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
59static int ip6_finish_output2(struct sk_buff *skb)
60{
61 struct dst_entry *dst = skb_dst(skb);
62 struct net_device *dev = dst->dev;
63 struct neighbour *neigh;
64 struct in6_addr *nexthop;
65 int ret;
66
67 skb->protocol = htons(ETH_P_IPV6);
68 skb->dev = dev;
69
70 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
71 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
72
73 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(skb->sk) &&
74 ((mroute6_socket(dev_net(dev), skb) &&
75 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
76 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
77 &ipv6_hdr(skb)->saddr))) {
78 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
79
80 /* Do not check for IFF_ALLMULTI; multicast routing
81 is not supported in any case.
82 */
83 if (newskb)
84 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
85 newskb, NULL, newskb->dev,
86 dev_loopback_xmit);
87
88 if (ipv6_hdr(skb)->hop_limit == 0) {
89 IP6_INC_STATS(dev_net(dev), idev,
90 IPSTATS_MIB_OUTDISCARDS);
91 kfree_skb(skb);
92 return 0;
93 }
94 }
95
96 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
97 skb->len);
98
99 if (IPV6_ADDR_MC_SCOPE(&ipv6_hdr(skb)->daddr) <=
100 IPV6_ADDR_SCOPE_NODELOCAL &&
101 !(dev->flags & IFF_LOOPBACK)) {
102 kfree_skb(skb);
103 return 0;
104 }
105 }
106
107 rcu_read_lock_bh();
108 nexthop = rt6_nexthop((struct rt6_info *)dst);
109 neigh = __ipv6_neigh_lookup_noref(dst->dev, nexthop);
110 if (unlikely(!neigh))
111 neigh = __neigh_create(&nd_tbl, nexthop, dst->dev, false);
112 if (!IS_ERR(neigh)) {
113 ret = dst_neigh_output(dst, neigh, skb);
114 rcu_read_unlock_bh();
115 return ret;
116 }
117 rcu_read_unlock_bh();
118
119 IP6_INC_STATS(dev_net(dst->dev),
120 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
121 kfree_skb(skb);
122 return -EINVAL;
123}
124
125static int ip6_finish_output(struct sk_buff *skb)
126{
127 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
128 dst_allfrag(skb_dst(skb)) ||
129 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
130 return ip6_fragment(skb, ip6_finish_output2);
131 else
132 return ip6_finish_output2(skb);
133}
134
135int ip6_output(struct sock *sk, struct sk_buff *skb)
136{
137 struct net_device *dev = skb_dst(skb)->dev;
138 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(dev_net(dev), idev,
141 IPSTATS_MIB_OUTDISCARDS);
142 kfree_skb(skb);
143 return 0;
144 }
145
146 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
147 ip6_finish_output,
148 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
149}
150
151/*
152 * xmit an sk_buff (used by TCP, SCTP and DCCP)
153 */
154
155int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
156 struct ipv6_txoptions *opt, int tclass)
157{
158 struct net *net = sock_net(sk);
159 struct ipv6_pinfo *np = inet6_sk(sk);
160 struct in6_addr *first_hop = &fl6->daddr;
161 struct dst_entry *dst = skb_dst(skb);
162 struct ipv6hdr *hdr;
163 u8 proto = fl6->flowi6_proto;
164 int seg_len = skb->len;
165 int hlimit = -1;
166 u32 mtu;
167
168 if (opt) {
169 unsigned int head_room;
170
171 /* First: exthdrs may take lots of space (~8K for now)
172 MAX_HEADER is not enough.
173 */
174 head_room = opt->opt_nflen + opt->opt_flen;
175 seg_len += head_room;
176 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
177
178 if (skb_headroom(skb) < head_room) {
179 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
180 if (skb2 == NULL) {
181 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
182 IPSTATS_MIB_OUTDISCARDS);
183 kfree_skb(skb);
184 return -ENOBUFS;
185 }
186 consume_skb(skb);
187 skb = skb2;
188 skb_set_owner_w(skb, sk);
189 }
190 if (opt->opt_flen)
191 ipv6_push_frag_opts(skb, opt, &proto);
192 if (opt->opt_nflen)
193 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
194 }
195
196 skb_push(skb, sizeof(struct ipv6hdr));
197 skb_reset_network_header(skb);
198 hdr = ipv6_hdr(skb);
199
200 /*
201 * Fill in the IPv6 header
202 */
203 if (np)
204 hlimit = np->hop_limit;
205 if (hlimit < 0)
206 hlimit = ip6_dst_hoplimit(dst);
207
208 ip6_flow_hdr(hdr, tclass, fl6->flowlabel);
209
210 hdr->payload_len = htons(seg_len);
211 hdr->nexthdr = proto;
212 hdr->hop_limit = hlimit;
213
214 hdr->saddr = fl6->saddr;
215 hdr->daddr = *first_hop;
216
217 skb->protocol = htons(ETH_P_IPV6);
218 skb->priority = sk->sk_priority;
219 skb->mark = sk->sk_mark;
220
221 mtu = dst_mtu(dst);
222 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
223 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
224 IPSTATS_MIB_OUT, skb->len);
225 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT, skb, NULL,
226 dst->dev, dst_output);
227 }
228
229 skb->dev = dst->dev;
230 ipv6_local_error(sk, EMSGSIZE, fl6, mtu);
231 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
232 kfree_skb(skb);
233 return -EMSGSIZE;
234}
235
236EXPORT_SYMBOL(ip6_xmit);
237
238static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
239{
240 struct ip6_ra_chain *ra;
241 struct sock *last = NULL;
242
243 read_lock(&ip6_ra_lock);
244 for (ra = ip6_ra_chain; ra; ra = ra->next) {
245 struct sock *sk = ra->sk;
246 if (sk && ra->sel == sel &&
247 (!sk->sk_bound_dev_if ||
248 sk->sk_bound_dev_if == skb->dev->ifindex)) {
249 if (last) {
250 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
251 if (skb2)
252 rawv6_rcv(last, skb2);
253 }
254 last = sk;
255 }
256 }
257
258 if (last) {
259 rawv6_rcv(last, skb);
260 read_unlock(&ip6_ra_lock);
261 return 1;
262 }
263 read_unlock(&ip6_ra_lock);
264 return 0;
265}
266
267static int ip6_forward_proxy_check(struct sk_buff *skb)
268{
269 struct ipv6hdr *hdr = ipv6_hdr(skb);
270 u8 nexthdr = hdr->nexthdr;
271 __be16 frag_off;
272 int offset;
273
274 if (ipv6_ext_hdr(nexthdr)) {
275 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
276 if (offset < 0)
277 return 0;
278 } else
279 offset = sizeof(struct ipv6hdr);
280
281 if (nexthdr == IPPROTO_ICMPV6) {
282 struct icmp6hdr *icmp6;
283
284 if (!pskb_may_pull(skb, (skb_network_header(skb) +
285 offset + 1 - skb->data)))
286 return 0;
287
288 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
289
290 switch (icmp6->icmp6_type) {
291 case NDISC_ROUTER_SOLICITATION:
292 case NDISC_ROUTER_ADVERTISEMENT:
293 case NDISC_NEIGHBOUR_SOLICITATION:
294 case NDISC_NEIGHBOUR_ADVERTISEMENT:
295 case NDISC_REDIRECT:
296 /* For reaction involving unicast neighbor discovery
297 * message destined to the proxied address, pass it to
298 * input function.
299 */
300 return 1;
301 default:
302 break;
303 }
304 }
305
306 /*
307 * The proxying router can't forward traffic sent to a link-local
308 * address, so signal the sender and discard the packet. This
309 * behavior is clarified by the MIPv6 specification.
310 */
311 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
312 dst_link_failure(skb);
313 return -1;
314 }
315
316 return 0;
317}
318
319static inline int ip6_forward_finish(struct sk_buff *skb)
320{
321 return dst_output(skb);
322}
323
324static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
325{
326 unsigned int mtu;
327 struct inet6_dev *idev;
328
329 if (dst_metric_locked(dst, RTAX_MTU)) {
330 mtu = dst_metric_raw(dst, RTAX_MTU);
331 if (mtu)
332 return mtu;
333 }
334
335 mtu = IPV6_MIN_MTU;
336 rcu_read_lock();
337 idev = __in6_dev_get(dst->dev);
338 if (idev)
339 mtu = idev->cnf.mtu6;
340 rcu_read_unlock();
341
342 return mtu;
343}
344
345static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
346{
347 if (skb->len <= mtu)
348 return false;
349
350 /* ipv6 conntrack defrag sets max_frag_size + local_df */
351 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
352 return true;
353
354 if (skb->local_df)
355 return false;
356
357 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
358 return false;
359
360 return true;
361}
362
363int ip6_forward(struct sk_buff *skb)
364{
365 struct dst_entry *dst = skb_dst(skb);
366 struct ipv6hdr *hdr = ipv6_hdr(skb);
367 struct inet6_skb_parm *opt = IP6CB(skb);
368 struct net *net = dev_net(dst->dev);
369 u32 mtu;
370
371 if (net->ipv6.devconf_all->forwarding == 0)
372 goto error;
373
374 if (skb->pkt_type != PACKET_HOST)
375 goto drop;
376
377 if (skb_warn_if_lro(skb))
378 goto drop;
379
380 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
381 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
382 IPSTATS_MIB_INDISCARDS);
383 goto drop;
384 }
385
386 skb_forward_csum(skb);
387
388 /*
389 * We DO NOT make any processing on
390 * RA packets, pushing them to user level AS IS
391 * without ane WARRANTY that application will be able
392 * to interpret them. The reason is that we
393 * cannot make anything clever here.
394 *
395 * We are not end-node, so that if packet contains
396 * AH/ESP, we cannot make anything.
397 * Defragmentation also would be mistake, RA packets
398 * cannot be fragmented, because there is no warranty
399 * that different fragments will go along one path. --ANK
400 */
401 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
402 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
403 return 0;
404 }
405
406 /*
407 * check and decrement ttl
408 */
409 if (hdr->hop_limit <= 1) {
410 /* Force OUTPUT device used as source address */
411 skb->dev = dst->dev;
412 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
413 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
414 IPSTATS_MIB_INHDRERRORS);
415
416 kfree_skb(skb);
417 return -ETIMEDOUT;
418 }
419
420 /* XXX: idev->cnf.proxy_ndp? */
421 if (net->ipv6.devconf_all->proxy_ndp &&
422 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
423 int proxied = ip6_forward_proxy_check(skb);
424 if (proxied > 0)
425 return ip6_input(skb);
426 else if (proxied < 0) {
427 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
428 IPSTATS_MIB_INDISCARDS);
429 goto drop;
430 }
431 }
432
433 if (!xfrm6_route_forward(skb)) {
434 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
435 IPSTATS_MIB_INDISCARDS);
436 goto drop;
437 }
438 dst = skb_dst(skb);
439
440 /* IPv6 specs say nothing about it, but it is clear that we cannot
441 send redirects to source routed frames.
442 We don't send redirects to frames decapsulated from IPsec.
443 */
444 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
445 struct in6_addr *target = NULL;
446 struct inet_peer *peer;
447 struct rt6_info *rt;
448
449 /*
450 * incoming and outgoing devices are the same
451 * send a redirect.
452 */
453
454 rt = (struct rt6_info *) dst;
455 if (rt->rt6i_flags & RTF_GATEWAY)
456 target = &rt->rt6i_gateway;
457 else
458 target = &hdr->daddr;
459
460 peer = inet_getpeer_v6(net->ipv6.peers, &rt->rt6i_dst.addr, 1);
461
462 /* Limit redirects both by destination (here)
463 and by source (inside ndisc_send_redirect)
464 */
465 if (inet_peer_xrlim_allow(peer, 1*HZ))
466 ndisc_send_redirect(skb, target);
467 if (peer)
468 inet_putpeer(peer);
469 } else {
470 int addrtype = ipv6_addr_type(&hdr->saddr);
471
472 /* This check is security critical. */
473 if (addrtype == IPV6_ADDR_ANY ||
474 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
475 goto error;
476 if (addrtype & IPV6_ADDR_LINKLOCAL) {
477 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
478 ICMPV6_NOT_NEIGHBOUR, 0);
479 goto error;
480 }
481 }
482
483 mtu = ip6_dst_mtu_forward(dst);
484 if (mtu < IPV6_MIN_MTU)
485 mtu = IPV6_MIN_MTU;
486
487 if (ip6_pkt_too_big(skb, mtu)) {
488 /* Again, force OUTPUT device used as source address */
489 skb->dev = dst->dev;
490 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
491 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
492 IPSTATS_MIB_INTOOBIGERRORS);
493 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
494 IPSTATS_MIB_FRAGFAILS);
495 kfree_skb(skb);
496 return -EMSGSIZE;
497 }
498
499 if (skb_cow(skb, dst->dev->hard_header_len)) {
500 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
501 IPSTATS_MIB_OUTDISCARDS);
502 goto drop;
503 }
504
505 hdr = ipv6_hdr(skb);
506
507 /* Mangling hops number delayed to point after skb COW */
508
509 hdr->hop_limit--;
510
511 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
512 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
513 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
514 ip6_forward_finish);
515
516error:
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
518drop:
519 kfree_skb(skb);
520 return -EINVAL;
521}
522
523static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
524{
525 to->pkt_type = from->pkt_type;
526 to->priority = from->priority;
527 to->protocol = from->protocol;
528 skb_dst_drop(to);
529 skb_dst_set(to, dst_clone(skb_dst(from)));
530 to->dev = from->dev;
531 to->mark = from->mark;
532
533#ifdef CONFIG_NET_SCHED
534 to->tc_index = from->tc_index;
535#endif
536 nf_copy(to, from);
537 skb_copy_secmark(to, from);
538}
539
540int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
541{
542 struct sk_buff *frag;
543 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
544 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
545 struct ipv6hdr *tmp_hdr;
546 struct frag_hdr *fh;
547 unsigned int mtu, hlen, left, len;
548 int hroom, troom;
549 __be32 frag_id = 0;
550 int ptr, offset = 0, err=0;
551 u8 *prevhdr, nexthdr = 0;
552 struct net *net = dev_net(skb_dst(skb)->dev);
553
554 hlen = ip6_find_1stfragopt(skb, &prevhdr);
555 nexthdr = *prevhdr;
556
557 mtu = ip6_skb_dst_mtu(skb);
558
559 /* We must not fragment if the socket is set to force MTU discovery
560 * or if the skb it not generated by a local socket.
