408
409 if (nexthdr == IPPROTO_ICMPV6) {
410 struct icmp6hdr *icmp6;
411
412 if (!pskb_may_pull(skb, (skb_network_header(skb) +
413 offset + 1 - skb->data)))
414 return 0;
415
416 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
417
418 switch (icmp6->icmp6_type) {
419 case NDISC_ROUTER_SOLICITATION:
420 case NDISC_ROUTER_ADVERTISEMENT:
421 case NDISC_NEIGHBOUR_SOLICITATION:
422 case NDISC_NEIGHBOUR_ADVERTISEMENT:
423 case NDISC_REDIRECT:
424 /* For reaction involving unicast neighbor discovery
425 * message destined to the proxied address, pass it to
426 * input function.
427 */
428 return 1;
429 default:
430 break;
431 }
432 }
433
434 /*
435 * The proxying router can't forward traffic sent to a link-local
436 * address, so signal the sender and discard the packet. This
437 * behavior is clarified by the MIPv6 specification.
438 */
439 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
440 dst_link_failure(skb);
441 return -1;
442 }
443
444 return 0;
445}
446
447static inline int ip6_forward_finish(struct net *net, struct sock *sk,
448 struct sk_buff *skb)
449{
450 struct dst_entry *dst = skb_dst(skb);
451
452 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
453 __IP6_ADD_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
454
455#ifdef CONFIG_NET_SWITCHDEV
456 if (skb->offload_l3_fwd_mark) {
457 consume_skb(skb);
458 return 0;
459 }
460#endif
461
462 skb_clear_tstamp(skb);
463 return dst_output(net, sk, skb);
464}
465
466static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
467{
468 if (skb->len <= mtu)
469 return false;
470
471 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
472 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
473 return true;
474
475 if (skb->ignore_df)
476 return false;
477
478 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
479 return false;
480
481 return true;
482}
483
484int ip6_forward(struct sk_buff *skb)
485{
486 struct dst_entry *dst = skb_dst(skb);
487 struct ipv6hdr *hdr = ipv6_hdr(skb);
488 struct inet6_skb_parm *opt = IP6CB(skb);
489 struct net *net = dev_net(dst->dev);
490 struct inet6_dev *idev;
491 SKB_DR(reason);
492 u32 mtu;
493
494 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
495 if (net->ipv6.devconf_all->forwarding == 0)
496 goto error;
497
498 if (skb->pkt_type != PACKET_HOST)
499 goto drop;
500
501 if (unlikely(skb->sk))
502 goto drop;
503
504 if (skb_warn_if_lro(skb))
505 goto drop;
506
507 if (!net->ipv6.devconf_all->disable_policy &&
508 (!idev || !idev->cnf.disable_policy) &&
509 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
510 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
511 goto drop;
512 }
513
514 skb_forward_csum(skb);
515
516 /*
517 * We DO NOT make any processing on
518 * RA packets, pushing them to user level AS IS
519 * without ane WARRANTY that application will be able
520 * to interpret them. The reason is that we
521 * cannot make anything clever here.
522 *
523 * We are not end-node, so that if packet contains
524 * AH/ESP, we cannot make anything.
525 * Defragmentation also would be mistake, RA packets
526 * cannot be fragmented, because there is no warranty
527 * that different fragments will go along one path. --ANK
528 */
529 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
530 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
531 return 0;
532 }
533
534 /*
535 * check and decrement ttl
536 */
537 if (hdr->hop_limit <= 1) {
538 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
539 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
540
541 kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
542 return -ETIMEDOUT;
543 }
544
545 /* XXX: idev->cnf.proxy_ndp? */
546 if (net->ipv6.devconf_all->proxy_ndp &&
547 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
548 int proxied = ip6_forward_proxy_check(skb);
549 if (proxied > 0) {
550 /* It's tempting to decrease the hop limit
551 * here by 1, as we do at the end of the
552 * function too.
553 *
554 * But that would be incorrect, as proxying is
555 * not forwarding. The ip6_input function
556 * will handle this packet locally, and it
557 * depends on the hop limit being unchanged.
558 *
559 * One example is the NDP hop limit, that
560 * always has to stay 255, but other would be
561 * similar checks around RA packets, where the
562 * user can even change the desired limit.
563 */
564 return ip6_input(skb);
565 } else if (proxied < 0) {
566 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
567 goto drop;
568 }
569 }
570
571 if (!xfrm6_route_forward(skb)) {
572 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
573 SKB_DR_SET(reason, XFRM_POLICY);
574 goto drop;
575 }
576 dst = skb_dst(skb);
577
578 /* IPv6 specs say nothing about it, but it is clear that we cannot
579 send redirects to source routed frames.
580 We don't send redirects to frames decapsulated from IPsec.
581 */
582 if (IP6CB(skb)->iif == dst->dev->ifindex &&
583 opt->srcrt == 0 && !skb_sec_path(skb)) {
584 struct in6_addr *target = NULL;
585 struct inet_peer *peer;
586 struct rt6_info *rt;
587
588 /*
589 * incoming and outgoing devices are the same
590 * send a redirect.
