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#include <net/l3mdev.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 neighbour *neigh;
65 struct in6_addr *nexthop;
66 int ret;
67
68 skb->protocol = htons(ETH_P_IPV6);
69 skb->dev = dev;
70
71 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
72 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
73
74 if (!(dev->flags & IFF_LOOPBACK) && sk_mc_loop(sk) &&
75 ((mroute6_socket(net, skb) &&
76 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
77 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
78 &ipv6_hdr(skb)->saddr))) {
79 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
80
81 /* Do not check for IFF_ALLMULTI; multicast routing
82 is not supported in any case.
83 */
84 if (newskb)
85 NF_HOOK(NFPROTO_IPV6, NF_INET_POST_ROUTING,
86 net, sk, newskb, NULL, newskb->dev,
87 dev_loopback_xmit);
88
89 if (ipv6_hdr(skb)->hop_limit == 0) {
90 IP6_INC_STATS(net, idev,
91 IPSTATS_MIB_OUTDISCARDS);
92 kfree_skb(skb);
93 return 0;
94 }
95 }
96
97 IP6_UPD_PO_STATS(net, idev, IPSTATS_MIB_OUTMCAST, 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, &ipv6_hdr(skb)->daddr);
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(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
120 kfree_skb(skb);
121 return -EINVAL;
122}
123
124static int ip6_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
125{
126 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
127 dst_allfrag(skb_dst(skb)) ||
128 (IP6CB(skb)->frag_max_size && skb->len > IP6CB(skb)->frag_max_size))
129 return ip6_fragment(net, sk, skb, ip6_finish_output2);
130 else
131 return ip6_finish_output2(net, sk, skb);
132}
133
134int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
135{
136 struct net_device *dev = skb_dst(skb)->dev;
137 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
138
139 if (unlikely(idev->cnf.disable_ipv6)) {
140 IP6_INC_STATS(net, idev, IPSTATS_MIB_OUTDISCARDS);
141 kfree_skb(skb);
142 return 0;
143 }
144
145 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
146 net, sk, 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 * Note : socket lock is not held for SYNACK packets, but might be modified
154 * by calls to skb_set_owner_w() and ipv6_local_error(),
155 * which are using proper atomic operations or spinlocks.
156 */
157int ip6_xmit(const struct sock *sk, struct sk_buff *skb, struct flowi6 *fl6,
158 struct ipv6_txoptions *opt, int tclass)
159{
160 struct net *net = sock_net(sk);
161 const struct ipv6_pinfo *np = inet6_sk(sk);
162 struct in6_addr *first_hop = &fl6->daddr;
163 struct dst_entry *dst = skb_dst(skb);
164 struct ipv6hdr *hdr;
165 u8 proto = fl6->flowi6_proto;
166 int seg_len = skb->len;
167 int hlimit = -1;
168 u32 mtu;
169
170 if (opt) {
171 unsigned int head_room;
172
173 /* First: exthdrs may take lots of space (~8K for now)
174 MAX_HEADER is not enough.
175 */
176 head_room = opt->opt_nflen + opt->opt_flen;
177 seg_len += head_room;
178 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
179
180 if (skb_headroom(skb) < head_room) {
181 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
182 if (!skb2) {
183 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
184 IPSTATS_MIB_OUTDISCARDS);
185 kfree_skb(skb);
186 return -ENOBUFS;
187 }
188 consume_skb(skb);
189 skb = skb2;
190 /* skb_set_owner_w() changes sk->sk_wmem_alloc atomically,
191 * it is safe to call in our context (socket lock not held)
192 */
193 skb_set_owner_w(skb, (struct sock *)sk);
194 }
195 if (opt->opt_flen)
196 ipv6_push_frag_opts(skb, opt, &proto);
197 if (opt->opt_nflen)
198 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
199 }
200
201 skb_push(skb, sizeof(struct ipv6hdr));
202 skb_reset_network_header(skb);
203 hdr = ipv6_hdr(skb);
204
205 /*
206 * Fill in the IPv6 header
207 */
208 if (np)
209 hlimit = np->hop_limit;
210 if (hlimit < 0)
211 hlimit = ip6_dst_hoplimit(dst);
212
213 ip6_flow_hdr(hdr, tclass, ip6_make_flowlabel(net, skb, fl6->flowlabel,
214 np->autoflowlabel, fl6));
215
216 hdr->payload_len = htons(seg_len);
217 hdr->nexthdr = proto;
218 hdr->hop_limit = hlimit;
219
220 hdr->saddr = fl6->saddr;
221 hdr->daddr = *first_hop;
222
223 skb->protocol = htons(ETH_P_IPV6);
224 skb->priority = sk->sk_priority;
225 skb->mark = sk->sk_mark;
226
227 mtu = dst_mtu(dst);
228 if ((skb->len <= mtu) || skb->ignore_df || skb_is_gso(skb)) {
229 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
230 IPSTATS_MIB_OUT, skb->len);
231 /* hooks should never assume socket lock is held.
232 * we promote our socket to non const
233 */
234 return NF_HOOK(NFPROTO_IPV6, NF_INET_LOCAL_OUT,
235 net, (struct sock *)sk, skb, NULL, dst->dev,
236 dst_output);
237 }
238
239 skb->dev = dst->dev;
240 /* ipv6_local_error() does not require socket lock,
241 * we promote our socket to non const
242 */
243 ipv6_local_error((struct sock *)sk, EMSGSIZE, fl6, mtu);
244
245 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
246 kfree_skb(skb);
247 return -EMSGSIZE;
248}
249EXPORT_SYMBOL(ip6_xmit);
250
251static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
252{
253 struct ip6_ra_chain *ra;
254 struct sock *last = NULL;
255
256 read_lock(&ip6_ra_lock);
257 for (ra = ip6_ra_chain; ra; ra = ra->next) {
258 struct sock *sk = ra->sk;
259 if (sk && ra->sel == sel &&
260 (!sk->sk_bound_dev_if ||
261 sk->sk_bound_dev_if == skb->dev->ifindex)) {
262 if (last) {
263 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
264 if (skb2)
265 rawv6_rcv(last, skb2);
266 }
267 last = sk;
268 }
269 }
270
271 if (last) {
272 rawv6_rcv(last, skb);
273 read_unlock(&ip6_ra_lock);
274 return 1;
275 }
276 read_unlock(&ip6_ra_lock);
277 return 0;
278}
279
280static int ip6_forward_proxy_check(struct sk_buff *skb)
281{
282 struct ipv6hdr *hdr = ipv6_hdr(skb);
283 u8 nexthdr = hdr->nexthdr;
284 __be16 frag_off;
285 int offset;
286
287 if (ipv6_ext_hdr(nexthdr)) {
288 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr, &frag_off);
289 if (offset < 0)
290 return 0;
291 } else
292 offset = sizeof(struct ipv6hdr);
293
294 if (nexthdr == IPPROTO_ICMPV6) {
295 struct icmp6hdr *icmp6;
296
297 if (!pskb_may_pull(skb, (skb_network_header(skb) +
298 offset + 1 - skb->data)))
299 return 0;
300
301 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
302
303 switch (icmp6->icmp6_type) {
304 case NDISC_ROUTER_SOLICITATION:
305 case NDISC_ROUTER_ADVERTISEMENT:
306 case NDISC_NEIGHBOUR_SOLICITATION:
307 case NDISC_NEIGHBOUR_ADVERTISEMENT:
308 case NDISC_REDIRECT:
309 /* For reaction involving unicast neighbor discovery
310 * message destined to the proxied address, pass it to
311 * input function.