561 */
562 if (unlikely(!skb->local_df && skb->len > mtu) ||
563 (IP6CB(skb)->frag_max_size &&
564 IP6CB(skb)->frag_max_size > mtu)) {
565 if (skb->sk && dst_allfrag(skb_dst(skb)))
566 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
567
568 skb->dev = skb_dst(skb)->dev;
569 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
570 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
571 IPSTATS_MIB_FRAGFAILS);
572 kfree_skb(skb);
573 return -EMSGSIZE;
574 }
575
576 if (np && np->frag_size < mtu) {
577 if (np->frag_size)
578 mtu = np->frag_size;
579 }
580 mtu -= hlen + sizeof(struct frag_hdr);
581
582 if (skb_has_frag_list(skb)) {
583 int first_len = skb_pagelen(skb);
584 struct sk_buff *frag2;
585
586 if (first_len - hlen > mtu ||
587 ((first_len - hlen) & 7) ||
588 skb_cloned(skb))
589 goto slow_path;
590
591 skb_walk_frags(skb, frag) {
592 /* Correct geometry. */
593 if (frag->len > mtu ||
594 ((frag->len & 7) && frag->next) ||
595 skb_headroom(frag) < hlen)
596 goto slow_path_clean;
597
598 /* Partially cloned skb? */
599 if (skb_shared(frag))
600 goto slow_path_clean;
601
602 BUG_ON(frag->sk);
603 if (skb->sk) {
604 frag->sk = skb->sk;
605 frag->destructor = sock_wfree;
606 }
607 skb->truesize -= frag->truesize;
608 }
609
610 err = 0;
611 offset = 0;
612 frag = skb_shinfo(skb)->frag_list;
613 skb_frag_list_init(skb);
614 /* BUILD HEADER */
615
616 *prevhdr = NEXTHDR_FRAGMENT;
617 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
618 if (!tmp_hdr) {
619 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
620 IPSTATS_MIB_FRAGFAILS);
621 return -ENOMEM;
622 }
623
624 __skb_pull(skb, hlen);
625 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
626 __skb_push(skb, hlen);
627 skb_reset_network_header(skb);
628 memcpy(skb_network_header(skb), tmp_hdr, hlen);
629
630 ipv6_select_ident(fh, rt);
631 fh->nexthdr = nexthdr;
632 fh->reserved = 0;
633 fh->frag_off = htons(IP6_MF);
634 frag_id = fh->identification;
635
636 first_len = skb_pagelen(skb);
637 skb->data_len = first_len - skb_headlen(skb);
638 skb->len = first_len;
639 ipv6_hdr(skb)->payload_len = htons(first_len -
640 sizeof(struct ipv6hdr));
641
642 dst_hold(&rt->dst);
643
644 for (;;) {
645 /* Prepare header of the next frame,
646 * before previous one went down. */
647 if (frag) {
648 frag->ip_summed = CHECKSUM_NONE;
649 skb_reset_transport_header(frag);
650 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
651 __skb_push(frag, hlen);
652 skb_reset_network_header(frag);
653 memcpy(skb_network_header(frag), tmp_hdr,
654 hlen);
655 offset += skb->len - hlen - sizeof(struct frag_hdr);
656 fh->nexthdr = nexthdr;
657 fh->reserved = 0;
658 fh->frag_off = htons(offset);
659 if (frag->next != NULL)
660 fh->frag_off |= htons(IP6_MF);
661 fh->identification = frag_id;
662 ipv6_hdr(frag)->payload_len =
663 htons(frag->len -
664 sizeof(struct ipv6hdr));
665 ip6_copy_metadata(frag, skb);
666 }
667
668 err = output(skb);
669 if(!err)
670 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
671 IPSTATS_MIB_FRAGCREATES);
672
673 if (err || !frag)
674 break;
675
676 skb = frag;
677 frag = skb->next;
678 skb->next = NULL;
679 }
680
681 kfree(tmp_hdr);
682
683 if (err == 0) {
684 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
685 IPSTATS_MIB_FRAGOKS);
686 ip6_rt_put(rt);
687 return 0;
688 }
689
690 while (frag) {
691 skb = frag->next;
692 kfree_skb(frag);
693 frag = skb;
694 }
695
696 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
697 IPSTATS_MIB_FRAGFAILS);
698 ip6_rt_put(rt);
699 return err;
700
701slow_path_clean:
702 skb_walk_frags(skb, frag2) {
703 if (frag2 == frag)
704 break;
705 frag2->sk = NULL;
706 frag2->destructor = NULL;
707 skb->truesize += frag2->truesize;
708 }
709 }
710
711slow_path:
712 if ((skb->ip_summed == CHECKSUM_PARTIAL) &&
713 skb_checksum_help(skb))
714 goto fail;
715
716 left = skb->len - hlen; /* Space per frame */
717 ptr = hlen; /* Where to start from */
718
719 /*
720 * Fragment the datagram.
721 */
722
723 *prevhdr = NEXTHDR_FRAGMENT;
724 hroom = LL_RESERVED_SPACE(rt->dst.dev);
725 troom = rt->dst.dev->needed_tailroom;
726
727 /*
728 * Keep copying data until we run out.
729 */
730 while(left > 0) {
731 len = left;
732 /* IF: it doesn't fit, use 'mtu' - the data space left */
733 if (len > mtu)
734 len = mtu;
735 /* IF: we are not sending up to and including the packet end
736 then align the next start on an eight byte boundary */
737 if (len < left) {
738 len &= ~7;
739 }
740 /*
741 * Allocate buffer.
742 */
743
744 if ((frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
745 hroom + troom, GFP_ATOMIC)) == NULL) {
746 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
747 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
748 IPSTATS_MIB_FRAGFAILS);
749 err = -ENOMEM;
750 goto fail;
751 }
752
753 /*
754 * Set up data on packet
755 */
756
757 ip6_copy_metadata(frag, skb);
758 skb_reserve(frag, hroom);
759 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
760 skb_reset_network_header(frag);
761 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
762 frag->transport_header = (frag->network_header + hlen +
763 sizeof(struct frag_hdr));
764
765 /*
766 * Charge the memory for the fragment to any owner
767 * it might possess
768 */
769 if (skb->sk)
770 skb_set_owner_w(frag, skb->sk);
771
772 /*
773 * Copy the packet header into the new buffer.
774 */
775 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
776
777 /*
778 * Build fragment header.
779 */
780 fh->nexthdr = nexthdr;
781 fh->reserved = 0;
782 if (!frag_id) {
783 ipv6_select_ident(fh, rt);
784 frag_id = fh->identification;
785 } else
786 fh->identification = frag_id;
787
788 /*
789 * Copy a block of the IP datagram.
790 */
791 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
792 BUG();
793 left -= len;
794
795 fh->frag_off = htons(offset);
796 if (left > 0)
797 fh->frag_off |= htons(IP6_MF);
798 ipv6_hdr(frag)->payload_len = htons(frag->len -
799 sizeof(struct ipv6hdr));
800
801 ptr += len;
802 offset += len;
803
804 /*
805 * Put this fragment into the sending queue.
806 */
807 err = output(frag);
808 if (err)
809 goto fail;
810
811 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
812 IPSTATS_MIB_FRAGCREATES);
813 }
814 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
815 IPSTATS_MIB_FRAGOKS);
816 consume_skb(skb);
817 return err;
818
819fail:
820 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
821 IPSTATS_MIB_FRAGFAILS);
822 kfree_skb(skb);
823 return err;
824}
825
826static inline int ip6_rt_check(const struct rt6key *rt_key,
827 const struct in6_addr *fl_addr,
828 const struct in6_addr *addr_cache)
829{
830 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
831 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache));
832}
833
834static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
835 struct dst_entry *dst,
836 const struct flowi6 *fl6)
837{
838 struct ipv6_pinfo *np = inet6_sk(sk);
839 struct rt6_info *rt;
840
841 if (!dst)
842 goto out;
843
844 if (dst->ops->family != AF_INET6) {
845 dst_release(dst);
846 return NULL;
847 }
848
849 rt = (struct rt6_info *)dst;
850 /* Yes, checking route validity in not connected
851 * case is not very simple. Take into account,
852 * that we do not support routing by source, TOS,
853 * and MSG_DONTROUTE --ANK (980726)
854 *
855 * 1. ip6_rt_check(): If route was host route,
856 * check that cached destination is current.
857 * If it is network route, we still may
858 * check its validity using saved pointer
859 * to the last used address: daddr_cache.
860 * We do not want to save whole address now,
861 * (because main consumer of this service
862 * is tcp, which has not this problem),
863 * so that the last trick works only on connected
864 * sockets.
865 * 2. oif also should be the same.
866 */
867 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
868#ifdef CONFIG_IPV6_SUBTREES
869 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
870#endif
871 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
872 dst_release(dst);
873 dst = NULL;
874 }
875
876out:
877 return dst;
878}
879
880static int ip6_dst_lookup_tail(struct sock *sk,
881 struct dst_entry **dst, struct flowi6 *fl6)
882{
883 struct net *net = sock_net(sk);
884#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
885 struct neighbour *n;
886 struct rt6_info *rt;
887#endif
888 int err;
889
890 if (*dst == NULL)
891 *dst = ip6_route_output(net, sk, fl6);
892
893 if ((err = (*dst)->error))
894 goto out_err_release;
895
896 if (ipv6_addr_any(&fl6->saddr)) {
897 struct rt6_info *rt = (struct rt6_info *) *dst;
898 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
899 sk ? inet6_sk(sk)->srcprefs : 0,
900 &fl6->saddr);
901 if (err)
902 goto out_err_release;
903 }
904
905#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
906 /*
907 * Here if the dst entry we've looked up
908 * has a neighbour entry that is in the INCOMPLETE
909 * state and the src address from the flow is
910 * marked as OPTIMISTIC, we release the found
911 * dst entry and replace it instead with the
912 * dst entry of the nexthop router
913 */
914 rt = (struct rt6_info *) *dst;
915 rcu_read_lock_bh();
916 n = __ipv6_neigh_lookup_noref(rt->dst.dev, rt6_nexthop(rt));
917 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
918 rcu_read_unlock_bh();
919
920 if (err) {
921 struct inet6_ifaddr *ifp;
922 struct flowi6 fl_gw6;
923 int redirect;
924
925 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
926 (*dst)->dev, 1);
927
928 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
929 if (ifp)
930 in6_ifa_put(ifp);
931
932 if (redirect) {
933 /*
934 * We need to get the dst entry for the
935 * default router instead
936 */
937 dst_release(*dst);
938 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
939 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
940 *dst = ip6_route_output(net, sk, &fl_gw6);
941 if ((err = (*dst)->error))
942 goto out_err_release;
943 }
944 }
945#endif
946
947 return 0;
948
949out_err_release:
950 if (err == -ENETUNREACH)
951 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
952 dst_release(*dst);
953 *dst = NULL;
954 return err;
955}
956
957/**
958 * ip6_dst_lookup - perform route lookup on flow
959 * @sk: socket which provides route info
960 * @dst: pointer to dst_entry * for result
961 * @fl6: flow to lookup
962 *
963 * This function performs a route lookup on the given flow.
964 *
965 * It returns zero on success, or a standard errno code on error.
966 */
967int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi6 *fl6)
968{
969 *dst = NULL;
970 return ip6_dst_lookup_tail(sk, dst, fl6);
971}
972EXPORT_SYMBOL_GPL(ip6_dst_lookup);
973
974/**
975 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
976 * @sk: socket which provides route info
977 * @fl6: flow to lookup
978 * @final_dst: final destination address for ipsec lookup
979 *
980 * This function performs a route lookup on the given flow.
981 *
982 * It returns a valid dst pointer on success, or a pointer encoded
983 * error code.
984 */
985struct dst_entry *ip6_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
986 const struct in6_addr *final_dst)
987{
988 struct dst_entry *dst = NULL;
989 int err;
990
991 err = ip6_dst_lookup_tail(sk, &dst, fl6);
992 if (err)
993 return ERR_PTR(err);
994 if (final_dst)
995 fl6->daddr = *final_dst;
996
997 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
998}
999EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1000
1001/**
1002 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1003 * @sk: socket which provides the dst cache and route info
1004 * @fl6: flow to lookup
1005 * @final_dst: final destination address for ipsec lookup
1006 *
1007 * This function performs a route lookup on the given flow with the
1008 * possibility of using the cached route in the socket if it is valid.
1009 * It will take the socket dst lock when operating on the dst cache.
1010 * As a result, this function can only be used in process context.
1011 *
1012 * It returns a valid dst pointer on success, or a pointer encoded
1013 * error code.
1014 */
1015struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1016 const struct in6_addr *final_dst)
1017{
1018 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1019 int err;
1020
1021 dst = ip6_sk_dst_check(sk, dst, fl6);
1022
1023 err = ip6_dst_lookup_tail(sk, &dst, fl6);
1024 if (err)
1025 return ERR_PTR(err);
1026 if (final_dst)
1027 fl6->daddr = *final_dst;
1028
1029 return xfrm_lookup(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1030}
1031EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1032
1033static inline int ip6_ufo_append_data(struct sock *sk,
1034 int getfrag(void *from, char *to, int offset, int len,
1035 int odd, struct sk_buff *skb),
1036 void *from, int length, int hh_len, int fragheaderlen,
1037 int transhdrlen, int mtu,unsigned int flags,
1038 struct rt6_info *rt)
1039
1040{
1041 struct sk_buff *skb;
1042 struct frag_hdr fhdr;
1043 int err;
1044
1045 /* There is support for UDP large send offload by network
1046 * device, so create one single skb packet containing complete
1047 * udp datagram
1048 */
1049 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1050 skb = sock_alloc_send_skb(sk,
1051 hh_len + fragheaderlen + transhdrlen + 20,
1052 (flags & MSG_DONTWAIT), &err);
1053 if (skb == NULL)
1054 return err;
1055
1056 /* reserve space for Hardware header */
1057 skb_reserve(skb, hh_len);
1058
1059 /* create space for UDP/IP header */
1060 skb_put(skb,fragheaderlen + transhdrlen);
1061
1062 /* initialize network header pointer */
1063 skb_reset_network_header(skb);
1064
1065 /* initialize protocol header pointer */
1066 skb->transport_header = skb->network_header + fragheaderlen;
1067
1068 skb->protocol = htons(ETH_P_IPV6);
1069 skb->csum = 0;
1070
1071 __skb_queue_tail(&sk->sk_write_queue, skb);
1072 } else if (skb_is_gso(skb)) {
1073 goto append;
1074 }
1075
1076 skb->ip_summed = CHECKSUM_PARTIAL;
1077 /* Specify the length of each IPv6 datagram fragment.