591 */
592
593 rt = (struct rt6_info *) dst;
594 if (rt->rt6i_flags & RTF_GATEWAY)
595 target = &rt->rt6i_gateway;
596 else
597 target = &hdr->daddr;
598
599 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
600
601 /* Limit redirects both by destination (here)
602 and by source (inside ndisc_send_redirect)
603 */
604 if (inet_peer_xrlim_allow(peer, 1*HZ))
605 ndisc_send_redirect(skb, target);
606 if (peer)
607 inet_putpeer(peer);
608 } else {
609 int addrtype = ipv6_addr_type(&hdr->saddr);
610
611 /* This check is security critical. */
612 if (addrtype == IPV6_ADDR_ANY ||
613 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
614 goto error;
615 if (addrtype & IPV6_ADDR_LINKLOCAL) {
616 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
617 ICMPV6_NOT_NEIGHBOUR, 0);
618 goto error;
619 }
620 }
621
622 mtu = ip6_dst_mtu_maybe_forward(dst, true);
623 if (mtu < IPV6_MIN_MTU)
624 mtu = IPV6_MIN_MTU;
625
626 if (ip6_pkt_too_big(skb, mtu)) {
627 /* Again, force OUTPUT device used as source address */
628 skb->dev = dst->dev;
629 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
630 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
631 __IP6_INC_STATS(net, ip6_dst_idev(dst),
632 IPSTATS_MIB_FRAGFAILS);
633 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
634 return -EMSGSIZE;
635 }
636
637 if (skb_cow(skb, dst->dev->hard_header_len)) {
638 __IP6_INC_STATS(net, ip6_dst_idev(dst),
639 IPSTATS_MIB_OUTDISCARDS);
640 goto drop;
641 }
642
643 hdr = ipv6_hdr(skb);
644
645 /* Mangling hops number delayed to point after skb COW */
646
647 hdr->hop_limit--;
648
649 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
650 net, NULL, skb, skb->dev, dst->dev,
651 ip6_forward_finish);
652
653error:
654 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
655 SKB_DR_SET(reason, IP_INADDRERRORS);
656drop:
657 kfree_skb_reason(skb, reason);
658 return -EINVAL;
659}
660
661static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
662{
663 to->pkt_type = from->pkt_type;
664 to->priority = from->priority;
665 to->protocol = from->protocol;
666 skb_dst_drop(to);
667 skb_dst_set(to, dst_clone(skb_dst(from)));
668 to->dev = from->dev;
669 to->mark = from->mark;
670
671 skb_copy_hash(to, from);
672
673#ifdef CONFIG_NET_SCHED
674 to->tc_index = from->tc_index;
675#endif
676 nf_copy(to, from);
677 skb_ext_copy(to, from);
678 skb_copy_secmark(to, from);
679}
680
681int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
682 u8 nexthdr, __be32 frag_id,
683 struct ip6_fraglist_iter *iter)
684{
685 unsigned int first_len;
686 struct frag_hdr *fh;
687
688 /* BUILD HEADER */
689 *prevhdr = NEXTHDR_FRAGMENT;
690 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
691 if (!iter->tmp_hdr)
692 return -ENOMEM;
693
694 iter->frag = skb_shinfo(skb)->frag_list;
695 skb_frag_list_init(skb);
696
697 iter->offset = 0;
698 iter->hlen = hlen;
699 iter->frag_id = frag_id;
700 iter->nexthdr = nexthdr;
701
702 __skb_pull(skb, hlen);
703 fh = __skb_push(skb, sizeof(struct frag_hdr));
704 __skb_push(skb, hlen);
705 skb_reset_network_header(skb);
706 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
707
708 fh->nexthdr = nexthdr;
709 fh->reserved = 0;
710 fh->frag_off = htons(IP6_MF);
711 fh->identification = frag_id;
712
713 first_len = skb_pagelen(skb);
714 skb->data_len = first_len - skb_headlen(skb);
715 skb->len = first_len;
716 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
717
718 return 0;
719}
720EXPORT_SYMBOL(ip6_fraglist_init);
721
722void ip6_fraglist_prepare(struct sk_buff *skb,
723 struct ip6_fraglist_iter *iter)
724{
725 struct sk_buff *frag = iter->frag;
726 unsigned int hlen = iter->hlen;
727 struct frag_hdr *fh;
728
729 frag->ip_summed = CHECKSUM_NONE;
730 skb_reset_transport_header(frag);
731 fh = __skb_push(frag, sizeof(struct frag_hdr));
732 __skb_push(frag, hlen);
733 skb_reset_network_header(frag);
734 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
735 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
736 fh->nexthdr = iter->nexthdr;
737 fh->reserved = 0;
738 fh->frag_off = htons(iter->offset);
739 if (frag->next)
740 fh->frag_off |= htons(IP6_MF);
741 fh->identification = iter->frag_id;
742 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
743 ip6_copy_metadata(frag, skb);
744}
745EXPORT_SYMBOL(ip6_fraglist_prepare);
746
747void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
748 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
749 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
750{
751 state->prevhdr = prevhdr;
752 state->nexthdr = nexthdr;
753 state->frag_id = frag_id;
754
755 state->hlen = hlen;
756 state->mtu = mtu;
757
758 state->left = skb->len - hlen; /* Space per frame */
759 state->ptr = hlen; /* Where to start from */
760
761 state->hroom = hdr_room;
762 state->troom = needed_tailroom;
763
764 state->offset = 0;
765}
766EXPORT_SYMBOL(ip6_frag_init);
767
768struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
769{
770 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
771 struct sk_buff *frag;
772 struct frag_hdr *fh;
773 unsigned int len;
774
775 len = state->left;
776 /* IF: it doesn't fit, use 'mtu' - the data space left */
777 if (len > state->mtu)
778 len = state->mtu;
779 /* IF: we are not sending up to and including the packet end
780 then align the next start on an eight byte boundary */
781 if (len < state->left)
782 len &= ~7;
783
784 /* Allocate buffer */
785 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
786 state->hroom + state->troom, GFP_ATOMIC);
787 if (!frag)
788 return ERR_PTR(-ENOMEM);
789
790 /*
791 * Set up data on packet
792 */
793
794 ip6_copy_metadata(frag, skb);
795 skb_reserve(frag, state->hroom);
796 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
797 skb_reset_network_header(frag);
798 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
799 frag->transport_header = (frag->network_header + state->hlen +
800 sizeof(struct frag_hdr));
801
802 /*
803 * Charge the memory for the fragment to any owner
804 * it might possess
805 */
806 if (skb->sk)
807 skb_set_owner_w(frag, skb->sk);
808
809 /*
810 * Copy the packet header into the new buffer.
811 */
812 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
813
814 fragnexthdr_offset = skb_network_header(frag);
815 fragnexthdr_offset += prevhdr - skb_network_header(skb);
816 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
817
818 /*
819 * Build fragment header.
820 */
821 fh->nexthdr = state->nexthdr;
822 fh->reserved = 0;
823 fh->identification = state->frag_id;
824
825 /*
826 * Copy a block of the IP datagram.