312 */
313 return 1;
314 default:
315 break;
316 }
317 }
318
319 /*
320 * The proxying router can't forward traffic sent to a link-local
321 * address, so signal the sender and discard the packet. This
322 * behavior is clarified by the MIPv6 specification.
323 */
324 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
325 dst_link_failure(skb);
326 return -1;
327 }
328
329 return 0;
330}
331
332static inline int ip6_forward_finish(struct net *net, struct sock *sk,
333 struct sk_buff *skb)
334{
335 return dst_output(net, sk, skb);
336}
337
338static unsigned int ip6_dst_mtu_forward(const struct dst_entry *dst)
339{
340 unsigned int mtu;
341 struct inet6_dev *idev;
342
343 if (dst_metric_locked(dst, RTAX_MTU)) {
344 mtu = dst_metric_raw(dst, RTAX_MTU);
345 if (mtu)
346 return mtu;
347 }
348
349 mtu = IPV6_MIN_MTU;
350 rcu_read_lock();
351 idev = __in6_dev_get(dst->dev);
352 if (idev)
353 mtu = idev->cnf.mtu6;
354 rcu_read_unlock();
355
356 return mtu;
357}
358
359static bool ip6_pkt_too_big(const struct sk_buff *skb, unsigned int mtu)
360{
361 if (skb->len <= mtu)
362 return false;
363
364 /* ipv6 conntrack defrag sets max_frag_size + ignore_df */
365 if (IP6CB(skb)->frag_max_size && IP6CB(skb)->frag_max_size > mtu)
366 return true;
367
368 if (skb->ignore_df)
369 return false;
370
371 if (skb_is_gso(skb) && skb_gso_network_seglen(skb) <= mtu)
372 return false;
373
374 return true;
375}
376
377int ip6_forward(struct sk_buff *skb)
378{
379 struct dst_entry *dst = skb_dst(skb);
380 struct ipv6hdr *hdr = ipv6_hdr(skb);
381 struct inet6_skb_parm *opt = IP6CB(skb);
382 struct net *net = dev_net(dst->dev);
383 u32 mtu;
384
385 if (net->ipv6.devconf_all->forwarding == 0)
386 goto error;
387
388 if (skb->pkt_type != PACKET_HOST)
389 goto drop;
390
391 if (unlikely(skb->sk))
392 goto drop;
393
394 if (skb_warn_if_lro(skb))
395 goto drop;
396
397 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
398 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
399 IPSTATS_MIB_INDISCARDS);
400 goto drop;
401 }
402
403 skb_forward_csum(skb);
404
405 /*
406 * We DO NOT make any processing on
407 * RA packets, pushing them to user level AS IS
408 * without ane WARRANTY that application will be able
409 * to interpret them. The reason is that we
410 * cannot make anything clever here.
411 *
412 * We are not end-node, so that if packet contains
413 * AH/ESP, we cannot make anything.
414 * Defragmentation also would be mistake, RA packets
415 * cannot be fragmented, because there is no warranty
416 * that different fragments will go along one path. --ANK
417 */
418 if (unlikely(opt->flags & IP6SKB_ROUTERALERT)) {
419 if (ip6_call_ra_chain(skb, ntohs(opt->ra)))
420 return 0;
421 }
422
423 /*
424 * check and decrement ttl
425 */
426 if (hdr->hop_limit <= 1) {
427 /* Force OUTPUT device used as source address */
428 skb->dev = dst->dev;
429 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT, 0);
430 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
431 IPSTATS_MIB_INHDRERRORS);
432
433 kfree_skb(skb);
434 return -ETIMEDOUT;
435 }
436
437 /* XXX: idev->cnf.proxy_ndp? */
438 if (net->ipv6.devconf_all->proxy_ndp &&
439 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
440 int proxied = ip6_forward_proxy_check(skb);
441 if (proxied > 0)
442 return ip6_input(skb);
443 else if (proxied < 0) {
444 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
445 IPSTATS_MIB_INDISCARDS);
446 goto drop;
447 }
448 }
449
450 if (!xfrm6_route_forward(skb)) {
451 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
452 IPSTATS_MIB_INDISCARDS);
453 goto drop;
454 }
455 dst = skb_dst(skb);
456
457 /* IPv6 specs say nothing about it, but it is clear that we cannot
458 send redirects to source routed frames.
459 We don't send redirects to frames decapsulated from IPsec.
460 */
461 if (skb->dev == dst->dev && opt->srcrt == 0 && !skb_sec_path(skb)) {
462 struct in6_addr *target = NULL;
463 struct inet_peer *peer;
464 struct rt6_info *rt;
465
466 /*
467 * incoming and outgoing devices are the same
468 * send a redirect.