1078 * It has to be a multiple of 8.
1079 */
1080 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1081 sizeof(struct frag_hdr)) & ~7;
1082 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1083 ipv6_select_ident(&fhdr, rt);
1084 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1085
1086append:
1087 return skb_append_datato_frags(sk, skb, getfrag, from,
1088 (length - transhdrlen));
1089}
1090
1091static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1092 gfp_t gfp)
1093{
1094 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1095}
1096
1097static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1098 gfp_t gfp)
1099{
1100 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1101}
1102
1103static void ip6_append_data_mtu(unsigned int *mtu,
1104 int *maxfraglen,
1105 unsigned int fragheaderlen,
1106 struct sk_buff *skb,
1107 struct rt6_info *rt,
1108 unsigned int orig_mtu)
1109{
1110 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1111 if (skb == NULL) {
1112 /* first fragment, reserve header_len */
1113 *mtu = orig_mtu - rt->dst.header_len;
1114
1115 } else {
1116 /*
1117 * this fragment is not first, the headers
1118 * space is regarded as data space.
1119 */
1120 *mtu = orig_mtu;
1121 }
1122 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1123 + fragheaderlen - sizeof(struct frag_hdr);
1124 }
1125}
1126
1127int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1128 int offset, int len, int odd, struct sk_buff *skb),
1129 void *from, int length, int transhdrlen,
1130 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1131 struct rt6_info *rt, unsigned int flags, int dontfrag)
1132{
1133 struct inet_sock *inet = inet_sk(sk);
1134 struct ipv6_pinfo *np = inet6_sk(sk);
1135 struct inet_cork *cork;
1136 struct sk_buff *skb, *skb_prev = NULL;
1137 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1138 int exthdrlen;
1139 int dst_exthdrlen;
1140 int hh_len;
1141 int copy;
1142 int err;
1143 int offset = 0;
1144 __u8 tx_flags = 0;
1145
1146 if (flags&MSG_PROBE)
1147 return 0;
1148 cork = &inet->cork.base;
1149 if (skb_queue_empty(&sk->sk_write_queue)) {
1150 /*
1151 * setup for corking
1152 */
1153 if (opt) {
1154 if (WARN_ON(np->cork.opt))
1155 return -EINVAL;
1156
1157 np->cork.opt = kzalloc(opt->tot_len, sk->sk_allocation);
1158 if (unlikely(np->cork.opt == NULL))
1159 return -ENOBUFS;
1160
1161 np->cork.opt->tot_len = opt->tot_len;
1162 np->cork.opt->opt_flen = opt->opt_flen;
1163 np->cork.opt->opt_nflen = opt->opt_nflen;
1164
1165 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1166 sk->sk_allocation);
1167 if (opt->dst0opt && !np->cork.opt->dst0opt)
1168 return -ENOBUFS;
1169
1170 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1171 sk->sk_allocation);
1172 if (opt->dst1opt && !np->cork.opt->dst1opt)
1173 return -ENOBUFS;
1174
1175 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1176 sk->sk_allocation);
1177 if (opt->hopopt && !np->cork.opt->hopopt)
1178 return -ENOBUFS;
1179
1180 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1181 sk->sk_allocation);
1182 if (opt->srcrt && !np->cork.opt->srcrt)
1183 return -ENOBUFS;
1184
1185 /* need source address above miyazawa*/
1186 }
1187 dst_hold(&rt->dst);
1188 cork->dst = &rt->dst;
1189 inet->cork.fl.u.ip6 = *fl6;
1190 np->cork.hop_limit = hlimit;
1191 np->cork.tclass = tclass;
1192 if (rt->dst.flags & DST_XFRM_TUNNEL)
1193 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1194 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1195 else
1196 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1197 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1198 if (np->frag_size < mtu) {
1199 if (np->frag_size)
1200 mtu = np->frag_size;
1201 }
1202 cork->fragsize = mtu;
1203 if (dst_allfrag(rt->dst.path))
1204 cork->flags |= IPCORK_ALLFRAG;
1205 cork->length = 0;
1206 exthdrlen = (opt ? opt->opt_flen : 0);
1207 length += exthdrlen;
1208 transhdrlen += exthdrlen;
1209 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1210 } else {
1211 rt = (struct rt6_info *)cork->dst;
1212 fl6 = &inet->cork.fl.u.ip6;
1213 opt = np->cork.opt;
1214 transhdrlen = 0;
1215 exthdrlen = 0;
1216 dst_exthdrlen = 0;
1217 mtu = cork->fragsize;
1218 }
1219 orig_mtu = mtu;
1220
1221 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1222
1223 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1224 (opt ? opt->opt_nflen : 0);
1225 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1226 sizeof(struct frag_hdr);
1227
1228 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1229 unsigned int maxnonfragsize, headersize;
1230
1231 headersize = sizeof(struct ipv6hdr) +
1232 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1233 (dst_allfrag(&rt->dst) ?
1234 sizeof(struct frag_hdr) : 0) +
1235 rt->rt6i_nfheader_len;
1236
1237 if (ip6_sk_local_df(sk))
1238 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1239 else
1240 maxnonfragsize = mtu;
1241
1242 /* dontfrag active */
1243 if ((cork->length + length > mtu - headersize) && dontfrag &&
1244 (sk->sk_protocol == IPPROTO_UDP ||
1245 sk->sk_protocol == IPPROTO_RAW)) {
1246 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1247 sizeof(struct ipv6hdr));
1248 goto emsgsize;
1249 }
1250
1251 if (cork->length + length > maxnonfragsize - headersize) {
1252emsgsize:
1253 ipv6_local_error(sk, EMSGSIZE, fl6,
1254 mtu - headersize +
1255 sizeof(struct ipv6hdr));
1256 return -EMSGSIZE;
1257 }
1258 }
1259
1260 /* For UDP, check if TX timestamp is enabled */
1261 if (sk->sk_type == SOCK_DGRAM)
1262 sock_tx_timestamp(sk, &tx_flags);
1263
1264 /*
1265 * Let's try using as much space as possible.
1266 * Use MTU if total length of the message fits into the MTU.
1267 * Otherwise, we need to reserve fragment header and
1268 * fragment alignment (= 8-15 octects, in total).
1269 *
1270 * Note that we may need to "move" the data from the tail of
1271 * of the buffer to the new fragment when we split
1272 * the message.
1273 *
1274 * FIXME: It may be fragmented into multiple chunks
1275 * at once if non-fragmentable extension headers
1276 * are too large.
1277 * --yoshfuji
1278 */
1279
1280 skb = skb_peek_tail(&sk->sk_write_queue);
1281 cork->length += length;
1282 if (((length > mtu) ||
1283 (skb && skb_is_gso(skb))) &&
1284 (sk->sk_protocol == IPPROTO_UDP) &&
1285 (rt->dst.dev->features & NETIF_F_UFO)) {
1286 err = ip6_ufo_append_data(sk, getfrag, from, length,
1287 hh_len, fragheaderlen,
1288 transhdrlen, mtu, flags, rt);
1289 if (err)
1290 goto error;
1291 return 0;
1292 }
1293
1294 if (!skb)
1295 goto alloc_new_skb;
1296
1297 while (length > 0) {
1298 /* Check if the remaining data fits into current packet. */
1299 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1300 if (copy < length)
1301 copy = maxfraglen - skb->len;
1302
1303 if (copy <= 0) {
1304 char *data;
1305 unsigned int datalen;
1306 unsigned int fraglen;
1307 unsigned int fraggap;
1308 unsigned int alloclen;
1309alloc_new_skb:
1310 /* There's no room in the current skb */
1311 if (skb)
1312 fraggap = skb->len - maxfraglen;
1313 else
1314 fraggap = 0;
1315 /* update mtu and maxfraglen if necessary */
1316 if (skb == NULL || skb_prev == NULL)
1317 ip6_append_data_mtu(&mtu, &maxfraglen,
1318 fragheaderlen, skb, rt,
1319 orig_mtu);
1320
1321 skb_prev = skb;
1322
1323 /*
1324 * If remaining data exceeds the mtu,
1325 * we know we need more fragment(s).
1326 */
1327 datalen = length + fraggap;
1328
1329 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1330 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1331 if ((flags & MSG_MORE) &&
1332 !(rt->dst.dev->features&NETIF_F_SG))
1333 alloclen = mtu;
1334 else
1335 alloclen = datalen + fragheaderlen;
1336
1337 alloclen += dst_exthdrlen;
1338
1339 if (datalen != length + fraggap) {
1340 /*
1341 * this is not the last fragment, the trailer
1342 * space is regarded as data space.
1343 */
1344 datalen += rt->dst.trailer_len;
1345 }
1346
1347 alloclen += rt->dst.trailer_len;
1348 fraglen = datalen + fragheaderlen;
1349
1350 /*
1351 * We just reserve space for fragment header.
1352 * Note: this may be overallocation if the message
1353 * (without MSG_MORE) fits into the MTU.
1354 */
1355 alloclen += sizeof(struct frag_hdr);
1356
1357 if (transhdrlen) {
1358 skb = sock_alloc_send_skb(sk,
1359 alloclen + hh_len,
1360 (flags & MSG_DONTWAIT), &err);
1361 } else {
1362 skb = NULL;
1363 if (atomic_read(&sk->sk_wmem_alloc) <=
1364 2 * sk->sk_sndbuf)
1365 skb = sock_wmalloc(sk,
1366 alloclen + hh_len, 1,
1367 sk->sk_allocation);
1368 if (unlikely(skb == NULL))
1369 err = -ENOBUFS;
1370 else {
1371 /* Only the initial fragment
1372 * is time stamped.