827 */
828 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
829 len));
830 state->left -= len;
831
832 fh->frag_off = htons(state->offset);
833 if (state->left > 0)
834 fh->frag_off |= htons(IP6_MF);
835 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
836
837 state->ptr += len;
838 state->offset += len;
839
840 return frag;
841}
842EXPORT_SYMBOL(ip6_frag_next);
843
844int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
845 int (*output)(struct net *, struct sock *, struct sk_buff *))
846{
847 struct sk_buff *frag;
848 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
849 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
850 inet6_sk(skb->sk) : NULL;
851 bool mono_delivery_time = skb->mono_delivery_time;
852 struct ip6_frag_state state;
853 unsigned int mtu, hlen, nexthdr_offset;
854 ktime_t tstamp = skb->tstamp;
855 int hroom, err = 0;
856 __be32 frag_id;
857 u8 *prevhdr, nexthdr = 0;
858
859 err = ip6_find_1stfragopt(skb, &prevhdr);
860 if (err < 0)
861 goto fail;
862 hlen = err;
863 nexthdr = *prevhdr;
864 nexthdr_offset = prevhdr - skb_network_header(skb);
865
866 mtu = ip6_skb_dst_mtu(skb);
867
868 /* We must not fragment if the socket is set to force MTU discovery
869 * or if the skb it not generated by a local socket.
870 */
871 if (unlikely(!skb->ignore_df && skb->len > mtu))
872 goto fail_toobig;
873
874 if (IP6CB(skb)->frag_max_size) {
875 if (IP6CB(skb)->frag_max_size > mtu)
876 goto fail_toobig;
877
878 /* don't send fragments larger than what we received */
879 mtu = IP6CB(skb)->frag_max_size;
880 if (mtu < IPV6_MIN_MTU)
881 mtu = IPV6_MIN_MTU;
882 }
883
884 if (np && np->frag_size < mtu) {
885 if (np->frag_size)
886 mtu = np->frag_size;
887 }
888 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
889 goto fail_toobig;
890 mtu -= hlen + sizeof(struct frag_hdr);
891
892 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
893 &ipv6_hdr(skb)->saddr);
894
895 if (skb->ip_summed == CHECKSUM_PARTIAL &&
896 (err = skb_checksum_help(skb)))
897 goto fail;
898
899 prevhdr = skb_network_header(skb) + nexthdr_offset;
900 hroom = LL_RESERVED_SPACE(rt->dst.dev);
901 if (skb_has_frag_list(skb)) {
902 unsigned int first_len = skb_pagelen(skb);
903 struct ip6_fraglist_iter iter;
904 struct sk_buff *frag2;
905
906 if (first_len - hlen > mtu ||
907 ((first_len - hlen) & 7) ||
908 skb_cloned(skb) ||
909 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
910 goto slow_path;
911
912 skb_walk_frags(skb, frag) {
913 /* Correct geometry. */
914 if (frag->len > mtu ||
915 ((frag->len & 7) && frag->next) ||
916 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
917 goto slow_path_clean;
918
919 /* Partially cloned skb? */
920 if (skb_shared(frag))
921 goto slow_path_clean;
922
923 BUG_ON(frag->sk);
924 if (skb->sk) {
925 frag->sk = skb->sk;
926 frag->destructor = sock_wfree;
927 }
928 skb->truesize -= frag->truesize;
929 }
930
931 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
932 &iter);
933 if (err < 0)
934 goto fail;
935
936 /* We prevent @rt from being freed. */
937 rcu_read_lock();
938
939 for (;;) {
940 /* Prepare header of the next frame,
941 * before previous one went down. */
942 if (iter.frag)
943 ip6_fraglist_prepare(skb, &iter);
944
945 skb_set_delivery_time(skb, tstamp, mono_delivery_time);
946 err = output(net, sk, skb);
947 if (!err)
948 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
949 IPSTATS_MIB_FRAGCREATES);
950
951 if (err || !iter.frag)
952 break;
953
954 skb = ip6_fraglist_next(&iter);
955 }
956
957 kfree(iter.tmp_hdr);
958
959 if (err == 0) {
960 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
961 IPSTATS_MIB_FRAGOKS);
962 rcu_read_unlock();
963 return 0;
964 }
965
966 kfree_skb_list(iter.frag);
967
968 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
969 IPSTATS_MIB_FRAGFAILS);
970 rcu_read_unlock();
971 return err;
972
973slow_path_clean:
974 skb_walk_frags(skb, frag2) {
975 if (frag2 == frag)
976 break;
977 frag2->sk = NULL;
978 frag2->destructor = NULL;
979 skb->truesize += frag2->truesize;
980 }
981 }
982
983slow_path:
984 /*
985 * Fragment the datagram.
986 */
987
988 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
989 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
990 &state);
991
992 /*
993 * Keep copying data until we run out.
994 */
995
996 while (state.left > 0) {
997 frag = ip6_frag_next(skb, &state);
998 if (IS_ERR(frag)) {
999 err = PTR_ERR(frag);
1000 goto fail;
1001 }
1002
1003 /*
1004 * Put this fragment into the sending queue.
1005 */
1006 skb_set_delivery_time(frag, tstamp, mono_delivery_time);
1007 err = output(net, sk, frag);
1008 if (err)
1009 goto fail;
1010
1011 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1012 IPSTATS_MIB_FRAGCREATES);
1013 }
1014 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1015 IPSTATS_MIB_FRAGOKS);
1016 consume_skb(skb);
1017 return err;
1018
1019fail_toobig:
1020 if (skb->sk && dst_allfrag(skb_dst(skb)))
1021 sk_gso_disable(skb->sk);
1022
1023 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1024 err = -EMSGSIZE;
1025
1026fail:
1027 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1028 IPSTATS_MIB_FRAGFAILS);
1029 kfree_skb(skb);
1030 return err;
1031}
1032
1033static inline int ip6_rt_check(const struct rt6key *rt_key,
1034 const struct in6_addr *fl_addr,
1035 const struct in6_addr *addr_cache)
1036{
1037 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1038 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1039}
1040
1041static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1042 struct dst_entry *dst,
1043 const struct flowi6 *fl6)
1044{
1045 struct ipv6_pinfo *np = inet6_sk(sk);
1046 struct rt6_info *rt;
1047
1048 if (!dst)
1049 goto out;
1050
1051 if (dst->ops->family != AF_INET6) {
1052 dst_release(dst);
1053 return NULL;
1054 }
1055
1056 rt = (struct rt6_info *)dst;
1057 /* Yes, checking route validity in not connected
1058 * case is not very simple. Take into account,
1059 * that we do not support routing by source, TOS,
1060 * and MSG_DONTROUTE --ANK (980726)
1061 *
1062 * 1. ip6_rt_check(): If route was host route,
1063 * check that cached destination is current.
1064 * If it is network route, we still may
1065 * check its validity using saved pointer
1066 * to the last used address: daddr_cache.