469 */
470
471 rt = (struct rt6_info *) dst;
472 if (rt->rt6i_flags & RTF_GATEWAY)
473 target = &rt->rt6i_gateway;
474 else
475 target = &hdr->daddr;
476
477 peer = inet_getpeer_v6(net->ipv6.peers, &hdr->daddr, 1);
478
479 /* Limit redirects both by destination (here)
480 and by source (inside ndisc_send_redirect)
481 */
482 if (inet_peer_xrlim_allow(peer, 1*HZ))
483 ndisc_send_redirect(skb, target);
484 if (peer)
485 inet_putpeer(peer);
486 } else {
487 int addrtype = ipv6_addr_type(&hdr->saddr);
488
489 /* This check is security critical. */
490 if (addrtype == IPV6_ADDR_ANY ||
491 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
492 goto error;
493 if (addrtype & IPV6_ADDR_LINKLOCAL) {
494 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
495 ICMPV6_NOT_NEIGHBOUR, 0);
496 goto error;
497 }
498 }
499
500 mtu = ip6_dst_mtu_forward(dst);
501 if (mtu < IPV6_MIN_MTU)
502 mtu = IPV6_MIN_MTU;
503
504 if (ip6_pkt_too_big(skb, mtu)) {
505 /* Again, force OUTPUT device used as source address */
506 skb->dev = dst->dev;
507 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
508 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
509 IPSTATS_MIB_INTOOBIGERRORS);
510 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
511 IPSTATS_MIB_FRAGFAILS);
512 kfree_skb(skb);
513 return -EMSGSIZE;
514 }
515
516 if (skb_cow(skb, dst->dev->hard_header_len)) {
517 IP6_INC_STATS_BH(net, ip6_dst_idev(dst),
518 IPSTATS_MIB_OUTDISCARDS);
519 goto drop;
520 }
521
522 hdr = ipv6_hdr(skb);
523
524 /* Mangling hops number delayed to point after skb COW */
525
526 hdr->hop_limit--;
527
528 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
529 IP6_ADD_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTOCTETS, skb->len);
530 return NF_HOOK(NFPROTO_IPV6, NF_INET_FORWARD,
531 net, NULL, skb, skb->dev, dst->dev,
532 ip6_forward_finish);
533
534error:
535 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
536drop:
537 kfree_skb(skb);
538 return -EINVAL;
539}
540
541static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
542{
543 to->pkt_type = from->pkt_type;
544 to->priority = from->priority;
545 to->protocol = from->protocol;
546 skb_dst_drop(to);
547 skb_dst_set(to, dst_clone(skb_dst(from)));
548 to->dev = from->dev;
549 to->mark = from->mark;
550
551#ifdef CONFIG_NET_SCHED
552 to->tc_index = from->tc_index;
553#endif
554 nf_copy(to, from);
555 skb_copy_secmark(to, from);
556}
557
558int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
559 int (*output)(struct net *, struct sock *, struct sk_buff *))
560{
561 struct sk_buff *frag;
562 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
563 struct ipv6_pinfo *np = skb->sk && !dev_recursion_level() ?
564 inet6_sk(skb->sk) : NULL;
565 struct ipv6hdr *tmp_hdr;
566 struct frag_hdr *fh;
567 unsigned int mtu, hlen, left, len;
568 int hroom, troom;
569 __be32 frag_id;
570 int ptr, offset = 0, err = 0;
571 u8 *prevhdr, nexthdr = 0;
572
573 hlen = ip6_find_1stfragopt(skb, &prevhdr);
574 nexthdr = *prevhdr;
575
576 mtu = ip6_skb_dst_mtu(skb);
577
578 /* We must not fragment if the socket is set to force MTU discovery
579 * or if the skb it not generated by a local socket.
580 */
581 if (unlikely(!skb->ignore_df && skb->len > mtu))
582 goto fail_toobig;
583
584 if (IP6CB(skb)->frag_max_size) {
585 if (IP6CB(skb)->frag_max_size > mtu)
586 goto fail_toobig;
587
588 /* don't send fragments larger than what we received */
589 mtu = IP6CB(skb)->frag_max_size;
590 if (mtu < IPV6_MIN_MTU)
591 mtu = IPV6_MIN_MTU;
592 }
593
594 if (np && np->frag_size < mtu) {
595 if (np->frag_size)
596 mtu = np->frag_size;
597 }
598 if (mtu < hlen + sizeof(struct frag_hdr) + 8)
599 goto fail_toobig;
600 mtu -= hlen + sizeof(struct frag_hdr);
601
602 frag_id = ipv6_select_ident(net, &ipv6_hdr(skb)->daddr,
603 &ipv6_hdr(skb)->saddr);
604
605 if (skb->ip_summed == CHECKSUM_PARTIAL &&
606 (err = skb_checksum_help(skb)))
607 goto fail;
608
609 hroom = LL_RESERVED_SPACE(rt->dst.dev);
610 if (skb_has_frag_list(skb)) {
611 int first_len = skb_pagelen(skb);
612 struct sk_buff *frag2;
613
614 if (first_len - hlen > mtu ||
615 ((first_len - hlen) & 7) ||
616 skb_cloned(skb) ||
617 skb_headroom(skb) < (hroom + sizeof(struct frag_hdr)))
618 goto slow_path;
619
620 skb_walk_frags(skb, frag) {
621 /* Correct geometry. */
622 if (frag->len > mtu ||
623 ((frag->len & 7) && frag->next) ||
624 skb_headroom(frag) < (hlen + hroom + sizeof(struct frag_hdr)))
625 goto slow_path_clean;
626
627 /* Partially cloned skb? */
628 if (skb_shared(frag))
629 goto slow_path_clean;
630
631 BUG_ON(frag->sk);
632 if (skb->sk) {
633 frag->sk = skb->sk;
634 frag->destructor = sock_wfree;
635 }
636 skb->truesize -= frag->truesize;
637 }
638
639 err = 0;
640 offset = 0;
641 /* BUILD HEADER */
642
643 *prevhdr = NEXTHDR_FRAGMENT;
644 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
645 if (!tmp_hdr) {
646 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
647 IPSTATS_MIB_FRAGFAILS);
648 err = -ENOMEM;
649 goto fail;
650 }
651 frag = skb_shinfo(skb)->frag_list;
652 skb_frag_list_init(skb);
653
654 __skb_pull(skb, hlen);
655 fh = (struct frag_hdr *)__skb_push(skb, sizeof(struct frag_hdr));
656 __skb_push(skb, hlen);
657 skb_reset_network_header(skb);
658 memcpy(skb_network_header(skb), tmp_hdr, hlen);
659
660 fh->nexthdr = nexthdr;
661 fh->reserved = 0;
662 fh->frag_off = htons(IP6_MF);
663 fh->identification = frag_id;
664
665 first_len = skb_pagelen(skb);