1373 */
1374 tx_flags = 0;
1375 }
1376 }
1377 if (skb == NULL)
1378 goto error;
1379 /*
1380 * Fill in the control structures
1381 */
1382 skb->protocol = htons(ETH_P_IPV6);
1383 skb->ip_summed = CHECKSUM_NONE;
1384 skb->csum = 0;
1385 /* reserve for fragmentation and ipsec header */
1386 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1387 dst_exthdrlen);
1388
1389 if (sk->sk_type == SOCK_DGRAM)
1390 skb_shinfo(skb)->tx_flags = tx_flags;
1391
1392 /*
1393 * Find where to start putting bytes
1394 */
1395 data = skb_put(skb, fraglen);
1396 skb_set_network_header(skb, exthdrlen);
1397 data += fragheaderlen;
1398 skb->transport_header = (skb->network_header +
1399 fragheaderlen);
1400 if (fraggap) {
1401 skb->csum = skb_copy_and_csum_bits(
1402 skb_prev, maxfraglen,
1403 data + transhdrlen, fraggap, 0);
1404 skb_prev->csum = csum_sub(skb_prev->csum,
1405 skb->csum);
1406 data += fraggap;
1407 pskb_trim_unique(skb_prev, maxfraglen);
1408 }
1409 copy = datalen - transhdrlen - fraggap;
1410
1411 if (copy < 0) {
1412 err = -EINVAL;
1413 kfree_skb(skb);
1414 goto error;
1415 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1416 err = -EFAULT;
1417 kfree_skb(skb);
1418 goto error;
1419 }
1420
1421 offset += copy;
1422 length -= datalen - fraggap;
1423 transhdrlen = 0;
1424 exthdrlen = 0;
1425 dst_exthdrlen = 0;
1426
1427 /*
1428 * Put the packet on the pending queue
1429 */
1430 __skb_queue_tail(&sk->sk_write_queue, skb);
1431 continue;
1432 }
1433
1434 if (copy > length)
1435 copy = length;
1436
1437 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1438 unsigned int off;
1439
1440 off = skb->len;
1441 if (getfrag(from, skb_put(skb, copy),
1442 offset, copy, off, skb) < 0) {
1443 __skb_trim(skb, off);
1444 err = -EFAULT;
1445 goto error;
1446 }
1447 } else {
1448 int i = skb_shinfo(skb)->nr_frags;
1449 struct page_frag *pfrag = sk_page_frag(sk);
1450
1451 err = -ENOMEM;
1452 if (!sk_page_frag_refill(sk, pfrag))
1453 goto error;
1454
1455 if (!skb_can_coalesce(skb, i, pfrag->page,
1456 pfrag->offset)) {
1457 err = -EMSGSIZE;
1458 if (i == MAX_SKB_FRAGS)
1459 goto error;
1460
1461 __skb_fill_page_desc(skb, i, pfrag->page,
1462 pfrag->offset, 0);
1463 skb_shinfo(skb)->nr_frags = ++i;
1464 get_page(pfrag->page);
1465 }
1466 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1467 if (getfrag(from,
1468 page_address(pfrag->page) + pfrag->offset,
1469 offset, copy, skb->len, skb) < 0)
1470 goto error_efault;
1471
1472 pfrag->offset += copy;
1473 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1474 skb->len += copy;
1475 skb->data_len += copy;
1476 skb->truesize += copy;
1477 atomic_add(copy, &sk->sk_wmem_alloc);
1478 }
1479 offset += copy;
1480 length -= copy;
1481 }
1482
1483 return 0;
1484
1485error_efault:
1486 err = -EFAULT;
1487error:
1488 cork->length -= length;
1489 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1490 return err;
1491}
1492EXPORT_SYMBOL_GPL(ip6_append_data);
1493
1494static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1495{
1496 if (np->cork.opt) {
1497 kfree(np->cork.opt->dst0opt);
1498 kfree(np->cork.opt->dst1opt);
1499 kfree(np->cork.opt->hopopt);
1500 kfree(np->cork.opt->srcrt);
1501 kfree(np->cork.opt);
1502 np->cork.opt = NULL;
1503 }
1504
1505 if (inet->cork.base.dst) {
1506 dst_release(inet->cork.base.dst);
1507 inet->cork.base.dst = NULL;
1508 inet->cork.base.flags &= ~IPCORK_ALLFRAG;
1509 }
1510 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1511}
1512
1513int ip6_push_pending_frames(struct sock *sk)
1514{
1515 struct sk_buff *skb, *tmp_skb;
1516 struct sk_buff **tail_skb;
1517 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1518 struct inet_sock *inet = inet_sk(sk);
1519 struct ipv6_pinfo *np = inet6_sk(sk);
1520 struct net *net = sock_net(sk);
1521 struct ipv6hdr *hdr;
1522 struct ipv6_txoptions *opt = np->cork.opt;
1523 struct rt6_info *rt = (struct rt6_info *)inet->cork.base.dst;
1524 struct flowi6 *fl6 = &inet->cork.fl.u.ip6;
1525 unsigned char proto = fl6->flowi6_proto;
1526 int err = 0;
1527
1528 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1529 goto out;
1530 tail_skb = &(skb_shinfo(skb)->frag_list);
1531
1532 /* move skb->data to ip header from ext header */
1533 if (skb->data < skb_network_header(skb))
1534 __skb_pull(skb, skb_network_offset(skb));
1535 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1536 __skb_pull(tmp_skb, skb_network_header_len(skb));
1537 *tail_skb = tmp_skb;
1538 tail_skb = &(tmp_skb->next);
1539 skb->len += tmp_skb->len;
1540 skb->data_len += tmp_skb->len;
1541 skb->truesize += tmp_skb->truesize;
1542 tmp_skb->destructor = NULL;
1543 tmp_skb->sk = NULL;
1544 }
1545
1546 /* Allow local fragmentation. */
1547 skb->local_df = ip6_sk_local_df(sk);
1548
1549 *final_dst = fl6->daddr;
1550 __skb_pull(skb, skb_network_header_len(skb));
1551 if (opt && opt->opt_flen)
1552 ipv6_push_frag_opts(skb, opt, &proto);
1553 if (opt && opt->opt_nflen)
1554 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1555
1556 skb_push(skb, sizeof(struct ipv6hdr));
1557 skb_reset_network_header(skb);
1558 hdr = ipv6_hdr(skb);
1559
1560 ip6_flow_hdr(hdr, np->cork.tclass, fl6->flowlabel);
1561 hdr->hop_limit = np->cork.hop_limit;
1562 hdr->nexthdr = proto;
1563 hdr->saddr = fl6->saddr;
1564 hdr->daddr = *final_dst;
1565
1566 skb->priority = sk->sk_priority;
1567 skb->mark = sk->sk_mark;
1568
1569 skb_dst_set(skb, dst_clone(&rt->dst));
1570 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1571 if (proto == IPPROTO_ICMPV6) {
1572 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1573
1574 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1575 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1576 }
1577
1578 err = ip6_local_out(skb);
1579 if (err) {
1580 if (err > 0)
1581 err = net_xmit_errno(err);
1582 if (err)
1583 goto error;
1584 }
1585
1586out:
1587 ip6_cork_release(inet, np);
1588 return err;
1589error:
1590 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1591 goto out;
1592}
1593EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1594
1595void ip6_flush_pending_frames(struct sock *sk)
1596{
1597 struct sk_buff *skb;
1598
1599 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
1600 if (skb_dst(skb))
1601 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1602 IPSTATS_MIB_OUTDISCARDS);
1603 kfree_skb(skb);
1604 }
1605
1606 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
1607}
1608EXPORT_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/gso.h>
46#include <net/ipv6.h>
47#include <net/ndisc.h>
48#include <net/protocol.h>
49#include <net/ip6_route.h>
50#include <net/addrconf.h>
51#include <net/rawv6.h>
52#include <net/icmp.h>
53#include <net/xfrm.h>
54#include <net/checksum.h>
55#include <linux/mroute6.h>
56#include <net/l3mdev.h>
57#include <net/lwtunnel.h>
58#include <net/ip_tunnels.h>
59
60static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
61{
62 struct dst_entry *dst = skb_dst(skb);
63 struct net_device *dev = dst->dev;
64 struct inet6_dev *idev = ip6_dst_idev(dst);
65 unsigned int hh_len = LL_RESERVED_SPACE(dev);
66 const struct in6_addr *daddr, *nexthop;
67 struct ipv6hdr *hdr;
68 struct neighbour *neigh;
69 int ret;
70
71 /* Be paranoid, rather than too clever. */
72 if (unlikely(hh_len > skb_headroom(skb)) && dev->header_ops) {
73 /* Make sure idev stays alive */
74 rcu_read_lock();
75 skb = skb_expand_head(skb, hh_len);
76 if (!skb) {
77 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
78 rcu_read_unlock();
79 return -ENOMEM;
80 }
81 rcu_read_unlock();
82 }
83
84 hdr = ipv6_hdr(skb);
85 daddr = &hdr->daddr;
86 if (ipv6_addr_is_multicast(daddr)) {
87 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
88 ((mroute6_is_socket(net, skb) &&
89 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
90 ipv6_chk_mcast_addr(dev, daddr, &hdr->saddr))) {
91 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
92
93 /* Do not check for IFF_ALLMULTI; multicast routing
94 is not supported in any case.
95 */
96 if (newskb)
97 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
98 net, sk, newskb, NULL, newskb->dev,
99 dev_loopback_xmit);
100
101 if (hdr->hop_limit == 0) {
102 IP6_INC_STATS(net, idev,
103 IPSTATS_MIB_OUTDISCARDS);
104 kfree_skb(skb);
105 return 0;
106 }
107 }
108
109 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, skb->len);
110 if (IPV6_ADDR_MC_SCOPE(daddr) <= IPV6_ADDR_SCOPE_NODELOCAL &&
111 !(dev->flags & IFF_LOOPBACK)) {
112 kfree_skb(skb);
113 return 0;
114 }
115 }
116
117 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
118 int res = lwtunnel_xmit(skb);
119
120 if (res != LWTUNNEL_XMIT_CONTINUE)
121 return res;
122 }
123
124 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUT, skb->len);
125
126 rcu_read_lock();
127 nexthop = rt6_nexthop(dst_rt6_info(dst), daddr);
128 neigh = __ipv6_neigh_lookup_noref(dev, nexthop);
129
130 if (IS_ERR_OR_NULL(neigh)) {
131 if (unlikely(!neigh))
132 neigh = __neigh_create(&nd_tbl, nexthop, dev, false);
133 if (IS_ERR(neigh)) {
134 rcu_read_unlock();
135 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTNOROUTES);
136 kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
137 return -EINVAL;
138 }
139 }
140 sock_confirm_neigh(skb, neigh);
141 ret = neigh_output(neigh, skb, false);
142 rcu_read_unlock();
143 return ret;
144}
145
146static int
147ip6_finish_output_gso_slowpath_drop(struct net *net, struct sock *sk,
148 struct sk_buff *skb, unsigned int mtu)
149{
150 struct sk_buff *segs, *nskb;
151 netdev_features_t features;
152 int ret = 0;
153
154 /* Please see corresponding comment in ip_finish_output_gso
155 * describing the cases where GSO segment length exceeds the
156 * egress MTU.
157 */
158 features = netif_skb_features(skb);
159 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
160 if (IS_ERR_OR_NULL(segs)) {
161 kfree_skb(skb);
162 return -ENOMEM;
163 }
164
165 consume_skb(skb);
166
167 skb_list_walk_safe(segs, segs, nskb) {
168 int err;
169
170 skb_mark_not_on_list(segs);
171 /* Last GSO segment can be smaller than gso_size (and MTU).
172 * Adding a fragment header would produce an "atomic fragment",
173 * which is considered harmful (RFC-8021). Avoid that.
174 */
175 err = segs->len > mtu ?
176 ip6_fragment(net, sk, segs, ip6_finish_output2) :
177 ip6_finish_output2(net, sk, segs);
178 if (err && ret == 0)
179 ret = err;
180 }
181
182 return ret;
183}
184
185static int ip6_finish_output_gso(struct net *net, struct sock *sk,
186 struct sk_buff *skb, unsigned int mtu)
187{
188 if (!(IP6CB(skb)->flags & IP6SKB_FAKEJUMBO) &&
189 !skb_gso_validate_network_len(skb, mtu))
190 return ip6_finish_output_gso_slowpath_drop(net, sk, skb, mtu);
191
192 return ip6_finish_output2(net, sk, skb);
193}
194
195static int __ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
196{
197 unsigned int mtu;
198
199#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
200 /* Policy lookup after SNAT yielded a new policy */
201 if (skb_dst(skb)->xfrm) {
202 IP6CB(skb)->flags |= IP6SKB_REROUTED;
203 return dst_output(net, sk, skb);
204 }
205#endif
206
207 mtu = ip6_skb_dst_mtu(skb);
208 if (skb_is_gso(skb))
209 return ip6_finish_output_gso(net, sk, skb, mtu);
210
211 if (skb->len > mtu ||
212 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
213 return ip6_fragment(net, sk, skb, ip6_finish_output2);
214
215 return ip6_finish_output2(net, sk, skb);
216}
217
218static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
219{
220 int ret;
221
222 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
223 switch (ret) {
224 case NET_XMIT_SUCCESS:
225 case NET_XMIT_CN:
226 return __ip6_finish_output(net, sk, skb) ? : ret;
227 default:
228 kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
229 return ret;
230 }
231}
232
233int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
234{
235 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
236 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
237
238 skb->protocol = htons(ETH_P_IPV6);
239 skb->dev = dev;
240
241 if (unlikely(!idev || READ_ONCE(idev->cnf.disable_ipv6))) {
242 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
243 kfree_skb_reason(skb, SKB_DROP_REASON_IPV6DISABLED);
244 return 0;
245 }
246
247 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
248 net, sk, skb, indev, dev,
249 ip6_finish_output,
250 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
251}
252EXPORT_SYMBOL(ip6_output);
253
254bool ip6_autoflowlabel(struct net *net, const struct sock *sk)
255{
256 if (!inet6_test_bit(AUTOFLOWLABEL_SET, sk))
257 return ip6_default_np_autolabel(net);
258 return inet6_test_bit(AUTOFLOWLABEL, sk);
259}
260
261/*
262 * xmit an sk_buff (used by TCP, SCTP and DCCP)
263 * Note : socket lock is not held for SYNACK packets, but might be modified
264 * by calls to skb_set_owner_w() and ipv6_local_error(),
265 * which are using proper atomic operations or spinlocks.
266 */
267int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
268 __u32 mark, struct ipv6_txoptions *opt, int tclass, u32 priority)
269{
270 struct net *net = sock_net(sk);
271 const struct ipv6_pinfo *np = inet6_sk(sk);
272 struct in6_addr *first_hop = &fl6->daddr;
273 struct dst_entry *dst = skb_dst(skb);
274 struct net_device *dev = dst->dev;
275 struct inet6_dev *idev = ip6_dst_idev(dst);
276 struct hop_jumbo_hdr *hop_jumbo;
277 int hoplen = sizeof(*hop_jumbo);
278 unsigned int head_room;
279 struct ipv6hdr *hdr;
280 u8 proto = fl6->flowi6_proto;
281 int seg_len = skb->len;
282 int hlimit = -1;
283 u32 mtu;
284
285 head_room = sizeof(struct ipv6hdr) + hoplen + LL_RESERVED_SPACE(dev);
286 if (opt)
287 head_room += opt->opt_nflen + opt->opt_flen;
288
289 if (unlikely(head_room > skb_headroom(skb))) {
290 /* Make sure idev stays alive */
291 rcu_read_lock();
292 skb = skb_expand_head(skb, head_room);
293 if (!skb) {
294 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
295 rcu_read_unlock();
296 return -ENOBUFS;
297 }
298 rcu_read_unlock();
299 }
300
301 if (opt) {
302 seg_len += opt->opt_nflen + opt->opt_flen;
303
304 if (opt->opt_flen)
305 ipv6_push_frag_opts(skb, opt, &proto);
306
307 if (opt->opt_nflen)
308 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop,
309 &fl6->saddr);
310 }
311
312 if (unlikely(seg_len > IPV6_MAXPLEN)) {
313 hop_jumbo = skb_push(skb, hoplen);
314
315 hop_jumbo->nexthdr = proto;
316 hop_jumbo->hdrlen = 0;
317 hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
318 hop_jumbo->tlv_len = 4;
319 hop_jumbo->jumbo_payload_len = htonl(seg_len + hoplen);
320
321 proto = IPPROTO_HOPOPTS;
322 seg_len = 0;
323 IP6CB(skb)->flags |= IP6SKB_FAKEJUMBO;
324 }
325
326 skb_push(skb, sizeof(struct ipv6hdr));
327 skb_reset_network_header(skb);
328 hdr = ipv6_hdr(skb);
329
330 /*
331 * Fill in the IPv6 header
332 */
333 if (np)
334 hlimit = READ_ONCE(np->hop_limit);
335 if (hlimit < 0)
336 hlimit = ip6_dst_hoplimit(dst);
337
338 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
339 ip6_autoflowlabel(net, sk), fl6));
340
341 hdr->payload_len = htons(seg_len);
342 hdr->nexthdr = proto;
343 hdr->hop_limit = hlimit;
344
345 hdr->saddr = fl6->saddr;
346 hdr->daddr = *first_hop;
347
348 skb->protocol = htons(ETH_P_IPV6);
349 skb->priority = priority;
350 skb->mark = mark;
351
352 mtu = dst_mtu(dst);
353 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
354 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTREQUESTS);
355
356 /* if egress device is enslaved to an L3 master device pass the
357 * skb to its handler for processing
358 */
359 skb = l3mdev_ip6_out((struct sock *)sk, skb);
360 if (unlikely(!skb))
361 return 0;
362
363 /* hooks should never assume socket lock is held.