1067 * We do not want to save whole address now,
1068 * (because main consumer of this service
1069 * is tcp, which has not this problem),
1070 * so that the last trick works only on connected
1071 * sockets.
1072 * 2. oif also should be the same.
1073 */
1074 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1075#ifdef CONFIG_IPV6_SUBTREES
1076 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1077#endif
1078 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1079 dst_release(dst);
1080 dst = NULL;
1081 }
1082
1083out:
1084 return dst;
1085}
1086
1087static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1088 struct dst_entry **dst, struct flowi6 *fl6)
1089{
1090#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1091 struct neighbour *n;
1092 struct rt6_info *rt;
1093#endif
1094 int err;
1095 int flags = 0;
1096
1097 /* The correct way to handle this would be to do
1098 * ip6_route_get_saddr, and then ip6_route_output; however,
1099 * the route-specific preferred source forces the
1100 * ip6_route_output call _before_ ip6_route_get_saddr.
1101 *
1102 * In source specific routing (no src=any default route),
1103 * ip6_route_output will fail given src=any saddr, though, so
1104 * that's why we try it again later.
1105 */
1106 if (ipv6_addr_any(&fl6->saddr)) {
1107 struct fib6_info *from;
1108 struct rt6_info *rt;
1109
1110 *dst = ip6_route_output(net, sk, fl6);
1111 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
1112
1113 rcu_read_lock();
1114 from = rt ? rcu_dereference(rt->from) : NULL;
1115 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1116 sk ? inet6_sk(sk)->srcprefs : 0,
1117 &fl6->saddr);
1118 rcu_read_unlock();
1119
1120 if (err)
1121 goto out_err_release;
1122
1123 /* If we had an erroneous initial result, pretend it
1124 * never existed and let the SA-enabled version take
1125 * over.
1126 */
1127 if ((*dst)->error) {
1128 dst_release(*dst);
1129 *dst = NULL;
1130 }
1131
1132 if (fl6->flowi6_oif)
1133 flags |= RT6_LOOKUP_F_IFACE;
1134 }
1135
1136 if (!*dst)
1137 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1138
1139 err = (*dst)->error;
1140 if (err)
1141 goto out_err_release;
1142
1143#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1144 /*
1145 * Here if the dst entry we've looked up
1146 * has a neighbour entry that is in the INCOMPLETE
1147 * state and the src address from the flow is
1148 * marked as OPTIMISTIC, we release the found
1149 * dst entry and replace it instead with the
1150 * dst entry of the nexthop router
1151 */
1152 rt = (struct rt6_info *) *dst;
1153 rcu_read_lock_bh();
1154 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1155 rt6_nexthop(rt, &fl6->daddr));
1156 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
1157 rcu_read_unlock_bh();
1158
1159 if (err) {
1160 struct inet6_ifaddr *ifp;
1161 struct flowi6 fl_gw6;
1162 int redirect;
1163
1164 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1165 (*dst)->dev, 1);
1166
1167 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1168 if (ifp)
1169 in6_ifa_put(ifp);
1170
1171 if (redirect) {
1172 /*
1173 * We need to get the dst entry for the
1174 * default router instead
1175 */
1176 dst_release(*dst);
1177 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1178 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1179 *dst = ip6_route_output(net, sk, &fl_gw6);
1180 err = (*dst)->error;
1181 if (err)
1182 goto out_err_release;
1183 }
1184 }
1185#endif
1186 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1187 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1188 err = -EAFNOSUPPORT;
1189 goto out_err_release;
1190 }
1191
1192 return 0;
1193
1194out_err_release:
1195 dst_release(*dst);
1196 *dst = NULL;
1197
1198 if (err == -ENETUNREACH)
1199 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1200 return err;
1201}
1202
1203/**
1204 * ip6_dst_lookup - perform route lookup on flow
1205 * @net: Network namespace to perform lookup in
1206 * @sk: socket which provides route info
1207 * @dst: pointer to dst_entry * for result
1208 * @fl6: flow to lookup
1209 *
1210 * This function performs a route lookup on the given flow.
1211 *
1212 * It returns zero on success, or a standard errno code on error.
1213 */
1214int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1215 struct flowi6 *fl6)
1216{
1217 *dst = NULL;
1218 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1219}
1220EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1221
1222/**
1223 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1224 * @net: Network namespace to perform lookup in
1225 * @sk: socket which provides route info
1226 * @fl6: flow to lookup
1227 * @final_dst: final destination address for ipsec lookup
1228 *
1229 * This function performs a route lookup on the given flow.
1230 *
1231 * It returns a valid dst pointer on success, or a pointer encoded
1232 * error code.
1233 */
1234struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1235 const struct in6_addr *final_dst)
1236{
1237 struct dst_entry *dst = NULL;
1238 int err;
1239
1240 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1241 if (err)
1242 return ERR_PTR(err);
1243 if (final_dst)
1244 fl6->daddr = *final_dst;
1245
1246 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1247}
1248EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1249
1250/**
1251 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1252 * @sk: socket which provides the dst cache and route info
1253 * @fl6: flow to lookup
1254 * @final_dst: final destination address for ipsec lookup
1255 * @connected: whether @sk is connected or not
1256 *
1257 * This function performs a route lookup on the given flow with the
1258 * possibility of using the cached route in the socket if it is valid.
1259 * It will take the socket dst lock when operating on the dst cache.
1260 * As a result, this function can only be used in process context.
1261 *
1262 * In addition, for a connected socket, cache the dst in the socket
1263 * if the current cache is not valid.
1264 *
1265 * It returns a valid dst pointer on success, or a pointer encoded
1266 * error code.
1267 */
1268struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1269 const struct in6_addr *final_dst,
1270 bool connected)
1271{
1272 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1273
1274 dst = ip6_sk_dst_check(sk, dst, fl6);
1275 if (dst)
1276 return dst;
1277
1278 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1279 if (connected && !IS_ERR(dst))
1280 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1281
1282 return dst;
1283}
1284EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1285
1286/**
1287 * ip6_dst_lookup_tunnel - perform route lookup on tunnel
1288 * @skb: Packet for which lookup is done
1289 * @dev: Tunnel device
1290 * @net: Network namespace of tunnel device
1291 * @sock: Socket which provides route info
1292 * @saddr: Memory to store the src ip address
1293 * @info: Tunnel information
1294 * @protocol: IP protocol
1295 * @use_cache: Flag to enable cache usage
1296 * This function performs a route lookup on a tunnel
1297 *
1298 * It returns a valid dst pointer and stores src address to be used in
1299 * tunnel in param saddr on success, else a pointer encoded error code.