666 skb->data_len = first_len - skb_headlen(skb);
667 skb->len = first_len;
668 ipv6_hdr(skb)->payload_len = htons(first_len -
669 sizeof(struct ipv6hdr));
670
671 dst_hold(&rt->dst);
672
673 for (;;) {
674 /* Prepare header of the next frame,
675 * before previous one went down. */
676 if (frag) {
677 frag->ip_summed = CHECKSUM_NONE;
678 skb_reset_transport_header(frag);
679 fh = (struct frag_hdr *)__skb_push(frag, sizeof(struct frag_hdr));
680 __skb_push(frag, hlen);
681 skb_reset_network_header(frag);
682 memcpy(skb_network_header(frag), tmp_hdr,
683 hlen);
684 offset += skb->len - hlen - sizeof(struct frag_hdr);
685 fh->nexthdr = nexthdr;
686 fh->reserved = 0;
687 fh->frag_off = htons(offset);
688 if (frag->next)
689 fh->frag_off |= htons(IP6_MF);
690 fh->identification = frag_id;
691 ipv6_hdr(frag)->payload_len =
692 htons(frag->len -
693 sizeof(struct ipv6hdr));
694 ip6_copy_metadata(frag, skb);
695 }
696
697 err = output(net, sk, skb);
698 if (!err)
699 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
700 IPSTATS_MIB_FRAGCREATES);
701
702 if (err || !frag)
703 break;
704
705 skb = frag;
706 frag = skb->next;
707 skb->next = NULL;
708 }
709
710 kfree(tmp_hdr);
711
712 if (err == 0) {
713 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
714 IPSTATS_MIB_FRAGOKS);
715 ip6_rt_put(rt);
716 return 0;
717 }
718
719 kfree_skb_list(frag);
720
721 IP6_INC_STATS(net, ip6_dst_idev(&rt->dst),
722 IPSTATS_MIB_FRAGFAILS);
723 ip6_rt_put(rt);
724 return err;
725
726slow_path_clean:
727 skb_walk_frags(skb, frag2) {
728 if (frag2 == frag)
729 break;
730 frag2->sk = NULL;
731 frag2->destructor = NULL;
732 skb->truesize += frag2->truesize;
733 }
734 }
735
736slow_path:
737 left = skb->len - hlen; /* Space per frame */
738 ptr = hlen; /* Where to start from */
739
740 /*
741 * Fragment the datagram.
742 */
743
744 *prevhdr = NEXTHDR_FRAGMENT;
745 troom = rt->dst.dev->needed_tailroom;
746
747 /*
748 * Keep copying data until we run out.
749 */
750 while (left > 0) {
751 len = left;
752 /* IF: it doesn't fit, use 'mtu' - the data space left */
753 if (len > mtu)
754 len = mtu;
755 /* IF: we are not sending up to and including the packet end
756 then align the next start on an eight byte boundary */
757 if (len < left) {
758 len &= ~7;
759 }
760
761 /* Allocate buffer */
762 frag = alloc_skb(len + hlen + sizeof(struct frag_hdr) +
763 hroom + troom, GFP_ATOMIC);
764 if (!frag) {
765 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
766 IPSTATS_MIB_FRAGFAILS);
767 err = -ENOMEM;
768 goto fail;
769 }
770
771 /*
772 * Set up data on packet
773 */
774
775 ip6_copy_metadata(frag, skb);
776 skb_reserve(frag, hroom);
777 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
778 skb_reset_network_header(frag);
779 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
780 frag->transport_header = (frag->network_header + hlen +
781 sizeof(struct frag_hdr));
782
783 /*
784 * Charge the memory for the fragment to any owner
785 * it might possess
786 */
787 if (skb->sk)
788 skb_set_owner_w(frag, skb->sk);
789
790 /*
791 * Copy the packet header into the new buffer.
792 */
793 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
794
795 /*
796 * Build fragment header.
797 */
798 fh->nexthdr = nexthdr;
799 fh->reserved = 0;
800 fh->identification = frag_id;
801
802 /*
803 * Copy a block of the IP datagram.
804 */
805 BUG_ON(skb_copy_bits(skb, ptr, skb_transport_header(frag),
806 len));
807 left -= len;
808
809 fh->frag_off = htons(offset);
810 if (left > 0)
811 fh->frag_off |= htons(IP6_MF);
812 ipv6_hdr(frag)->payload_len = htons(frag->len -
813 sizeof(struct ipv6hdr));
814
815 ptr += len;
816 offset += len;
817
818 /*
819 * Put this fragment into the sending queue.
820 */
821 err = output(net, sk, frag);
822 if (err)
823 goto fail;
824
825 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
826 IPSTATS_MIB_FRAGCREATES);
827 }
828 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
829 IPSTATS_MIB_FRAGOKS);
830 consume_skb(skb);
831 return err;
832
833fail_toobig:
834 if (skb->sk && dst_allfrag(skb_dst(skb)))
835 sk_nocaps_add(skb->sk, NETIF_F_GSO_MASK);
836
837 skb->dev = skb_dst(skb)->dev;
838 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
839 err = -EMSGSIZE;
840
841fail:
842 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
843 IPSTATS_MIB_FRAGFAILS);
844 kfree_skb(skb);
845 return err;
846}
847
848static inline int ip6_rt_check(const struct rt6key *rt_key,
849 const struct in6_addr *fl_addr,
850 const struct in6_addr *addr_cache)
851{
852 return (rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
853 (!addr_cache || !ipv6_addr_equal(fl_addr, addr_cache));
854}
855
856static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
857 struct dst_entry *dst,
858 const struct flowi6 *fl6)
859{
860 struct ipv6_pinfo *np = inet6_sk(sk);
861 struct rt6_info *rt;
862
863 if (!dst)
864 goto out;
865
866 if (dst->ops->family != AF_INET6) {
867 dst_release(dst);
868 return NULL;
869 }
870
871 rt = (struct rt6_info *)dst;
872 /* Yes, checking route validity in not connected
873 * case is not very simple. Take into account,
874 * that we do not support routing by source, TOS,
875 * and MSG_DONTROUTE --ANK (980726)
876 *
877 * 1. ip6_rt_check(): If route was host route,
878 * check that cached destination is current.
879 * If it is network route, we still may
880 * check its validity using saved pointer
881 * to the last used address: daddr_cache.