364 * we promote our socket to non const
365 */
366 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
367 net, (struct sock *)sk, skb, NULL, dev,
368 dst_output);
369 }
370
371 skb->dev = dev;
372 /* ipv6_local_error() does not require socket lock,
373 * we promote our socket to non const
374 */
375 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
376
377 IP6_INC_STATS(net, idev, IPSTATS_MIB_FRAGFAILS);
378 kfree_skb(skb);
379 return -EMSGSIZE;
380}
381EXPORT_SYMBOL(ip6_xmit);
382
383static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
384{
385 struct ip6_ra_chain *ra;
386 struct sock *last = NULL;
387
388 read_lock(&ip6_ra_lock);
389 for (ra = ip6_ra_chain; ra; ra = ra->next) {
390 struct sock *sk = ra->sk;
391 if (sk && ra->sel == sel &&
392 (!sk->sk_bound_dev_if ||
393 sk->sk_bound_dev_if == skb->dev->ifindex)) {
394
395 if (inet6_test_bit(RTALERT_ISOLATE, sk) &&
396 !net_eq(sock_net(sk), dev_net(skb->dev))) {
397 continue;
398 }
399 if (last) {
400 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
401 if (skb2)
402 rawv6_rcv(last, skb2);
403 }
404 last = sk;
405 }
406 }
407
408 if (last) {
409 rawv6_rcv(last, skb);
410 read_unlock(&ip6_ra_lock);
411 return 1;
412 }
413 read_unlock(&ip6_ra_lock);
414 return 0;
415}
416
417static int ip6_forward_proxy_check(struct sk_buff *skb)
418{
419 struct ipv6hdr *hdr = ipv6_hdr(skb);
420 u8 nexthdr = hdr->nexthdr;
421 __be16 frag_off;
422 int offset;
423
424 if (ipv6_ext_hdr(nexthdr)) {
425 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
426 if (offset < 0)
427 return 0;
428 } else
429 offset = sizeof(struct ipv6hdr);
430
431 if (nexthdr == IPPROTO_ICMPV6) {
432 struct icmp6hdr *icmp6;
433
434 if (!pskb_may_pull(skb, (skb_network_header(skb) +
435 offset + 1 - skb->data)))
436 return 0;
437
438 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
439
440 switch (icmp6->icmp6_type) {
441 case NDISC_ROUTER_SOLICITATION:
442 case NDISC_ROUTER_ADVERTISEMENT:
443 case NDISC_NEIGHBOUR_SOLICITATION:
444 case NDISC_NEIGHBOUR_ADVERTISEMENT:
445 case NDISC_REDIRECT:
446 /* For reaction involving unicast neighbor discovery
447 * message destined to the proxied address, pass it to
448 * input function.
449 */
450 return 1;
451 default:
452 break;
453 }
454 }
455
456 /*
457 * The proxying router can't forward traffic sent to a link-local
458 * address, so signal the sender and discard the packet. This
459 * behavior is clarified by the MIPv6 specification.
460 */
461 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
462 dst_link_failure(skb);
463 return -1;
464 }
465
466 return 0;
467}
468
469static inline int ip6_forward_finish(struct net *net, struct sock *sk,
470 struct sk_buff *skb)
471{
472#ifdef CONFIG_NET_SWITCHDEV
473 if (skb->offload_l3_fwd_mark) {
474 consume_skb(skb);
475 return 0;
476 }
477#endif
478
479 skb_clear_tstamp(skb);
480 return dst_output(net, sk, skb);
481}
482
483static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
484{
485 if (skb->len <= mtu)
486 return false;
487
488 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
489 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
490 return true;
491
492 if (skb->ignore_df)
493 return false;
494
495 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
496 return false;
497
498 return true;
499}
500
501int ip6_forward(struct sk_buff *skb)
502{
503 struct dst_entry *dst = skb_dst(skb);
504 struct ipv6hdr *hdr = ipv6_hdr(skb);
505 struct inet6_skb_parm *opt = IP6CB(skb);
506 struct net *net = dev_net(dst->dev);
507 struct inet6_dev *idev;
508 SKB_DR(reason);
509 u32 mtu;
510
511 idev = __in6_dev_get_safely(dev_get_by_index_rcu(net, IP6CB(skb)->iif));
512 if (READ_ONCE(net->ipv6.devconf_all->forwarding) == 0)
513 goto error;
514
515 if (skb->pkt_type != PACKET_HOST)
516 goto drop;
517
518 if (unlikely(skb->sk))
519 goto drop;
520
521 if (skb_warn_if_lro(skb))
522 goto drop;
523
524 if (!READ_ONCE(net->ipv6.devconf_all->disable_policy) &&
525 (!idev || !READ_ONCE(idev->cnf.disable_policy)) &&
526 !xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
527 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
528 goto drop;
529 }
530
531 skb_forward_csum(skb);
532
533 /*
534 * We DO NOT make any processing on
535 * RA packets, pushing them to user level AS IS
536 * without ane WARRANTY that application will be able
537 * to interpret them. The reason is that we
538 * cannot make anything clever here.
539 *
540 * We are not end-node, so that if packet contains
541 * AH/ESP, we cannot make anything.
542 * Defragmentation also would be mistake, RA packets
543 * cannot be fragmented, because there is no warranty
544 * that different fragments will go along one path. --ANK
545 */
546 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
547 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
548 return 0;
549 }
550
551 /*
552 * check and decrement ttl
553 */
554 if (hdr->hop_limit <= 1) {
555 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
556 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
557
558 kfree_skb_reason(skb, SKB_DROP_REASON_IP_INHDR);
559 return -ETIMEDOUT;
560 }
561
562 /* XXX: idev->cnf.proxy_ndp? */
563 if (READ_ONCE(net->ipv6.devconf_all->proxy_ndp) &&
564 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
565 int proxied = ip6_forward_proxy_check(skb);
566 if (proxied > 0) {
567 /* It's tempting to decrease the hop limit
568 * here by 1, as we do at the end of the
569 * function too.
570 *
571 * But that would be incorrect, as proxying is
572 * not forwarding. The ip6_input function
573 * will handle this packet locally, and it
574 * depends on the hop limit being unchanged.
575 *
576 * One example is the NDP hop limit, that
577 * always has to stay 255, but other would be
578 * similar checks around RA packets, where the
579 * user can even change the desired limit.
580 */
581 return ip6_input(skb);
582 } else if (proxied < 0) {
583 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
584 goto drop;
585 }
586 }
587
588 if (!xfrm6_route_forward(skb)) {
589 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INDISCARDS);
590 SKB_DR_SET(reason, XFRM_POLICY);
591 goto drop;
592 }
593 dst = skb_dst(skb);
594
595 /* IPv6 specs say nothing about it, but it is clear that we cannot
596 send redirects to source routed frames.
597 We don't send redirects to frames decapsulated from IPsec.
598 */
599 if (IP6CB(skb)->iif == dst->dev->ifindex &&
600 opt->srcrt == 0 && !skb_sec_path(skb)) {
601 struct in6_addr *target = NULL;
602 struct inet_peer *peer;
603 struct rt6_info *rt;
604
605 /*
606 * incoming and outgoing devices are the same
607 * send a redirect.
608 */
609
610 rt = dst_rt6_info(dst);
611 if (rt->rt6i_flags & RTF_GATEWAY)
612 target = &rt->rt6i_gateway;
613 else
614 target = &hdr->daddr;
615
616 rcu_read_lock();
617 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr);
618
619 /* Limit redirects both by destination (here)
620 and by source (inside ndisc_send_redirect)
621 */
622 if (inet_peer_xrlim_allow(peer, 1*HZ))
623 ndisc_send_redirect(skb, target);
624 rcu_read_unlock();
625 } else {
626 int addrtype = ipv6_addr_type(&hdr->saddr);
627
628 /* This check is security critical. */
629 if (addrtype == IPV6_ADDR_ANY ||
630 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
631 goto error;
632 if (addrtype & IPV6_ADDR_LINKLOCAL) {
633 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
634 ICMPV6_NOT_NEIGHBOUR, 0);
635 goto error;
636 }
637 }
638
639 __IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
640
641 mtu = ip6_dst_mtu_maybe_forward(dst, true);
642 if (mtu < IPV6_MIN_MTU)
643 mtu = IPV6_MIN_MTU;
644
645 if (ip6_pkt_too_big(skb, mtu)) {
646 /* Again, force OUTPUT device used as source address */
647 skb->dev = dst->dev;
648 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
649 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INTOOBIGERRORS);
650 __IP6_INC_STATS(net, ip6_dst_idev(dst),
651 IPSTATS_MIB_FRAGFAILS);
652 kfree_skb_reason(skb, SKB_DROP_REASON_PKT_TOO_BIG);
653 return -EMSGSIZE;
654 }
655
656 if (skb_cow(skb, dst->dev->hard_header_len)) {
657 __IP6_INC_STATS(net, ip6_dst_idev(dst),
658 IPSTATS_MIB_OUTDISCARDS);
659 goto drop;
660 }
661
662 hdr = ipv6_hdr(skb);
663
664 /* Mangling hops number delayed to point after skb COW */
665
666 hdr->hop_limit--;
667
668 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
669 net, NULL, skb, skb->dev, dst->dev,
670 ip6_forward_finish);
671
672error:
673 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INADDRERRORS);
674 SKB_DR_SET(reason, IP_INADDRERRORS);
675drop:
676 kfree_skb_reason(skb, reason);
677 return -EINVAL;
678}
679
680static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
681{
682 to->pkt_type = from->pkt_type;
683 to->priority = from->priority;
684 to->protocol = from->protocol;
685 skb_dst_drop(to);
686 skb_dst_set(to, dst_clone(skb_dst(from)));
687 to->dev = from->dev;
688 to->mark = from->mark;
689
690 skb_copy_hash(to, from);
691
692#ifdef CONFIG_NET_SCHED
693 to->tc_index = from->tc_index;
694#endif
695 nf_copy(to, from);
696 skb_ext_copy(to, from);
697 skb_copy_secmark(to, from);
698}
699
700int ip6_fraglist_init(struct sk_buff *skb, unsigned int hlen, u8 *prevhdr,
701 u8 nexthdr, __be32 frag_id,
702 struct ip6_fraglist_iter *iter)
703{
704 unsigned int first_len;
705 struct frag_hdr *fh;
706
707 /* BUILD HEADER */
708 *prevhdr = NEXTHDR_FRAGMENT;
709 iter->tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
710 if (!iter->tmp_hdr)
711 return -ENOMEM;
712
713 iter->frag = skb_shinfo(skb)->frag_list;
714 skb_frag_list_init(skb);
715
716 iter->offset = 0;
717 iter->hlen = hlen;
718 iter->frag_id = frag_id;
719 iter->nexthdr = nexthdr;
720
721 __skb_pull(skb, hlen);
722 fh = __skb_push(skb, sizeof(struct frag_hdr));
723 __skb_push(skb, hlen);
724 skb_reset_network_header(skb);
725 memcpy(skb_network_header(skb), iter->tmp_hdr, hlen);
726
727 fh->nexthdr = nexthdr;
728 fh->reserved = 0;
729 fh->frag_off = htons(IP6_MF);
730 fh->identification = frag_id;
731
732 first_len = skb_pagelen(skb);
733 skb->data_len = first_len - skb_headlen(skb);
734 skb->len = first_len;
735 ipv6_hdr(skb)->payload_len = htons(first_len - sizeof(struct ipv6hdr));
736
737 return 0;
738}
739EXPORT_SYMBOL(ip6_fraglist_init);
740
741void ip6_fraglist_prepare(struct sk_buff *skb,
742 struct ip6_fraglist_iter *iter)
743{
744 struct sk_buff *frag = iter->frag;
745 unsigned int hlen = iter->hlen;
746 struct frag_hdr *fh;
747
748 frag->ip_summed = CHECKSUM_NONE;
749 skb_reset_transport_header(frag);
750 fh = __skb_push(frag, sizeof(struct frag_hdr));
751 __skb_push(frag, hlen);
752 skb_reset_network_header(frag);
753 memcpy(skb_network_header(frag), iter->tmp_hdr, hlen);
754 iter->offset += skb->len - hlen - sizeof(struct frag_hdr);
755 fh->nexthdr = iter->nexthdr;
756 fh->reserved = 0;
757 fh->frag_off = htons(iter->offset);
758 if (frag->next)
759 fh->frag_off |= htons(IP6_MF);
760 fh->identification = iter->frag_id;
761 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
762 ip6_copy_metadata(frag, skb);
763}
764EXPORT_SYMBOL(ip6_fraglist_prepare);
765
766void ip6_frag_init(struct sk_buff *skb, unsigned int hlen, unsigned int mtu,
767 unsigned short needed_tailroom, int hdr_room, u8 *prevhdr,
768 u8 nexthdr, __be32 frag_id, struct ip6_frag_state *state)
769{
770 state->prevhdr = prevhdr;
771 state->nexthdr = nexthdr;
772 state->frag_id = frag_id;
773
774 state->hlen = hlen;
775 state->mtu = mtu;
776
777 state->left = skb->len - hlen; /* Space per frame */
778 state->ptr = hlen; /* Where to start from */
779
780 state->hroom = hdr_room;
781 state->troom = needed_tailroom;
782
783 state->offset = 0;
784}
785EXPORT_SYMBOL(ip6_frag_init);
786
787struct sk_buff *ip6_frag_next(struct sk_buff *skb, struct ip6_frag_state *state)
788{
789 u8 *prevhdr = state->prevhdr, *fragnexthdr_offset;
790 struct sk_buff *frag;
791 struct frag_hdr *fh;
792 unsigned int len;
793
794 len = state->left;
795 /* IF: it doesn't fit, use 'mtu' - the data space left */
796 if (len > state->mtu)
797 len = state->mtu;
798 /* IF: we are not sending up to and including the packet end
799 then align the next start on an eight byte boundary */
800 if (len < state->left)
801 len &= ~7;
802
803 /* Allocate buffer */
804 frag = alloc_skb(len + state->hlen + sizeof(struct frag_hdr) +
805 state->hroom + state->troom, GFP_ATOMIC);
806 if (!frag)
807 return ERR_PTR(-ENOMEM);
808
809 /*
810 * Set up data on packet
811 */
812
813 ip6_copy_metadata(frag, skb);
814 skb_reserve(frag, state->hroom);
815 skb_put(frag, len + state->hlen + sizeof(struct frag_hdr));
816 skb_reset_network_header(frag);
817 fh = (struct frag_hdr *)(skb_network_header(frag) + state->hlen);
818 frag->transport_header = (frag->network_header + state->hlen +
819 sizeof(struct frag_hdr));
820
821 /*
822 * Charge the memory for the fragment to any owner
823 * it might possess
824 */
825 if (skb->sk)
826 skb_set_owner_w(frag, skb->sk);
827
828 /*
829 * Copy the packet header into the new buffer.