1300 */
1301
1302struct dst_entry *ip6_dst_lookup_tunnel(struct sk_buff *skb,
1303 struct net_device *dev,
1304 struct net *net,
1305 struct socket *sock,
1306 struct in6_addr *saddr,
1307 const struct ip_tunnel_info *info,
1308 u8 protocol,
1309 bool use_cache)
1310{
1311 struct dst_entry *dst = NULL;
1312#ifdef CONFIG_DST_CACHE
1313 struct dst_cache *dst_cache;
1314#endif
1315 struct flowi6 fl6;
1316 __u8 prio;
1317
1318#ifdef CONFIG_DST_CACHE
1319 dst_cache = (struct dst_cache *)&info->dst_cache;
1320 if (use_cache) {
1321 dst = dst_cache_get_ip6(dst_cache, saddr);
1322 if (dst)
1323 return dst;
1324 }
1325#endif
1326 memset(&fl6, 0, sizeof(fl6));
1327 fl6.flowi6_mark = skb->mark;
1328 fl6.flowi6_proto = protocol;
1329 fl6.daddr = info->key.u.ipv6.dst;
1330 fl6.saddr = info->key.u.ipv6.src;
1331 prio = info->key.tos;
1332 fl6.flowlabel = ip6_make_flowinfo(prio, info->key.label);
1333
1334 dst = ipv6_stub->ipv6_dst_lookup_flow(net, sock->sk, &fl6,
1335 NULL);
1336 if (IS_ERR(dst)) {
1337 netdev_dbg(dev, "no route to %pI6\n", &fl6.daddr);
1338 return ERR_PTR(-ENETUNREACH);
1339 }
1340 if (dst->dev == dev) { /* is this necessary? */
1341 netdev_dbg(dev, "circular route to %pI6\n", &fl6.daddr);
1342 dst_release(dst);
1343 return ERR_PTR(-ELOOP);
1344 }
1345#ifdef CONFIG_DST_CACHE
1346 if (use_cache)
1347 dst_cache_set_ip6(dst_cache, dst, &fl6.saddr);
1348#endif
1349 *saddr = fl6.saddr;
1350 return dst;
1351}
1352EXPORT_SYMBOL_GPL(ip6_dst_lookup_tunnel);
1353
1354static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1355 gfp_t gfp)
1356{
1357 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1358}
1359
1360static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1361 gfp_t gfp)
1362{
1363 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1364}
1365
1366static void ip6_append_data_mtu(unsigned int *mtu,
1367 int *maxfraglen,
1368 unsigned int fragheaderlen,
1369 struct sk_buff *skb,
1370 struct rt6_info *rt,
1371 unsigned int orig_mtu)
1372{
1373 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1374 if (!skb) {
1375 /* first fragment, reserve header_len */
1376 *mtu = orig_mtu - rt->dst.header_len;
1377
1378 } else {
1379 /*
1380 * this fragment is not first, the headers
1381 * space is regarded as data space.
1382 */
1383 *mtu = orig_mtu;
1384 }
1385 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1386 + fragheaderlen - sizeof(struct frag_hdr);
1387 }
1388}
1389
1390static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1391 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1392 struct rt6_info *rt)
1393{
1394 struct ipv6_pinfo *np = inet6_sk(sk);
1395 unsigned int mtu;
1396 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1397
1398 /* callers pass dst together with a reference, set it first so
1399 * ip6_cork_release() can put it down even in case of an error.
1400 */
1401 cork->base.dst = &rt->dst;
1402
1403 /*
1404 * setup for corking
1405 */
1406 if (opt) {
1407 if (WARN_ON(v6_cork->opt))
1408 return -EINVAL;
1409
1410 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1411 if (unlikely(!nopt))
1412 return -ENOBUFS;
1413
1414 nopt->tot_len = sizeof(*opt);
1415 nopt->opt_flen = opt->opt_flen;
1416 nopt->opt_nflen = opt->opt_nflen;
1417
1418 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1419 if (opt->dst0opt && !nopt->dst0opt)
1420 return -ENOBUFS;
1421
1422 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1423 if (opt->dst1opt && !nopt->dst1opt)
1424 return -ENOBUFS;
1425
1426 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1427 if (opt->hopopt && !nopt->hopopt)
1428 return -ENOBUFS;
1429
1430 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1431 if (opt->srcrt && !nopt->srcrt)
1432 return -ENOBUFS;
1433
1434 /* need source address above miyazawa*/
1435 }
1436 v6_cork->hop_limit = ipc6->hlimit;
1437 v6_cork->tclass = ipc6->tclass;
1438 if (rt->dst.flags & DST_XFRM_TUNNEL)
1439 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1440 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1441 else
1442 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1443 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1444 if (np->frag_size < mtu) {
1445 if (np->frag_size)
1446 mtu = np->frag_size;
1447 }
1448 cork->base.fragsize = mtu;
1449 cork->base.gso_size = ipc6->gso_size;
1450 cork->base.tx_flags = 0;
1451 cork->base.mark = ipc6->sockc.mark;
1452 sock_tx_timestamp(sk, ipc6->sockc.tsflags, &cork->base.tx_flags);
1453
1454 if (dst_allfrag(xfrm_dst_path(&rt->dst)))
1455 cork->base.flags |= IPCORK_ALLFRAG;
1456 cork->base.length = 0;
1457
1458 cork->base.transmit_time = ipc6->sockc.transmit_time;
1459
1460 return 0;
1461}
1462
1463static int __ip6_append_data(struct sock *sk,
1464 struct sk_buff_head *queue,
1465 struct inet_cork_full *cork_full,
1466 struct inet6_cork *v6_cork,
1467 struct page_frag *pfrag,
1468 int getfrag(void *from, char *to, int offset,
1469 int len, int odd, struct sk_buff *skb),
1470 void *from, size_t length, int transhdrlen,
1471 unsigned int flags, struct ipcm6_cookie *ipc6)
1472{
1473 struct sk_buff *skb, *skb_prev = NULL;
1474 struct inet_cork *cork = &cork_full->base;
1475 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1476 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1477 struct ubuf_info *uarg = NULL;
1478 int exthdrlen = 0;
1479 int dst_exthdrlen = 0;
1480 int hh_len;
1481 int copy;
1482 int err;
1483 int offset = 0;
1484 bool zc = false;
1485 u32 tskey = 0;
1486 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1487 struct ipv6_txoptions *opt = v6_cork->opt;
1488 int csummode = CHECKSUM_NONE;
1489 unsigned int maxnonfragsize, headersize;
1490 unsigned int wmem_alloc_delta = 0;
1491 bool paged, extra_uref = false;
1492
1493 skb = skb_peek_tail(queue);
1494 if (!skb) {
1495 exthdrlen = opt ? opt->opt_flen : 0;
1496 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1497 }
1498
1499 paged = !!cork->gso_size;
1500 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1501 orig_mtu = mtu;
1502
1503 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1504 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1505 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1506
1507 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1508
1509 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1510 (opt ? opt->opt_nflen : 0);
1511
1512 headersize = sizeof(struct ipv6hdr) +
1513 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1514 (dst_allfrag(&rt->dst) ?