882 * We do not want to save whole address now,
883 * (because main consumer of this service
884 * is tcp, which has not this problem),
885 * so that the last trick works only on connected
886 * sockets.
887 * 2. oif also should be the same.
888 */
889 if (ip6_rt_check(&rt->rt6i_dst, &fl6->daddr, np->daddr_cache) ||
890#ifdef CONFIG_IPV6_SUBTREES
891 ip6_rt_check(&rt->rt6i_src, &fl6->saddr, np->saddr_cache) ||
892#endif
893 (!(fl6->flowi6_flags & FLOWI_FLAG_SKIP_NH_OIF) &&
894 (fl6->flowi6_oif && fl6->flowi6_oif != dst->dev->ifindex))) {
895 dst_release(dst);
896 dst = NULL;
897 }
898
899out:
900 return dst;
901}
902
903static int ip6_dst_lookup_tail(struct net *net, const struct sock *sk,
904 struct dst_entry **dst, struct flowi6 *fl6)
905{
906#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
907 struct neighbour *n;
908 struct rt6_info *rt;
909#endif
910 int err;
911 int flags = 0;
912
913 /* The correct way to handle this would be to do
914 * ip6_route_get_saddr, and then ip6_route_output; however,
915 * the route-specific preferred source forces the
916 * ip6_route_output call _before_ ip6_route_get_saddr.
917 *
918 * In source specific routing (no src=any default route),
919 * ip6_route_output will fail given src=any saddr, though, so
920 * that's why we try it again later.
921 */
922 if (ipv6_addr_any(&fl6->saddr) && (!*dst || !(*dst)->error)) {
923 struct rt6_info *rt;
924 bool had_dst = *dst != NULL;
925
926 if (!had_dst)
927 *dst = ip6_route_output(net, sk, fl6);
928 rt = (*dst)->error ? NULL : (struct rt6_info *)*dst;
929 err = ip6_route_get_saddr(net, rt, &fl6->daddr,
930 sk ? inet6_sk(sk)->srcprefs : 0,
931 &fl6->saddr);
932 if (err)
933 goto out_err_release;
934
935 /* If we had an erroneous initial result, pretend it
936 * never existed and let the SA-enabled version take
937 * over.
938 */
939 if (!had_dst && (*dst)->error) {
940 dst_release(*dst);
941 *dst = NULL;
942 }
943
944 if (fl6->flowi6_oif)
945 flags |= RT6_LOOKUP_F_IFACE;
946 }
947
948 if (!*dst)
949 *dst = ip6_route_output_flags(net, sk, fl6, flags);
950
951 err = (*dst)->error;
952 if (err)
953 goto out_err_release;
954
955#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
956 /*
957 * Here if the dst entry we've looked up
958 * has a neighbour entry that is in the INCOMPLETE
959 * state and the src address from the flow is
960 * marked as OPTIMISTIC, we release the found
961 * dst entry and replace it instead with the
962 * dst entry of the nexthop router
963 */
964 rt = (struct rt6_info *) *dst;
965 rcu_read_lock_bh();
966 n = __ipv6_neigh_lookup_noref(rt->dst.dev,
967 rt6_nexthop(rt, &fl6->daddr));
968 err = n && !(n->nud_state & NUD_VALID) ? -EINVAL : 0;
969 rcu_read_unlock_bh();
970
971 if (err) {
972 struct inet6_ifaddr *ifp;
973 struct flowi6 fl_gw6;
974 int redirect;
975
976 ifp = ipv6_get_ifaddr(net, &fl6->saddr,
977 (*dst)->dev, 1);
978
979 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
980 if (ifp)
981 in6_ifa_put(ifp);
982
983 if (redirect) {
984 /*
985 * We need to get the dst entry for the
986 * default router instead
987 */
988 dst_release(*dst);
989 memcpy(&fl_gw6, fl6, sizeof(struct flowi6));
990 memset(&fl_gw6.daddr, 0, sizeof(struct in6_addr));
991 *dst = ip6_route_output(net, sk, &fl_gw6);
992 err = (*dst)->error;
993 if (err)
994 goto out_err_release;
995 }
996 }
997#endif
998
999 return 0;
1000
1001out_err_release:
1002 if (err == -ENETUNREACH)
1003 IP6_INC_STATS(net, NULL, IPSTATS_MIB_OUTNOROUTES);
1004 dst_release(*dst);
1005 *dst = NULL;
1006 return err;
1007}
1008
1009/**
1010 * ip6_dst_lookup - perform route lookup on flow
1011 * @sk: socket which provides route info
1012 * @dst: pointer to dst_entry * for result
1013 * @fl6: flow to lookup
1014 *
1015 * This function performs a route lookup on the given flow.
1016 *
1017 * It returns zero on success, or a standard errno code on error.
1018 */
1019int ip6_dst_lookup(struct net *net, struct sock *sk, struct dst_entry **dst,
1020 struct flowi6 *fl6)
1021{
1022 *dst = NULL;
1023 return ip6_dst_lookup_tail(net, sk, dst, fl6);
1024}
1025EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1026
1027/**
1028 * ip6_dst_lookup_flow - perform route lookup on flow with ipsec
1029 * @sk: socket which provides route info
1030 * @fl6: flow to lookup
1031 * @final_dst: final destination address for ipsec lookup
1032 *
1033 * This function performs a route lookup on the given flow.
1034 *
1035 * It returns a valid dst pointer on success, or a pointer encoded
1036 * error code.
1037 */
1038struct dst_entry *ip6_dst_lookup_flow(const struct sock *sk, struct flowi6 *fl6,
1039 const struct in6_addr *final_dst)
1040{
1041 struct dst_entry *dst = NULL;
1042 int err;
1043
1044 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1045 if (err)
1046 return ERR_PTR(err);
1047 if (final_dst)
1048 fl6->daddr = *final_dst;
1049 if (!fl6->flowi6_oif)
1050 fl6->flowi6_oif = l3mdev_fib_oif(dst->dev);
1051
1052 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1053}
1054EXPORT_SYMBOL_GPL(ip6_dst_lookup_flow);
1055
1056/**
1057 * ip6_sk_dst_lookup_flow - perform socket cached route lookup on flow
1058 * @sk: socket which provides the dst cache and route info
1059 * @fl6: flow to lookup
1060 * @final_dst: final destination address for ipsec lookup
1061 *
1062 * This function performs a route lookup on the given flow with the
1063 * possibility of using the cached route in the socket if it is valid.