830 */
831 skb_copy_from_linear_data(skb, skb_network_header(frag), state->hlen);
832
833 fragnexthdr_offset = skb_network_header(frag);
834 fragnexthdr_offset += prevhdr - skb_network_header(skb);
835 *fragnexthdr_offset = NEXTHDR_FRAGMENT;
836
837 /*
838 * Build fragment header.
839 */
840 fh->nexthdr = state->nexthdr;
841 fh->reserved = 0;
842 fh->identification = state->frag_id;
843
844 /*
845 * Copy a block of the IP datagram.
846 */
847 BUG_ON(skb_copy_bits(skb, state->ptr, skb_transport_header(frag),
848 len));
849 state->left -= len;
850
851 fh->frag_off = htons(state->offset);
852 if (state->left > 0)
853 fh->frag_off |= htons(IP6_MF);
854 ipv6_hdr(frag)->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
855
856 state->ptr += len;
857 state->offset += len;
858
859 return frag;
860}
861EXPORT_SYMBOL(ip6_frag_next);
862
863int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
864 int (*output)(struct net *, struct sock *, struct sk_buff *))
865{
866 struct sk_buff *frag;
867 struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
868 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
869 inet6_sk(skb->sk) : NULL;
870 u8 tstamp_type = skb->tstamp_type;
871 struct ip6_frag_state state;
872 unsigned int mtu, hlen, nexthdr_offset;
873 ktime_t tstamp = skb->tstamp;
874 int hroom, err = 0;
875 __be32 frag_id;
876 u8 *prevhdr, nexthdr = 0;
877
878 err = ip6_find_1stfragopt(skb, &prevhdr);
879 if (err < 0)
880 goto fail;
881 hlen = err;
882 nexthdr = *prevhdr;
883 nexthdr_offset = prevhdr - skb_network_header(skb);
884
885 mtu = ip6_skb_dst_mtu(skb);
886
887 /* We must not fragment if the socket is set to force MTU discovery
888 * or if the skb it not generated by a local socket.
889 */
890 if (unlikely(!skb->ignore_df && skb->len > mtu))
891 goto fail_toobig;
892
893 if (IP6CB(skb)->frag_max_size) {
894 if (IP6CB(skb)->frag_max_size > mtu)
895 goto fail_toobig;
896
897 /* don't send fragments larger than what we received */
898 mtu = IP6CB(skb)->frag_max_size;
899 if (mtu < IPV6_MIN_MTU)
900 mtu = IPV6_MIN_MTU;
901 }
902
903 if (np) {
904 u32 frag_size = READ_ONCE(np->frag_size);
905
906 if (frag_size && frag_size < mtu)
907 mtu = frag_size;
908 }
909 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
910 goto fail_toobig;
911 mtu -= hlen + sizeof(struct frag_hdr);
912
913 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
914 &ipv6_hdr(skb)->saddr);
915
916 if (skb->ip_summed == CHECKSUM_PARTIAL &&
917 (err = skb_checksum_help(skb)))
918 goto fail;
919
920 prevhdr = skb_network_header(skb) + nexthdr_offset;
921 hroom = LL_RESERVED_SPACE(rt->dst.dev);
922 if (skb_has_frag_list(skb)) {
923 unsigned int first_len = skb_pagelen(skb);
924 struct ip6_fraglist_iter iter;
925 struct sk_buff *frag2;
926
927 if (first_len - hlen > mtu ||
928 ((first_len - hlen) & 7) ||
929 skb_cloned(skb) ||
930 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
931 goto slow_path;
932
933 skb_walk_frags(skb, frag) {
934 /* Correct geometry. */
935 if (frag->len > mtu ||
936 ((frag->len & 7) && frag->next) ||
937 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
938 goto slow_path_clean;
939
940 /* Partially cloned skb? */
941 if (skb_shared(frag))
942 goto slow_path_clean;
943
944 BUG_ON(frag->sk);
945 if (skb->sk) {
946 frag->sk = skb->sk;
947 frag->destructor = sock_wfree;
948 }
949 skb->truesize -= frag->truesize;
950 }
951
952 err = ip6_fraglist_init(skb, hlen, prevhdr, nexthdr, frag_id,
953 &iter);
954 if (err < 0)
955 goto fail;
956
957 /* We prevent @rt from being freed. */
958 rcu_read_lock();
959
960 for (;;) {
961 /* Prepare header of the next frame,
962 * before previous one went down. */
963 if (iter.frag)
964 ip6_fraglist_prepare(skb, &iter);
965
966 skb_set_delivery_time(skb, tstamp, tstamp_type);
967 err = output(net, sk, skb);
968 if (!err)
969 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
970 IPSTATS_MIB_FRAGCREATES);
971
972 if (err || !iter.frag)
973 break;
974
975 skb = ip6_fraglist_next(&iter);
976 }
977
978 kfree(iter.tmp_hdr);
979
980 if (err == 0) {
981 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
982 IPSTATS_MIB_FRAGOKS);
983 rcu_read_unlock();
984 return 0;
985 }
986
987 kfree_skb_list(iter.frag);
988
989 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
990 IPSTATS_MIB_FRAGFAILS);
991 rcu_read_unlock();
992 return err;
993
994slow_path_clean:
995 skb_walk_frags(skb, frag2) {
996 if (frag2 == frag)
997 break;
998 frag2->sk = NULL;
999 frag2->destructor = NULL;
1000 skb->truesize += frag2->truesize;
1001 }
1002 }
1003
1004slow_path:
1005 /*
1006 * Fragment the datagram.
1007 */
1008
1009 ip6_frag_init(skb, hlen, mtu, rt->dst.dev->needed_tailroom,
1010 LL_RESERVED_SPACE(rt->dst.dev), prevhdr, nexthdr, frag_id,
1011 &state);
1012
1013 /*
1014 * Keep copying data until we run out.
1015 */
1016
1017 while (state.left > 0) {
1018 frag = ip6_frag_next(skb, &state);
1019 if (IS_ERR(frag)) {
1020 err = PTR_ERR(frag);
1021 goto fail;
1022 }
1023
1024 /*
1025 * Put this fragment into the sending queue.
1026 */
1027 skb_set_delivery_time(frag, tstamp, tstamp_type);
1028 err = output(net, sk, frag);
1029 if (err)
1030 goto fail;
1031
1032 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1033 IPSTATS_MIB_FRAGCREATES);
1034 }
1035 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1036 IPSTATS_MIB_FRAGOKS);
1037 consume_skb(skb);
1038 return err;
1039
1040fail_toobig:
1041 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
1042 err = -EMSGSIZE;
1043
1044fail:
1045 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
1046 IPSTATS_MIB_FRAGFAILS);
1047 kfree_skb(skb);
1048 return err;
1049}
1050
1051static inline int ip6_rt_check(const struct rt6key *rt_key,
1052 const struct in6_addr *fl_addr,
1053 const struct in6_addr *addr_cache)
1054{
1055 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
1056 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
1057}
1058
1059static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
1060 struct dst_entry *dst,
1061 const struct flowi6 *fl6)
1062{
1063 struct ipv6_pinfo *np = inet6_sk(sk);
1064 struct rt6_info *rt;
1065
1066 if (!dst)
1067 goto out;
1068
1069 if (dst->ops->family != AF_INET6) {
1070 dst_release(dst);
1071 return NULL;
1072 }
1073
1074 rt = dst_rt6_info(dst);
1075 /* Yes, checking route validity in not connected
1076 * case is not very simple. Take into account,
1077 * that we do not support routing by source, TOS,
1078 * and MSG_DONTROUTE --ANK (980726)
1079 *
1080 * 1. ip6_rt_check(): If route was host route,
1081 * check that cached destination is current.
1082 * If it is network route, we still may
1083 * check its validity using saved pointer
1084 * to the last used address: daddr_cache.
1085 * We do not want to save whole address now,
1086 * (because main consumer of this service
1087 * is tcp, which has not this problem),
1088 * so that the last trick works only on connected
1089 * sockets.
1090 * 2. oif also should be the same.
1091 */
1092 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
1093#ifdef CONFIG_IPV6_SUBTREES
1094 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
1095#endif
1096 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex)) {
1097 dst_release(dst);
1098 dst = NULL;
1099 }
1100
1101out:
1102 return dst;
1103}
1104
1105static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
1106 struct dst_entry **dst, struct flowi6 *fl6)
1107{
1108#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1109 struct neighbour *n;
1110 struct rt6_info *rt;
1111#endif
1112 int err;
1113 int flags = 0;
1114
1115 /* The correct way to handle this would be to do
1116 * ip6_route_get_saddr, and then ip6_route_output; however,
1117 * the route-specific preferred source forces the
1118 * ip6_route_output call _before_ ip6_route_get_saddr.
1119 *
1120 * In source specific routing (no src=any default route),
1121 * ip6_route_output will fail given src=any saddr, though, so
1122 * that's why we try it again later.
1123 */
1124 if (ipv6_addr_any(&fl6->saddr)) {
1125 struct fib6_info *from;
1126 struct rt6_info *rt;
1127
1128 *dst = ip6_route_output(net, sk, fl6);
1129 rt = (*dst)->error ? NULL : dst_rt6_info(*dst);
1130
1131 rcu_read_lock();
1132 from = rt ? rcu_dereference(rt->from) : NULL;
1133 err = ip6_route_get_saddr(net, from, &fl6->daddr,
1134 sk ? READ_ONCE(inet6_sk(sk)->srcprefs) : 0,
1135 fl6->flowi6_l3mdev,
1136 &fl6->saddr);
1137 rcu_read_unlock();
1138
1139 if (err)
1140 goto out_err_release;
1141
1142 /* If we had an erroneous initial result, pretend it
1143 * never existed and let the SA-enabled version take
1144 * over.
1145 */
1146 if ((*dst)->error) {
1147 dst_release(*dst);
1148 *dst = NULL;
1149 }
1150
1151 if (fl6->flowi6_oif)
1152 flags |= RT6_LOOKUP_F_IFACE;
1153 }
1154
1155 if (!*dst)
1156 *dst = ip6_route_output_flags(net, sk, fl6, flags);
1157
1158 err = (*dst)->error;
1159 if (err)
1160 goto out_err_release;
1161
1162#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
1163 /*
1164 * Here if the dst entry we've looked up
1165 * has a neighbour entry that is in the INCOMPLETE
1166 * state and the src address from the flow is
1167 * marked as OPTIMISTIC, we release the found
1168 * dst entry and replace it instead with the
1169 * dst entry of the nexthop router
1170 */
1171 rt = dst_rt6_info(*dst);
1172 rcu_read_lock();
1173 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
1174 rt6_nexthop(rt, &fl6->daddr));
1175 err = n && !(READ_ONCE(n->nud_state) & NUD_VALID) ? -EINVAL : 0;
1176 rcu_read_unlock();
1177
1178 if (err) {
1179 struct inet6_ifaddr *ifp;
1180 struct flowi6 fl_gw6;
1181 int redirect;
1182
1183 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
1184 (*dst)->dev, 1);
1185
1186 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
1187 if (ifp)
1188 in6_ifa_put(ifp);
1189
1190 if (redirect) {
1191 /*
1192 * We need to get the dst entry for the
1193 * default router instead
1194 */
1195 dst_release(*dst);
1196 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
1197 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
1198 *dst = ip6_route_output(net, sk, &fl_gw6);
1199 err = (*dst)->error;
1200 if (err)
1201 goto out_err_release;
1202 }
1203 }
1204#endif
1205 if (ipv6_addr_v4mapped(&fl6->saddr) &&
1206 !(ipv6_addr_v4mapped(&fl6->daddr) || ipv6_addr_any(&fl6->daddr))) {
1207 err = -EAFNOSUPPORT;
1208 goto out_err_release;
1209 }
1210
1211 return 0;
1212
1213out_err_release:
1214 dst_release(*dst);
1215 *dst = NULL;
1216
1217 if (err == -ENETUNREACH)
1218 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1219 return err;
1220}
1221
1222/**
1223 * ip6_dst_lookup - perform route lookup on flow
1224 * @net: Network namespace to perform lookup in
1225 * @sk: socket which provides route info
1226 * @dst: pointer to dst_entry * for result
1227 * @fl6: flow to lookup
1228 *
1229 * This function performs a route lookup on the given flow.
1230 *
1231 * It returns zero on success, or a standard errno code on error.
1232 */
1233int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1234 struct flowi6 *fl6)
1235{
1236 *dst = NULL;
1237 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1238}
1239EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1240
1241/**
1242 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1243 * @net: Network namespace to perform lookup in
1244 * @sk: socket which provides route info
1245 * @fl6: flow to lookup
1246 * @final_dst: final destination address for ipsec lookup
1247 *
1248 * This function performs a route lookup on the given flow.
1249 *
1250 * It returns a valid dst pointer on success, or a pointer encoded
1251 * error code.
1252 */
1253struct dst_entry *ip6_dst_lookup_flow(struct net *net, const struct sock *sk, struct flowi6 *fl6,
1254 const struct in6_addr *final_dst)
1255{
1256 struct dst_entry *dst = NULL;
1257 int err;
1258
1259 err = ip6_dst_lookup_tail(net, sk, &dst, fl6);
1260 if (err)
1261 return ERR_PTR(err);
1262 if (final_dst)
1263 fl6->daddr = *final_dst;
1264
1265 return xfrm_lookup_route(net, dst, flowi6_to_flowi(fl6), sk, 0);
1266}
1267EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1268
1269/**
1270 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1271 * @sk: socket which provides the dst cache and route info
1272 * @fl6: flow to lookup
1273 * @final_dst: final destination address for ipsec lookup
1274 * @connected: whether @sk is connected or not
1275 *
1276 * This function performs a route lookup on the given flow with the
1277 * possibility of using the cached route in the socket if it is valid.