1515 sizeof(struct frag_hdr) : 0) +
1516 rt->rt6i_nfheader_len;
1517
1518 if (mtu <= fragheaderlen ||
1519 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1520 goto emsgsize;
1521
1522 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1523 sizeof(struct frag_hdr);
1524
1525 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1526 * the first fragment
1527 */
1528 if (headersize + transhdrlen > mtu)
1529 goto emsgsize;
1530
1531 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1532 (sk->sk_protocol == IPPROTO_UDP ||
1533 sk->sk_protocol == IPPROTO_ICMPV6 ||
1534 sk->sk_protocol == IPPROTO_RAW)) {
1535 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1536 sizeof(struct ipv6hdr));
1537 goto emsgsize;
1538 }
1539
1540 if (ip6_sk_ignore_df(sk))
1541 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1542 else
1543 maxnonfragsize = mtu;
1544
1545 if (cork->length + length > maxnonfragsize - headersize) {
1546emsgsize:
1547 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1548 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1549 return -EMSGSIZE;
1550 }
1551
1552 /* CHECKSUM_PARTIAL only with no extension headers and when
1553 * we are not going to fragment
1554 */
1555 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1556 headersize == sizeof(struct ipv6hdr) &&
1557 length <= mtu - headersize &&
1558 (!(flags & MSG_MORE) || cork->gso_size) &&
1559 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1560 csummode = CHECKSUM_PARTIAL;
1561
1562 if ((flags & MSG_ZEROCOPY) && length) {
1563 struct msghdr *msg = from;
1564
1565 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1566 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1567 return -EINVAL;
1568
1569 /* Leave uarg NULL if can't zerocopy, callers should
1570 * be able to handle it.
1571 */
1572 if ((rt->dst.dev->features & NETIF_F_SG) &&
1573 csummode == CHECKSUM_PARTIAL) {
1574 paged = true;
1575 zc = true;
1576 uarg = msg->msg_ubuf;
1577 }
1578 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1579 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1580 if (!uarg)
1581 return -ENOBUFS;
1582 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1583 if (rt->dst.dev->features & NETIF_F_SG &&
1584 csummode == CHECKSUM_PARTIAL) {
1585 paged = true;
1586 zc = true;
1587 } else {
1588 uarg_to_msgzc(uarg)->zerocopy = 0;
1589 skb_zcopy_set(skb, uarg, &extra_uref);
1590 }
1591 }
1592 }
1593
1594 /*
1595 * Let's try using as much space as possible.
1596 * Use MTU if total length of the message fits into the MTU.
1597 * Otherwise, we need to reserve fragment header and
1598 * fragment alignment (= 8-15 octects, in total).
1599 *
1600 * Note that we may need to "move" the data from the tail
1601 * of the buffer to the new fragment when we split
1602 * the message.
1603 *
1604 * FIXME: It may be fragmented into multiple chunks
1605 * at once if non-fragmentable extension headers
1606 * are too large.
1607 * --yoshfuji
1608 */
1609
1610 cork->length += length;
1611 if (!skb)
1612 goto alloc_new_skb;
1613
1614 while (length > 0) {
1615 /* Check if the remaining data fits into current packet. */
1616 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1617 if (copy < length)
1618 copy = maxfraglen - skb->len;
1619
1620 if (copy <= 0) {
1621 char *data;
1622 unsigned int datalen;
1623 unsigned int fraglen;
1624 unsigned int fraggap;
1625 unsigned int alloclen, alloc_extra;
1626 unsigned int pagedlen;
1627alloc_new_skb:
1628 /* There's no room in the current skb */
1629 if (skb)
1630 fraggap = skb->len - maxfraglen;
1631 else
1632 fraggap = 0;
1633 /* update mtu and maxfraglen if necessary */
1634 if (!skb || !skb_prev)
1635 ip6_append_data_mtu(&mtu, &maxfraglen,
1636 fragheaderlen, skb, rt,
1637 orig_mtu);
1638
1639 skb_prev = skb;
1640
1641 /*
1642 * If remaining data exceeds the mtu,
1643 * we know we need more fragment(s).
1644 */
1645 datalen = length + fraggap;
1646
1647 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1648 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1649 fraglen = datalen + fragheaderlen;
1650 pagedlen = 0;
1651
1652 alloc_extra = hh_len;
1653 alloc_extra += dst_exthdrlen;
1654 alloc_extra += rt->dst.trailer_len;
1655
1656 /* We just reserve space for fragment header.
1657 * Note: this may be overallocation if the message
1658 * (without MSG_MORE) fits into the MTU.
1659 */
1660 alloc_extra += sizeof(struct frag_hdr);
1661
1662 if ((flags & MSG_MORE) &&
1663 !(rt->dst.dev->features&NETIF_F_SG))
1664 alloclen = mtu;
1665 else if (!paged &&
1666 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1667 !(rt->dst.dev->features & NETIF_F_SG)))
1668 alloclen = fraglen;
1669 else {
1670 alloclen = fragheaderlen + transhdrlen;
1671 pagedlen = datalen - transhdrlen;
1672 }
1673 alloclen += alloc_extra;
1674
1675 if (datalen != length + fraggap) {
1676 /*
1677 * this is not the last fragment, the trailer
1678 * space is regarded as data space.