1064 * It will take the socket dst lock when operating on the dst cache.
1065 * As a result, this function can only be used in process context.
1066 *
1067 * It returns a valid dst pointer on success, or a pointer encoded
1068 * error code.
1069 */
1070struct dst_entry *ip6_sk_dst_lookup_flow(struct sock *sk, struct flowi6 *fl6,
1071 const struct in6_addr *final_dst)
1072{
1073 struct dst_entry *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1074 int err;
1075
1076 dst = ip6_sk_dst_check(sk, dst, fl6);
1077
1078 err = ip6_dst_lookup_tail(sock_net(sk), sk, &dst, fl6);
1079 if (err)
1080 return ERR_PTR(err);
1081 if (final_dst)
1082 fl6->daddr = *final_dst;
1083
1084 return xfrm_lookup_route(sock_net(sk), dst, flowi6_to_flowi(fl6), sk, 0);
1085}
1086EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup_flow);
1087
1088static inline int ip6_ufo_append_data(struct sock *sk,
1089 struct sk_buff_head *queue,
1090 int getfrag(void *from, char *to, int offset, int len,
1091 int odd, struct sk_buff *skb),
1092 void *from, int length, int hh_len, int fragheaderlen,
1093 int exthdrlen, int transhdrlen, int mtu,
1094 unsigned int flags, const struct flowi6 *fl6)
1095
1096{
1097 struct sk_buff *skb;
1098 int err;
1099
1100 /* There is support for UDP large send offload by network
1101 * device, so create one single skb packet containing complete
1102 * udp datagram
1103 */
1104 skb = skb_peek_tail(queue);
1105 if (!skb) {
1106 skb = sock_alloc_send_skb(sk,
1107 hh_len + fragheaderlen + transhdrlen + 20,
1108 (flags & MSG_DONTWAIT), &err);
1109 if (!skb)
1110 return err;
1111
1112 /* reserve space for Hardware header */
1113 skb_reserve(skb, hh_len);
1114
1115 /* create space for UDP/IP header */
1116 skb_put(skb, fragheaderlen + transhdrlen);
1117
1118 /* initialize network header pointer */
1119 skb_set_network_header(skb, exthdrlen);
1120
1121 /* initialize protocol header pointer */
1122 skb->transport_header = skb->network_header + fragheaderlen;
1123
1124 skb->protocol = htons(ETH_P_IPV6);
1125 skb->csum = 0;
1126
1127 __skb_queue_tail(queue, skb);
1128 } else if (skb_is_gso(skb)) {
1129 goto append;
1130 }
1131
1132 skb->ip_summed = CHECKSUM_PARTIAL;
1133 /* Specify the length of each IPv6 datagram fragment.
1134 * It has to be a multiple of 8.
1135 */
1136 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1137 sizeof(struct frag_hdr)) & ~7;
1138 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1139 skb_shinfo(skb)->ip6_frag_id = ipv6_select_ident(sock_net(sk),
1140 &fl6->daddr,
1141 &fl6->saddr);
1142
1143append:
1144 return skb_append_datato_frags(sk, skb, getfrag, from,
1145 (length - transhdrlen));
1146}
1147
1148static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1149 gfp_t gfp)
1150{
1151 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1152}
1153
1154static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1155 gfp_t gfp)
1156{
1157 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1158}
1159
1160static void ip6_append_data_mtu(unsigned int *mtu,
1161 int *maxfraglen,
1162 unsigned int fragheaderlen,
1163 struct sk_buff *skb,
1164 struct rt6_info *rt,
1165 unsigned int orig_mtu)
1166{
1167 if (!(rt->dst.flags & DST_XFRM_TUNNEL)) {
1168 if (!skb) {
1169 /* first fragment, reserve header_len */
1170 *mtu = orig_mtu - rt->dst.header_len;
1171
1172 } else {
1173 /*
1174 * this fragment is not first, the headers
1175 * space is regarded as data space.
1176 */
1177 *mtu = orig_mtu;
1178 }
1179 *maxfraglen = ((*mtu - fragheaderlen) & ~7)
1180 + fragheaderlen - sizeof(struct frag_hdr);
1181 }
1182}
1183
1184static int ip6_setup_cork(struct sock *sk, struct inet_cork_full *cork,
1185 struct inet6_cork *v6_cork,
1186 int hlimit, int tclass, struct ipv6_txoptions *opt,
1187 struct rt6_info *rt, struct flowi6 *fl6)
1188{
1189 struct ipv6_pinfo *np = inet6_sk(sk);
1190 unsigned int mtu;
1191
1192 /*
1193 * setup for corking
1194 */
1195 if (opt) {
1196 if (WARN_ON(v6_cork->opt))
1197 return -EINVAL;
1198
1199 v6_cork->opt = kzalloc(opt->tot_len, sk->sk_allocation);
1200 if (unlikely(!v6_cork->opt))
1201 return -ENOBUFS;
1202
1203 v6_cork->opt->tot_len = opt->tot_len;
1204 v6_cork->opt->opt_flen = opt->opt_flen;
1205 v6_cork->opt->opt_nflen = opt->opt_nflen;
1206
1207 v6_cork->opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1208 sk->sk_allocation);
1209 if (opt->dst0opt && !v6_cork->opt->dst0opt)
1210 return -ENOBUFS;
1211
1212 v6_cork->opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1213 sk->sk_allocation);
1214 if (opt->dst1opt && !v6_cork->opt->dst1opt)
1215 return -ENOBUFS;
1216
1217 v6_cork->opt->hopopt = ip6_opt_dup(opt->hopopt,
1218 sk->sk_allocation);
1219 if (opt->hopopt && !v6_cork->opt->hopopt)
1220 return -ENOBUFS;
1221
1222 v6_cork->opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1223 sk->sk_allocation);
1224 if (opt->srcrt && !v6_cork->opt->srcrt)
1225 return -ENOBUFS;
1226
1227 /* need source address above miyazawa*/
1228 }
1229 dst_hold(&rt->dst);
1230 cork->base.dst = &rt->dst;
1231 cork->fl.u.ip6 = *fl6;
1232 v6_cork->hop_limit = hlimit;
1233 v6_cork->tclass = tclass;
1234 if (rt->dst.flags & DST_XFRM_TUNNEL)
1235 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1236 rt->dst.dev->mtu : dst_mtu(&rt->dst);
1237 else
1238 mtu = np->pmtudisc >= IPV6_PMTUDISC_PROBE ?