1278 * It will take the socket dst lock when operating on the dst cache.
1279 * As a result, this function can only be used in process context.
1280 *
1281 * In addition, for a connected socket, cache the dst in the socket
1282 * if the current cache is not valid.
1283 *
1284 * It returns a valid dst pointer on success, or a pointer encoded
1285 * error code.
1286 */
1287struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1288 const struct in6_addr *final_dst,
1289 bool connected)
1290{
1291 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1292
1293 dst = ip6_sk_dst_check(sk, dst, fl6);
1294 if (dst)
1295 return dst;
1296
1297 dst = ip6_dst_lookup_flow(sock_net(sk), sk, fl6, final_dst);
1298 if (connected && !IS_ERR(dst))
1299 ip6_sk_dst_store_flow(sk, dst_clone(dst), fl6);
1300
1301 return dst;
1302}
1303EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1304
1305static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1306 gfp_t gfp)
1307{
1308 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1309}
1310
1311static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1312 gfp_t gfp)
1313{
1314 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1315}
1316
1317static void ip6_append_data_mtu(unsigned int *mtu,
1318 int *maxfraglen,
1319 unsigned int fragheaderlen,
1320 struct sk_buff *skb,
1321 struct rt6_info *rt,
1322 unsigned int orig_mtu)
1323{
1324 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1325 if (!skb) {
1326 /* first fragment, reserve header_len */
1327 *mtu = orig_mtu - rt->dst.header_len;
1328
1329 } else {
1330 /*
1331 * this fragment is not first, the headers
1332 * space is regarded as data space.
1333 */
1334 *mtu = orig_mtu;
1335 }
1336 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1337 + fragheaderlen - sizeof(struct frag_hdr);
1338 }
1339}
1340
1341static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1342 struct inet6_cork *v6_cork, struct ipcm6_cookie *ipc6,
1343 struct rt6_info *rt)
1344{
1345 struct ipv6_pinfo *np = inet6_sk(sk);
1346 unsigned int mtu, frag_size;
1347 struct ipv6_txoptions *nopt, *opt = ipc6->opt;
1348
1349 /* callers pass dst together with a reference, set it first so
1350 * ip6_cork_release() can put it down even in case of an error.
1351 */
1352 cork->base.dst = &rt->dst;
1353
1354 /*
1355 * setup for corking
1356 */
1357 if (opt) {
1358 if (WARN_ON(v6_cork->opt))
1359 return -EINVAL;
1360
1361 nopt = v6_cork->opt = kzalloc(sizeof(*opt), sk->sk_allocation);
1362 if (unlikely(!nopt))
1363 return -ENOBUFS;
1364
1365 nopt->tot_len = sizeof(*opt);
1366 nopt->opt_flen = opt->opt_flen;
1367 nopt->opt_nflen = opt->opt_nflen;
1368
1369 nopt->dst0opt = ip6_opt_dup(opt->dst0opt, sk->sk_allocation);
1370 if (opt->dst0opt && !nopt->dst0opt)
1371 return -ENOBUFS;
1372
1373 nopt->dst1opt = ip6_opt_dup(opt->dst1opt, sk->sk_allocation);
1374 if (opt->dst1opt && !nopt->dst1opt)
1375 return -ENOBUFS;
1376
1377 nopt->hopopt = ip6_opt_dup(opt->hopopt, sk->sk_allocation);
1378 if (opt->hopopt && !nopt->hopopt)
1379 return -ENOBUFS;
1380
1381 nopt->srcrt = ip6_rthdr_dup(opt->srcrt, sk->sk_allocation);
1382 if (opt->srcrt && !nopt->srcrt)
1383 return -ENOBUFS;
1384
1385 /* need source address above miyazawa*/
1386 }
1387 v6_cork->hop_limit = ipc6->hlimit;
1388 v6_cork->tclass = ipc6->tclass;
1389 if (rt->dst.flags & DST_XFRM_TUNNEL)
1390 mtu = READ_ONCE(np->pmtudisc) >= IPV6_PMTUDISC_PROBE ?
1391 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(&rt->dst);
1392 else
1393 mtu = READ_ONCE(np->pmtudisc) >= IPV6_PMTUDISC_PROBE ?
1394 READ_ONCE(rt->dst.dev->mtu) : dst_mtu(xfrm_dst_path(&rt->dst));
1395
1396 frag_size = READ_ONCE(np->frag_size);
1397 if (frag_size && frag_size < mtu)
1398 mtu = frag_size;
1399
1400 cork->base.fragsize = mtu;
1401 cork->base.gso_size = ipc6->gso_size;
1402 cork->base.tx_flags = 0;
1403 cork->base.mark = ipc6->sockc.mark;
1404 sock_tx_timestamp(sk, &ipc6->sockc, &cork->base.tx_flags);
1405 if (ipc6->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1406 cork->base.flags |= IPCORK_TS_OPT_ID;
1407 cork->base.ts_opt_id = ipc6->sockc.ts_opt_id;
1408 }
1409 cork->base.length = 0;
1410 cork->base.transmit_time = ipc6->sockc.transmit_time;
1411
1412 return 0;
1413}
1414
1415static int __ip6_append_data(struct sock *sk,
1416 struct sk_buff_head *queue,
1417 struct inet_cork_full *cork_full,
1418 struct inet6_cork *v6_cork,
1419 struct page_frag *pfrag,
1420 int getfrag(void *from, char *to, int offset,
1421 int len, int odd, struct sk_buff *skb),
1422 void *from, size_t length, int transhdrlen,
1423 unsigned int flags, struct ipcm6_cookie *ipc6)
1424{
1425 struct sk_buff *skb, *skb_prev = NULL;
1426 struct inet_cork *cork = &cork_full->base;
1427 struct flowi6 *fl6 = &cork_full->fl.u.ip6;
1428 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
1429 struct ubuf_info *uarg = NULL;
1430 int exthdrlen = 0;
1431 int dst_exthdrlen = 0;
1432 int hh_len;
1433 int copy;
1434 int err;
1435 int offset = 0;
1436 bool zc = false;
1437 u32 tskey = 0;
1438 struct rt6_info *rt = dst_rt6_info(cork->dst);
1439 bool paged, hold_tskey = false, extra_uref = false;
1440 struct ipv6_txoptions *opt = v6_cork->opt;
1441 int csummode = CHECKSUM_NONE;
1442 unsigned int maxnonfragsize, headersize;
1443 unsigned int wmem_alloc_delta = 0;
1444
1445 skb = skb_peek_tail(queue);
1446 if (!skb) {
1447 exthdrlen = opt ? opt->opt_flen : 0;
1448 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1449 }
1450
1451 paged = !!cork->gso_size;
1452 mtu = cork->gso_size ? IP6_MAX_MTU : cork->fragsize;
1453 orig_mtu = mtu;
1454
1455 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1456
1457 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1458 (opt ? opt->opt_nflen : 0);
1459
1460 headersize = sizeof(struct ipv6hdr) +
1461 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1462 rt->rt6i_nfheader_len;
1463
1464 if (mtu <= fragheaderlen ||
1465 ((mtu - fragheaderlen) & ~7) + fragheaderlen <= sizeof(struct frag_hdr))
1466 goto emsgsize;
1467
1468 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1469 sizeof(struct frag_hdr);
1470
1471 /* as per RFC 7112 section 5, the entire IPv6 Header Chain must fit
1472 * the first fragment
1473 */
1474 if (headersize + transhdrlen > mtu)
1475 goto emsgsize;
1476
1477 if (cork->length + length > mtu - headersize && ipc6->dontfrag &&
1478 (sk->sk_protocol == IPPROTO_UDP ||
1479 sk->sk_protocol == IPPROTO_ICMPV6 ||
1480 sk->sk_protocol == IPPROTO_RAW)) {
1481 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1482 sizeof(struct ipv6hdr));
1483 goto emsgsize;
1484 }
1485
1486 if (ip6_sk_ignore_df(sk))
1487 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1488 else
1489 maxnonfragsize = mtu;
1490
1491 if (cork->length + length > maxnonfragsize - headersize) {
1492emsgsize:
1493 pmtu = max_t(int, mtu - headersize + sizeof(struct ipv6hdr), 0);
1494 ipv6_local_error(sk, EMSGSIZE, fl6, pmtu);
1495 return -EMSGSIZE;
1496 }
1497
1498 /* CHECKSUM_PARTIAL only with no extension headers and when
1499 * we are not going to fragment
1500 */
1501 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1502 headersize == sizeof(struct ipv6hdr) &&
1503 length <= mtu - headersize &&
1504 (!(flags & MSG_MORE) || cork->gso_size) &&
1505 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1506 csummode = CHECKSUM_PARTIAL;
1507
1508 if ((flags & MSG_ZEROCOPY) && length) {
1509 struct msghdr *msg = from;
1510
1511 if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1512 if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1513 return -EINVAL;
1514
1515 /* Leave uarg NULL if can't zerocopy, callers should
1516 * be able to handle it.
1517 */
1518 if ((rt->dst.dev->features & NETIF_F_SG) &&
1519 csummode == CHECKSUM_PARTIAL) {
1520 paged = true;
1521 zc = true;
1522 uarg = msg->msg_ubuf;
1523 }
1524 } else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1525 uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1526 if (!uarg)
1527 return -ENOBUFS;
1528 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1529 if (rt->dst.dev->features & NETIF_F_SG &&
1530 csummode == CHECKSUM_PARTIAL) {
1531 paged = true;
1532 zc = true;
1533 } else {
1534 uarg_to_msgzc(uarg)->zerocopy = 0;
1535 skb_zcopy_set(skb, uarg, &extra_uref);
1536 }
1537 }
1538 } else if ((flags & MSG_SPLICE_PAGES) && length) {
1539 if (inet_test_bit(HDRINCL, sk))
1540 return -EPERM;
1541 if (rt->dst.dev->features & NETIF_F_SG &&
1542 getfrag == ip_generic_getfrag)
1543 /* We need an empty buffer to attach stuff to */
1544 paged = true;
1545 else
1546 flags &= ~MSG_SPLICE_PAGES;
1547 }
1548
1549 if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1550 READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1551 if (cork->flags & IPCORK_TS_OPT_ID) {
1552 tskey = cork->ts_opt_id;
1553 } else {
1554 tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1555 hold_tskey = true;
1556 }
1557 }
1558
1559 /*
1560 * Let's try using as much space as possible.
1561 * Use MTU if total length of the message fits into the MTU.
1562 * Otherwise, we need to reserve fragment header and
1563 * fragment alignment (= 8-15 octects, in total).
1564 *
1565 * Note that we may need to "move" the data from the tail
1566 * of the buffer to the new fragment when we split
1567 * the message.
1568 *
1569 * FIXME: It may be fragmented into multiple chunks
1570 * at once if non-fragmentable extension headers
1571 * are too large.
1572 * --yoshfuji
1573 */
1574
1575 cork->length += length;
1576 if (!skb)
1577 goto alloc_new_skb;
1578
1579 while (length > 0) {
1580 /* Check if the remaining data fits into current packet. */
1581 copy = (cork->length <= mtu ? mtu : maxfraglen) - skb->len;
1582 if (copy < length)
1583 copy = maxfraglen - skb->len;
1584
1585 if (copy <= 0) {
1586 char *data;
1587 unsigned int datalen;
1588 unsigned int fraglen;
1589 unsigned int fraggap;
1590 unsigned int alloclen, alloc_extra;
1591 unsigned int pagedlen;
1592alloc_new_skb:
1593 /* There's no room in the current skb */
1594 if (skb)
1595 fraggap = skb->len - maxfraglen;
1596 else
1597 fraggap = 0;
1598 /* update mtu and maxfraglen if necessary */
1599 if (!skb || !skb_prev)
1600 ip6_append_data_mtu(&mtu, &maxfraglen,
1601 fragheaderlen, skb, rt,
1602 orig_mtu);
1603
1604 skb_prev = skb;
1605
1606 /*
1607 * If remaining data exceeds the mtu,
1608 * we know we need more fragment(s).
1609 */
1610 datalen = length + fraggap;
1611
1612 if (datalen > (cork->length <= mtu ? mtu : maxfraglen) - fragheaderlen)
1613 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1614 fraglen = datalen + fragheaderlen;
1615 pagedlen = 0;
1616
1617 alloc_extra = hh_len;
1618 alloc_extra += dst_exthdrlen;
1619 alloc_extra += rt->dst.trailer_len;
1620
1621 /* We just reserve space for fragment header.
1622 * Note: this may be overallocation if the message
1623 * (without MSG_MORE) fits into the MTU.
1624 */
1625 alloc_extra += sizeof(struct frag_hdr);
1626
1627 if ((flags & MSG_MORE) &&
1628 !(rt->dst.dev->features&NETIF_F_SG))
1629 alloclen = mtu;
1630 else if (!paged &&
1631 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1632 !(rt->dst.dev->features & NETIF_F_SG)))
1633 alloclen = fraglen;
1634 else {
1635 alloclen = fragheaderlen + transhdrlen;
1636 pagedlen = datalen - transhdrlen;
1637 }
1638 alloclen += alloc_extra;
1639
1640 if (datalen != length + fraggap) {
1641 /*
1642 * this is not the last fragment, the trailer
1643 * space is regarded as data space.
1644 */
1645 datalen += rt->dst.trailer_len;
1646 }
1647
1648 fraglen = datalen + fragheaderlen;
1649
1650 copy = datalen - transhdrlen - fraggap - pagedlen;
1651 /* [!] NOTE: copy may be negative if pagedlen>0
1652 * because then the equation may reduces to -fraggap.