1679 */
1680 datalen += rt->dst.trailer_len;
1681 }
1682
1683 fraglen = datalen + fragheaderlen;
1684
1685 copy = datalen - transhdrlen - fraggap - pagedlen;
1686 if (copy < 0) {
1687 err = -EINVAL;
1688 goto error;
1689 }
1690 if (transhdrlen) {
1691 skb = sock_alloc_send_skb(sk, alloclen,
1692 (flags & MSG_DONTWAIT), &err);
1693 } else {
1694 skb = NULL;
1695 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1696 2 * sk->sk_sndbuf)
1697 skb = alloc_skb(alloclen,
1698 sk->sk_allocation);
1699 if (unlikely(!skb))
1700 err = -ENOBUFS;
1701 }
1702 if (!skb)
1703 goto error;
1704 /*
1705 * Fill in the control structures
1706 */
1707 skb->protocol = htons(ETH_P_IPV6);
1708 skb->ip_summed = csummode;
1709 skb->csum = 0;
1710 /* reserve for fragmentation and ipsec header */
1711 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1712 dst_exthdrlen);
1713
1714 /*
1715 * Find where to start putting bytes
1716 */
1717 data = skb_put(skb, fraglen - pagedlen);
1718 skb_set_network_header(skb, exthdrlen);
1719 data += fragheaderlen;
1720 skb->transport_header = (skb->network_header +
1721 fragheaderlen);
1722 if (fraggap) {
1723 skb->csum = skb_copy_and_csum_bits(
1724 skb_prev, maxfraglen,
1725 data + transhdrlen, fraggap);
1726 skb_prev->csum = csum_sub(skb_prev->csum,
1727 skb->csum);
1728 data += fraggap;
1729 pskb_trim_unique(skb_prev, maxfraglen);
1730 }
1731 if (copy > 0 &&
1732 getfrag(from, data + transhdrlen, offset,
1733 copy, fraggap, skb) < 0) {
1734 err = -EFAULT;
1735 kfree_skb(skb);
1736 goto error;
1737 }
1738
1739 offset += copy;
1740 length -= copy + transhdrlen;
1741 transhdrlen = 0;
1742 exthdrlen = 0;
1743 dst_exthdrlen = 0;
1744
1745 /* Only the initial fragment is time stamped */
1746 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1747 cork->tx_flags = 0;
1748 skb_shinfo(skb)->tskey = tskey;
1749 tskey = 0;
1750 skb_zcopy_set(skb, uarg, &extra_uref);
1751
1752 if ((flags & MSG_CONFIRM) && !skb_prev)
1753 skb_set_dst_pending_confirm(skb, 1);
1754
1755 /*
1756 * Put the packet on the pending queue
1757 */
1758 if (!skb->destructor) {
1759 skb->destructor = sock_wfree;
1760 skb->sk = sk;
1761 wmem_alloc_delta += skb->truesize;
1762 }
1763 __skb_queue_tail(queue, skb);
1764 continue;
1765 }
1766
1767 if (copy > length)
1768 copy = length;
1769
1770 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1771 skb_tailroom(skb) >= copy) {
1772 unsigned int off;
1773
1774 off = skb->len;
1775 if (getfrag(from, skb_put(skb, copy),
1776 offset, copy, off, skb) < 0) {
1777 __skb_trim(skb, off);
1778 err = -EFAULT;
1779 goto error;
1780 }
1781 } else if (!zc) {
1782 int i = skb_shinfo(skb)->nr_frags;
1783
1784 err = -ENOMEM;
1785 if (!sk_page_frag_refill(sk, pfrag))
1786 goto error;
1787
1788 skb_zcopy_downgrade_managed(skb);
1789 if (!skb_can_coalesce(skb, i, pfrag->page,
1790 pfrag->offset)) {
1791 err = -EMSGSIZE;
1792 if (i == MAX_SKB_FRAGS)
1793 goto error;
1794
1795 __skb_fill_page_desc(skb, i, pfrag->page,
1796 pfrag->offset, 0);
1797 skb_shinfo(skb)->nr_frags = ++i;
1798 get_page(pfrag->page);
1799 }
1800 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1801 if (getfrag(from,
1802 page_address(pfrag->page) + pfrag->offset,
1803 offset, copy, skb->len, skb) < 0)
1804 goto error_efault;
1805
1806 pfrag->offset += copy;
1807 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1808 skb->len += copy;
1809 skb->data_len += copy;
1810 skb->truesize += copy;
1811 wmem_alloc_delta += copy;
1812 } else {
1813 err = skb_zerocopy_iter_dgram(skb, from, copy);
1814 if (err < 0)
1815 goto error;
1816 }
1817 offset += copy;
1818 length -= copy;
1819 }
1820
1821 if (wmem_alloc_delta)
1822 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1823 return 0;
1824
1825error_efault:
1826 err = -EFAULT;
1827error:
1828 net_zcopy_put_abort(uarg, extra_uref);
1829 cork->length -= length;
1830 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1831 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1832 return err;
1833}
1834
1835int ip6_append_data(struct sock *sk,
1836 int getfrag(void *from, char *to, int offset, int len,
1837 int odd, struct sk_buff *skb),
1838 void *from, size_t length, int transhdrlen,
1839 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1840 struct rt6_info *rt, unsigned int flags)
1841{
1842 struct inet_sock *inet = inet_sk(sk);
1843 struct ipv6_pinfo *np = inet6_sk(sk);
1844 int exthdrlen;
1845 int err;
1846
1847 if (flags&MSG_PROBE)
1848 return 0;
1849 if (skb_queue_empty(&sk->sk_write_queue)) {
1850 /*
1851 * setup for corking
1852 */
1853 dst_hold(&rt->dst);
1854 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1855 ipc6, rt);
1856 if (err)
1857 return err;
1858
1859 inet->cork.fl.u.ip6 = *fl6;
1860 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1861 length += exthdrlen;
1862 transhdrlen += exthdrlen;
1863 } else {
1864 transhdrlen = 0;
1865 }
1866
1867 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1868 &np->cork, sk_page_frag(sk), getfrag,
1869 from, length, transhdrlen, flags, ipc6);
1870}
1871EXPORT_SYMBOL_GPL(ip6_append_data);
1872
1873static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1874{
1875 struct dst_entry *dst = cork->base.dst;
1876
1877 cork->base.dst = NULL;
1878 cork->base.flags &= ~IPCORK_ALLFRAG;
1879 skb_dst_set(skb, dst);
1880}
1881
1882static void ip6_cork_release(struct inet_cork_full *cork,
1883 struct inet6_cork *v6_cork)