1239 rt->dst.dev->mtu : dst_mtu(rt->dst.path);
1240 if (np->frag_size < mtu) {
1241 if (np->frag_size)
1242 mtu = np->frag_size;
1243 }
1244 cork->base.fragsize = mtu;
1245 if (dst_allfrag(rt->dst.path))
1246 cork->base.flags |= IPCORK_ALLFRAG;
1247 cork->base.length = 0;
1248
1249 return 0;
1250}
1251
1252static int __ip6_append_data(struct sock *sk,
1253 struct flowi6 *fl6,
1254 struct sk_buff_head *queue,
1255 struct inet_cork *cork,
1256 struct inet6_cork *v6_cork,
1257 struct page_frag *pfrag,
1258 int getfrag(void *from, char *to, int offset,
1259 int len, int odd, struct sk_buff *skb),
1260 void *from, int length, int transhdrlen,
1261 unsigned int flags, int dontfrag)
1262{
1263 struct sk_buff *skb, *skb_prev = NULL;
1264 unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu;
1265 int exthdrlen = 0;
1266 int dst_exthdrlen = 0;
1267 int hh_len;
1268 int copy;
1269 int err;
1270 int offset = 0;
1271 __u8 tx_flags = 0;
1272 u32 tskey = 0;
1273 struct rt6_info *rt = (struct rt6_info *)cork->dst;
1274 struct ipv6_txoptions *opt = v6_cork->opt;
1275 int csummode = CHECKSUM_NONE;
1276 unsigned int maxnonfragsize, headersize;
1277
1278 skb = skb_peek_tail(queue);
1279 if (!skb) {
1280 exthdrlen = opt ? opt->opt_flen : 0;
1281 dst_exthdrlen = rt->dst.header_len - rt->rt6i_nfheader_len;
1282 }
1283
1284 mtu = cork->fragsize;
1285 orig_mtu = mtu;
1286
1287 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1288
1289 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
1290 (opt ? opt->opt_nflen : 0);
1291 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen -
1292 sizeof(struct frag_hdr);
1293
1294 headersize = sizeof(struct ipv6hdr) +
1295 (opt ? opt->opt_flen + opt->opt_nflen : 0) +
1296 (dst_allfrag(&rt->dst) ?
1297 sizeof(struct frag_hdr) : 0) +
1298 rt->rt6i_nfheader_len;
1299
1300 if (cork->length + length > mtu - headersize && dontfrag &&
1301 (sk->sk_protocol == IPPROTO_UDP ||
1302 sk->sk_protocol == IPPROTO_RAW)) {
1303 ipv6_local_rxpmtu(sk, fl6, mtu - headersize +
1304 sizeof(struct ipv6hdr));
1305 goto emsgsize;
1306 }
1307
1308 if (ip6_sk_ignore_df(sk))
1309 maxnonfragsize = sizeof(struct ipv6hdr) + IPV6_MAXPLEN;
1310 else
1311 maxnonfragsize = mtu;
1312
1313 if (cork->length + length > maxnonfragsize - headersize) {
1314emsgsize:
1315 ipv6_local_error(sk, EMSGSIZE, fl6,
1316 mtu - headersize +
1317 sizeof(struct ipv6hdr));
1318 return -EMSGSIZE;
1319 }
1320
1321 /* CHECKSUM_PARTIAL only with no extension headers and when
1322 * we are not going to fragment
1323 */
1324 if (transhdrlen && sk->sk_protocol == IPPROTO_UDP &&
1325 headersize == sizeof(struct ipv6hdr) &&
1326 length < mtu - headersize &&
1327 !(flags & MSG_MORE) &&
1328 rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
1329 csummode = CHECKSUM_PARTIAL;
1330
1331 if (sk->sk_type == SOCK_DGRAM || sk->sk_type == SOCK_RAW) {
1332 sock_tx_timestamp(sk, &tx_flags);
1333 if (tx_flags & SKBTX_ANY_SW_TSTAMP &&
1334 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1335 tskey = sk->sk_tskey++;
1336 }
1337
1338 /*
1339 * Let's try using as much space as possible.
1340 * Use MTU if total length of the message fits into the MTU.
1341 * Otherwise, we need to reserve fragment header and
1342 * fragment alignment (= 8-15 octects, in total).
1343 *
1344 * Note that we may need to "move" the data from the tail of
1345 * of the buffer to the new fragment when we split
1346 * the message.
1347 *
1348 * FIXME: It may be fragmented into multiple chunks
1349 * at once if non-fragmentable extension headers
1350 * are too large.
1351 * --yoshfuji
1352 */
1353
1354 cork->length += length;
1355 if (((length > mtu) ||
1356 (skb && skb_is_gso(skb))) &&
1357 (sk->sk_protocol == IPPROTO_UDP) &&
1358 (rt->dst.dev->features & NETIF_F_UFO) &&
1359 (sk->sk_type == SOCK_DGRAM) && !udp_get_no_check6_tx(sk)) {
1360 err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
1361 hh_len, fragheaderlen, exthdrlen,
1362 transhdrlen, mtu, flags, fl6);
1363 if (err)
1364 goto error;
1365 return 0;
1366 }
1367
1368 if (!skb)
1369 goto alloc_new_skb;
1370
1371 while (length > 0) {
1372 /* Check if the remaining data fits into current packet. */
1373 copy = (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1374 if (copy < length)
1375 copy = maxfraglen - skb->len;
1376
1377 if (copy <= 0) {
1378 char *data;
1379 unsigned int datalen;
1380 unsigned int fraglen;
1381 unsigned int fraggap;
1382 unsigned int alloclen;
1383alloc_new_skb:
1384 /* There's no room in the current skb */
1385 if (skb)
1386 fraggap = skb->len - maxfraglen;
1387 else
1388 fraggap = 0;
1389 /* update mtu and maxfraglen if necessary */
1390 if (!skb || !skb_prev)
1391 ip6_append_data_mtu(&mtu, &maxfraglen,
1392 fragheaderlen, skb, rt,
1393 orig_mtu);
1394
1395 skb_prev = skb;
1396
1397 /*
1398 * If remaining data exceeds the mtu,
1399 * we know we need more fragment(s).