1653 */
1654 if (copy < 0 && !(flags & MSG_SPLICE_PAGES)) {
1655 err = -EINVAL;
1656 goto error;
1657 }
1658 if (transhdrlen) {
1659 skb = sock_alloc_send_skb(sk, alloclen,
1660 (flags & MSG_DONTWAIT), &err);
1661 } else {
1662 skb = NULL;
1663 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1664 2 * sk->sk_sndbuf)
1665 skb = alloc_skb(alloclen,
1666 sk->sk_allocation);
1667 if (unlikely(!skb))
1668 err = -ENOBUFS;
1669 }
1670 if (!skb)
1671 goto error;
1672 /*
1673 * Fill in the control structures
1674 */
1675 skb->protocol = htons(ETH_P_IPV6);
1676 skb->ip_summed = csummode;
1677 skb->csum = 0;
1678 /* reserve for fragmentation and ipsec header */
1679 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1680 dst_exthdrlen);
1681
1682 /*
1683 * Find where to start putting bytes
1684 */
1685 data = skb_put(skb, fraglen - pagedlen);
1686 skb_set_network_header(skb, exthdrlen);
1687 data += fragheaderlen;
1688 skb->transport_header = (skb->network_header +
1689 fragheaderlen);
1690 if (fraggap) {
1691 skb->csum = skb_copy_and_csum_bits(
1692 skb_prev, maxfraglen,
1693 data + transhdrlen, fraggap);
1694 skb_prev->csum = csum_sub(skb_prev->csum,
1695 skb->csum);
1696 data += fraggap;
1697 pskb_trim_unique(skb_prev, maxfraglen);
1698 }
1699 if (copy > 0 &&
1700 getfrag(from, data + transhdrlen, offset,
1701 copy, fraggap, skb) < 0) {
1702 err = -EFAULT;
1703 kfree_skb(skb);
1704 goto error;
1705 } else if (flags & MSG_SPLICE_PAGES) {
1706 copy = 0;
1707 }
1708
1709 offset += copy;
1710 length -= copy + transhdrlen;
1711 transhdrlen = 0;
1712 exthdrlen = 0;
1713 dst_exthdrlen = 0;
1714
1715 /* Only the initial fragment is time stamped */
1716 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1717 cork->tx_flags = 0;
1718 skb_shinfo(skb)->tskey = tskey;
1719 tskey = 0;
1720 skb_zcopy_set(skb, uarg, &extra_uref);
1721
1722 if ((flags & MSG_CONFIRM) && !skb_prev)
1723 skb_set_dst_pending_confirm(skb, 1);
1724
1725 /*
1726 * Put the packet on the pending queue
1727 */
1728 if (!skb->destructor) {
1729 skb->destructor = sock_wfree;
1730 skb->sk = sk;
1731 wmem_alloc_delta += skb->truesize;
1732 }
1733 __skb_queue_tail(queue, skb);
1734 continue;
1735 }
1736
1737 if (copy > length)
1738 copy = length;
1739
1740 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1741 skb_tailroom(skb) >= copy) {
1742 unsigned int off;
1743
1744 off = skb->len;
1745 if (getfrag(from, skb_put(skb, copy),
1746 offset, copy, off, skb) < 0) {
1747 __skb_trim(skb, off);
1748 err = -EFAULT;
1749 goto error;
1750 }
1751 } else if (flags & MSG_SPLICE_PAGES) {
1752 struct msghdr *msg = from;
1753
1754 err = -EIO;
1755 if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1756 goto error;
1757
1758 err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1759 sk->sk_allocation);
1760 if (err < 0)
1761 goto error;
1762 copy = err;
1763 wmem_alloc_delta += copy;
1764 } else if (!zc) {
1765 int i = skb_shinfo(skb)->nr_frags;
1766
1767 err = -ENOMEM;
1768 if (!sk_page_frag_refill(sk, pfrag))
1769 goto error;
1770
1771 skb_zcopy_downgrade_managed(skb);
1772 if (!skb_can_coalesce(skb, i, pfrag->page,
1773 pfrag->offset)) {
1774 err = -EMSGSIZE;
1775 if (i == MAX_SKB_FRAGS)
1776 goto error;
1777
1778 __skb_fill_page_desc(skb, i, pfrag->page,
1779 pfrag->offset, 0);
1780 skb_shinfo(skb)->nr_frags = ++i;
1781 get_page(pfrag->page);
1782 }
1783 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1784 if (getfrag(from,
1785 page_address(pfrag->page) + pfrag->offset,
1786 offset, copy, skb->len, skb) < 0)
1787 goto error_efault;
1788
1789 pfrag->offset += copy;
1790 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1791 skb->len += copy;
1792 skb->data_len += copy;
1793 skb->truesize += copy;
1794 wmem_alloc_delta += copy;
1795 } else {
1796 err = skb_zerocopy_iter_dgram(skb, from, copy);
1797 if (err < 0)
1798 goto error;
1799 }
1800 offset += copy;
1801 length -= copy;
1802 }
1803
1804 if (wmem_alloc_delta)
1805 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1806 return 0;
1807
1808error_efault:
1809 err = -EFAULT;
1810error:
1811 net_zcopy_put_abort(uarg, extra_uref);
1812 cork->length -= length;
1813 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1814 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1815 if (hold_tskey)
1816 atomic_dec(&sk->sk_tskey);
1817 return err;
1818}
1819
1820int ip6_append_data(struct sock *sk,
1821 int getfrag(void *from, char *to, int offset, int len,
1822 int odd, struct sk_buff *skb),
1823 void *from, size_t length, int transhdrlen,
1824 struct ipcm6_cookie *ipc6, struct flowi6 *fl6,
1825 struct rt6_info *rt, unsigned int flags)
1826{
1827 struct inet_sock *inet = inet_sk(sk);
1828 struct ipv6_pinfo *np = inet6_sk(sk);
1829 int exthdrlen;
1830 int err;
1831
1832 if (flags&MSG_PROBE)
1833 return 0;
1834 if (skb_queue_empty(&sk->sk_write_queue)) {
1835 /*
1836 * setup for corking
1837 */
1838 dst_hold(&rt->dst);
1839 err = ip6_setup_cork(sk, &inet->cork, &np->cork,
1840 ipc6, rt);
1841 if (err)
1842 return err;
1843
1844 inet->cork.fl.u.ip6 = *fl6;
1845 exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
1846 length += exthdrlen;
1847 transhdrlen += exthdrlen;
1848 } else {
1849 transhdrlen = 0;
1850 }
1851
1852 return __ip6_append_data(sk, &sk->sk_write_queue, &inet->cork,
1853 &np->cork, sk_page_frag(sk), getfrag,
1854 from, length, transhdrlen, flags, ipc6);
1855}
1856EXPORT_SYMBOL_GPL(ip6_append_data);
1857
1858static void ip6_cork_steal_dst(struct sk_buff *skb, struct inet_cork_full *cork)
1859{
1860 struct dst_entry *dst = cork->base.dst;
1861
1862 cork->base.dst = NULL;
1863 skb_dst_set(skb, dst);
1864}
1865
1866static void ip6_cork_release(struct inet_cork_full *cork,
1867 struct inet6_cork *v6_cork)
1868{
1869 if (v6_cork->opt) {
1870 struct ipv6_txoptions *opt = v6_cork->opt;
1871
1872 kfree(opt->dst0opt);
1873 kfree(opt->dst1opt);
1874 kfree(opt->hopopt);
1875 kfree(opt->srcrt);
1876 kfree(opt);
1877 v6_cork->opt = NULL;
1878 }
1879
1880 if (cork->base.dst) {
1881 dst_release(cork->base.dst);
1882 cork->base.dst = NULL;
1883 }
1884}
1885
1886struct sk_buff *__ip6_make_skb(struct sock *sk,
1887 struct sk_buff_head *queue,
1888 struct inet_cork_full *cork,
1889 struct inet6_cork *v6_cork)
1890{
1891 struct sk_buff *skb, *tmp_skb;
1892 struct sk_buff **tail_skb;
1893 struct in6_addr *final_dst;
1894 struct net *net = sock_net(sk);
1895 struct ipv6hdr *hdr;
1896 struct ipv6_txoptions *opt = v6_cork->opt;
1897 struct rt6_info *rt = dst_rt6_info(cork->base.dst);
1898 struct flowi6 *fl6 = &cork->fl.u.ip6;
1899 unsigned char proto = fl6->flowi6_proto;
1900
1901 skb = __skb_dequeue(queue);
1902 if (!skb)
1903 goto out;
1904 tail_skb = &(skb_shinfo(skb)->frag_list);
1905
1906 /* move skb->data to ip header from ext header */
1907 if (skb->data < skb_network_header(skb))
1908 __skb_pull(skb, skb_network_offset(skb));
1909 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1910 __skb_pull(tmp_skb, skb_network_header_len(skb));
1911 *tail_skb = tmp_skb;
1912 tail_skb = &(tmp_skb->next);
1913 skb->len += tmp_skb->len;
1914 skb->data_len += tmp_skb->len;
1915 skb->truesize += tmp_skb->truesize;
1916 tmp_skb->destructor = NULL;
1917 tmp_skb->sk = NULL;
1918 }
1919
1920 /* Allow local fragmentation. */
1921 skb->ignore_df = ip6_sk_ignore_df(sk);
1922 __skb_pull(skb, skb_network_header_len(skb));
1923
1924 final_dst = &fl6->daddr;
1925 if (opt && opt->opt_flen)
1926 ipv6_push_frag_opts(skb, opt, &proto);
1927 if (opt && opt->opt_nflen)
1928 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst, &fl6->saddr);
1929
1930 skb_push(skb, sizeof(struct ipv6hdr));
1931 skb_reset_network_header(skb);
1932 hdr = ipv6_hdr(skb);
1933
1934 ip6_flow_hdr(hdr, v6_cork->tclass,
1935 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1936 ip6_autoflowlabel(net, sk), fl6));
1937 hdr->hop_limit = v6_cork->hop_limit;
1938 hdr->nexthdr = proto;
1939 hdr->saddr = fl6->saddr;
1940 hdr->daddr = *final_dst;
1941
1942 skb->priority = READ_ONCE(sk->sk_priority);
1943 skb->mark = cork->base.mark;
1944 if (sk_is_tcp(sk))
1945 skb_set_delivery_time(skb, cork->base.transmit_time, SKB_CLOCK_MONOTONIC);
1946 else
1947 skb_set_delivery_type_by_clockid(skb, cork->base.transmit_time, sk->sk_clockid);
1948
1949 ip6_cork_steal_dst(skb, cork);
1950 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
1951 if (proto == IPPROTO_ICMPV6) {
1952 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1953 u8 icmp6_type;
1954
1955 if (sk->sk_socket->type == SOCK_RAW &&
1956 !(fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH))
1957 icmp6_type = fl6->fl6_icmp_type;
1958 else
1959 icmp6_type = icmp6_hdr(skb)->icmp6_type;
1960 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_type);
1961 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1962 }
1963
1964 ip6_cork_release(cork, v6_cork);
1965out:
1966 return skb;
1967}
1968
1969int ip6_send_skb(struct sk_buff *skb)
1970{
1971 struct net *net = sock_net(skb->sk);
1972 struct rt6_info *rt = dst_rt6_info(skb_dst(skb));
1973 int err;
1974
1975 rcu_read_lock();
1976 err = ip6_local_out(net, skb->sk, skb);
1977 if (err) {
1978 if (err > 0)
1979 err = net_xmit_errno(err);
1980 if (err)
1981 IP6_INC_STATS(net, rt->rt6i_idev,
1982 IPSTATS_MIB_OUTDISCARDS);
1983 }
1984
1985 rcu_read_unlock();
1986 return err;
1987}
1988
1989int ip6_push_pending_frames(struct sock *sk)
1990{
1991 struct sk_buff *skb;
1992
1993 skb = ip6_finish_skb(sk);
1994 if (!skb)
1995 return 0;
1996
1997 return ip6_send_skb(skb);
1998}
1999EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
2000
2001static void __ip6_flush_pending_frames(struct sock *sk,
2002 struct sk_buff_head *queue,
2003 struct inet_cork_full *cork,
2004 struct inet6_cork *v6_cork)
2005{
2006 struct sk_buff *skb;
2007
2008 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
2009 if (skb_dst(skb))
2010 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
2011 IPSTATS_MIB_OUTDISCARDS);
2012 kfree_skb(skb);
2013 }
2014
2015 ip6_cork_release(cork, v6_cork);
2016}
2017
2018void ip6_flush_pending_frames(struct sock *sk)
2019{
2020 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
2021 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
2022}
2023EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
2024
2025struct sk_buff *ip6_make_skb(struct sock *sk,
2026 int getfrag(void *from, char *to, int offset,
2027 int len, int odd, struct sk_buff *skb),
2028 void *from, size_t length, int transhdrlen,
2029 struct ipcm6_cookie *ipc6, struct rt6_info *rt,
2030 unsigned int flags, struct inet_cork_full *cork)
2031{
2032 struct inet6_cork v6_cork;
2033 struct sk_buff_head queue;
2034 int exthdrlen = (ipc6->opt ? ipc6->opt->opt_flen : 0);
2035 int err;
2036
2037 if (flags & MSG_PROBE) {
2038 dst_release(&rt->dst);
2039 return NULL;
2040 }
2041
2042 __skb_queue_head_init(&queue);
2043
2044 cork->base.flags = 0;
2045 cork->base.addr = 0;
2046 cork->base.opt = NULL;
2047 v6_cork.opt = NULL;
2048 err = ip6_setup_cork(sk, cork, &v6_cork, ipc6, rt);
2049 if (err) {
2050 ip6_cork_release(cork, &v6_cork);
2051 return ERR_PTR(err);
2052 }
2053 if (ipc6->dontfrag < 0)
2054 ipc6->dontfrag = inet6_test_bit(DONTFRAG, sk);
2055
2056 err = __ip6_append_data(sk, &queue, cork, &v6_cork,
2057 ¤t->task_frag, getfrag, from,
2058 length + exthdrlen, transhdrlen + exthdrlen,
2059 flags, ipc6);
2060 if (err) {
2061 __ip6_flush_pending_frames(sk, &queue, cork, &v6_cork);
2062 return ERR_PTR(err);
2063 }
2064
2065 return __ip6_make_skb(sk, &queue, cork, &v6_cork);
2066}