1884{
1885 if (v6_cork->opt) {
1886 struct ipv6_txoptions *opt = v6_cork->opt;
1887
1888 kfree(opt->dst0opt);
1889 kfree(opt->dst1opt);
1890 kfree(opt->hopopt);
1891 kfree(opt->srcrt);
1892 kfree(opt);
1893 v6_cork->opt = NULL;
1894 }
1895
1896 if (cork->base.dst) {
1897 dst_release(cork->base.dst);
1898 cork->base.dst = NULL;
1899 cork->base.flags &= ~IPCORK_ALLFRAG;
1900 }
1901}
1902
1903struct sk_buff *__ip6_make_skb(struct sock *sk,
1904 struct sk_buff_head *queue,
1905 struct inet_cork_full *cork,
1906 struct inet6_cork *v6_cork)
1907{
1908 struct sk_buff *skb, *tmp_skb;
1909 struct sk_buff **tail_skb;
1910 struct in6_addr *final_dst;
1911 struct ipv6_pinfo *np = inet6_sk(sk);
1912 struct net *net = sock_net(sk);
1913 struct ipv6hdr *hdr;
1914 struct ipv6_txoptions *opt = v6_cork->opt;
1915 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1916 struct flowi6 *fl6 = &cork->fl.u.ip6;
1917 unsigned char proto = fl6->flowi6_proto;
1918
1919 skb = __skb_dequeue(queue);
1920 if (!skb)
1921 goto out;
1922 tail_skb = &(skb_shinfo(skb)->frag_list);
1923
1924 /* move skb->data to ip header from ext header */
1925 if (skb->data < skb_network_header(skb))
1926 __skb_pull(skb, skb_network_offset(skb));
1927 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1928 __skb_pull(tmp_skb, skb_network_header_len(skb));
1929 *tail_skb = tmp_skb;
1930 tail_skb = &(tmp_skb->next);
1931 skb->len += tmp_skb->len;
1932 skb->data_len += tmp_skb->len;
1933 skb->truesize += tmp_skb->truesize;
1934 tmp_skb->destructor = NULL;
1935 tmp_skb->sk = NULL;
1936 }
1937
1938 /* Allow local fragmentation. */
1939 skb->ignore_df = ip6_sk_ignore_df(sk);
1940 __skb_pull(skb, skb_network_header_len(skb));
1941
1942 final_dst = &fl6->daddr;
1943 if (opt && opt->opt_flen)
1944 ipv6_push_frag_opts(skb, opt, &proto);
1945 if (opt && opt->opt_nflen)
1946 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1947
1948 skb_push(skb, sizeof(struct ipv6hdr));
1949 skb_reset_network_header(skb);
1950 hdr = ipv6_hdr(skb);
1951
1952 ip6_flow_hdr(hdr, v6_cork->tclass,
1953 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1954 ip6_autoflowlabel(net, np), fl6));
1955 hdr->hop_limit = v6_cork->hop_limit;
1956 hdr->nexthdr = proto;
1957 hdr->saddr = fl6->saddr;
1958 hdr->daddr = *final_dst;
1959
1960 skb->priority = sk->sk_priority;
1961 skb->mark = cork->base.mark;
1962 skb->tstamp = cork->base.transmit_time;
1963
1964 ip6_cork_steal_dst(skb, cork);
1965 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1966 if (proto == IPPROTO_ICMPV6) {
1967 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1968
1969 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1970 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1971 }
1972
1973 ip6_cork_release(cork, v6_cork);
1974out:
1975 return skb;
1976}
1977
1978int ip6_send_skb(struct sk_buff *skb)
1979{
1980 struct net *net = sock_net(skb->sk);
1981 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1982 int err;
1983
1984 err = ip6_local_out(net, skb->sk, skb);
1985 if (err) {
1986 if (err > 0)
1987 err = net_xmit_errno(err);
1988 if (err)
1989 IP6_INC_STATS(net, rt->rt6i_idev,
1990 IPSTATS_MIB_OUTDISCARDS);
1991 }
1992
1993 return err;
1994}
1995
1996int ip6_push_pending_frames(struct sock *sk)
1997{
1998 struct sk_buff *skb;
1999
2000 skb = ip6_finish_skb(sk);
2001 if (!skb)
2002 return 0;
2003
2004 return ip6_send_skb(skb);
2005}
2006EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2007
2008static void __ip6_flush_pending_frames(struct sock *sk,
2009 struct sk_buff_head *queue,
2010 struct inet_cork_full *cork,
2011 struct inet6_cork *v6_cork)
2012{
2013 struct sk_buff *skb;
2014
2015 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2016 if (skb_dst(skb))
2017 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2018 IPSTATS_MIB_OUTDISCARDS);
2019 kfree_skb(skb);
2020 }
2021
2022 ip6_cork_release(cork, v6_cork);
2023}
2024
2025void ip6_flush_pending_frames(struct sock *sk)
2026{
2027 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2028 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2029}
2030EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2031
2032struct sk_buff *ip6_make_skb(struct sock *sk,
2033 int getfrag(void *from, char *to, int offset,
2034 int len, int odd, struct sk_buff *skb),
2035 void *from, size_t length, int transhdrlen,
2036 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2037 unsigned int flags, struct inet_cork_full *cork)
2038{
2039 struct inet6_cork v6_cork;
2040 struct sk_buff_head queue;
2041 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2042 int err;
2043
2044 if (flags & MSG_PROBE) {
2045 dst_release(&rt->dst);
2046 return NULL;
2047 }
2048
2049 __skb_queue_head_init(&queue);
2050
2051 cork->base.flags = 0;
2052 cork->base.addr = 0;
2053 cork->base.opt = NULL;
2054 v6_cork.opt = NULL;
2055 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2056 if (err) {
2057 ip6_cork_release(cork, &v6_cork);
2058 return ERR_PTR(err);
2059 }
2060 if (ipc6->dontfrag < 0)
2061 ipc6->dontfrag = inet6_sk(sk)->dontfrag;
2062
2063 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2064 ¤t->task_frag, getfrag, from,
2065 length + exthdrlen, transhdrlen + exthdrlen,
2066 flags, ipc6);
2067 if (err) {
2068 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2069 return ERR_PTR(err);
2070 }
2071
2072 return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2073}