1400 */
1401 datalen = length + fraggap;
1402
1403 if (datalen > (cork->length <= mtu && !(cork->flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1404 datalen = maxfraglen - fragheaderlen - rt->dst.trailer_len;
1405 if ((flags & MSG_MORE) &&
1406 !(rt->dst.dev->features&NETIF_F_SG))
1407 alloclen = mtu;
1408 else
1409 alloclen = datalen + fragheaderlen;
1410
1411 alloclen += dst_exthdrlen;
1412
1413 if (datalen != length + fraggap) {
1414 /*
1415 * this is not the last fragment, the trailer
1416 * space is regarded as data space.
1417 */
1418 datalen += rt->dst.trailer_len;
1419 }
1420
1421 alloclen += rt->dst.trailer_len;
1422 fraglen = datalen + fragheaderlen;
1423
1424 /*
1425 * We just reserve space for fragment header.
1426 * Note: this may be overallocation if the message
1427 * (without MSG_MORE) fits into the MTU.
1428 */
1429 alloclen += sizeof(struct frag_hdr);
1430
1431 if (transhdrlen) {
1432 skb = sock_alloc_send_skb(sk,
1433 alloclen + hh_len,
1434 (flags & MSG_DONTWAIT), &err);
1435 } else {
1436 skb = NULL;
1437 if (atomic_read(&sk->sk_wmem_alloc) <=
1438 2 * sk->sk_sndbuf)
1439 skb = sock_wmalloc(sk,
1440 alloclen + hh_len, 1,
1441 sk->sk_allocation);
1442 if (unlikely(!skb))
1443 err = -ENOBUFS;
1444 }
1445 if (!skb)
1446 goto error;
1447 /*
1448 * Fill in the control structures
1449 */
1450 skb->protocol = htons(ETH_P_IPV6);
1451 skb->ip_summed = csummode;
1452 skb->csum = 0;
1453 /* reserve for fragmentation and ipsec header */
1454 skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
1455 dst_exthdrlen);
1456
1457 /* Only the initial fragment is time stamped */
1458 skb_shinfo(skb)->tx_flags = tx_flags;
1459 tx_flags = 0;
1460 skb_shinfo(skb)->tskey = tskey;
1461 tskey = 0;
1462
1463 /*
1464 * Find where to start putting bytes
1465 */
1466 data = skb_put(skb, fraglen);
1467 skb_set_network_header(skb, exthdrlen);
1468 data += fragheaderlen;
1469 skb->transport_header = (skb->network_header +
1470 fragheaderlen);
1471 if (fraggap) {
1472 skb->csum = skb_copy_and_csum_bits(
1473 skb_prev, maxfraglen,
1474 data + transhdrlen, fraggap, 0);
1475 skb_prev->csum = csum_sub(skb_prev->csum,
1476 skb->csum);
1477 data += fraggap;
1478 pskb_trim_unique(skb_prev, maxfraglen);
1479 }
1480 copy = datalen - transhdrlen - fraggap;
1481
1482 if (copy < 0) {
1483 err = -EINVAL;
1484 kfree_skb(skb);
1485 goto error;
1486 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1487 err = -EFAULT;
1488 kfree_skb(skb);
1489 goto error;
1490 }
1491
1492 offset += copy;
1493 length -= datalen - fraggap;
1494 transhdrlen = 0;
1495 exthdrlen = 0;
1496 dst_exthdrlen = 0;
1497
1498 /*
1499 * Put the packet on the pending queue
1500 */
1501 __skb_queue_tail(queue, skb);
1502 continue;
1503 }
1504
1505 if (copy > length)
1506 copy = length;
1507
1508 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1509 unsigned int off;
1510
1511 off = skb->len;
1512 if (getfrag(from, skb_put(skb, copy),
1513 offset, copy, off, skb) < 0) {
1514 __skb_trim(skb, off);
1515 err = -EFAULT;
1516 goto error;
1517 }
1518 } else {
1519 int i = skb_shinfo(skb)->nr_frags;
1520
1521 err = -ENOMEM;
1522 if (!sk_page_frag_refill(sk, pfrag))
1523 goto error;
1524
1525 if (!skb_can_coalesce(skb, i, pfrag->page,
1526 pfrag->offset)) {
1527 err = -EMSGSIZE;
1528 if (i == MAX_SKB_FRAGS)
1529 goto error;
1530
1531 __skb_fill_page_desc(skb, i, pfrag->page,
1532 pfrag->offset, 0);
1533 skb_shinfo(skb)->nr_frags = ++i;
1534 get_page(pfrag->page);
1535 }
1536 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1537 if (getfrag(from,
1538 page_address(pfrag->page) + pfrag->offset,
1539 offset, copy, skb->len, skb) < 0)
1540 goto error_efault;
1541
1542 pfrag->offset += copy;
1543 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1544 skb->len += copy;
1545 skb->data_len += copy;
1546 skb->truesize += copy;
1547 atomic_add(copy, &sk->sk_wmem_alloc);
1548 }
1549 offset += copy;
1550 length -= copy;
1551 }
1552
1553 return 0;
1554
1555error_efault:
1556 err = -EFAULT;
1557error:
1558 cork->length -= length;
1559 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
1560 return err;
1561}
1562
1563int ip6_append_data(struct sock *sk,
1564 int getfrag(void *from, char *to, int offset, int len,
1565 int odd, struct sk_buff *skb),
1566 void *from, int length, int transhdrlen, int hlimit,
1567 int tclass, struct ipv6_txoptions *opt, struct flowi6 *fl6,
1568 struct rt6_info *rt, unsigned int flags, int dontfrag)
1569{
1570 struct inet_sock *inet = inet_sk(sk);
1571 struct ipv6_pinfo *np = inet6_sk(sk);
1572 int exthdrlen;
1573 int err;
1574
1575 if (flags&MSG_PROBE)
1576 return 0;
1577 if (skb_queue_empty(&sk->sk_write_queue)) {
1578 /*
1579 * setup for corking
1580 */
1581 err = ip6_setup_cork(sk, &inet->cork, &np->cork, hlimit,
1582 tclass, opt, rt, fl6);
1583 if (err)
1584 return err;
1585
1586 exthdrlen = (opt ? opt->opt_flen : 0);
1587 length += exthdrlen;
1588 transhdrlen += exthdrlen;
1589 } else {
1590 fl6 = &inet->cork.fl.u.ip6;
1591 transhdrlen = 0;
1592 }
1593
1594 return __ip6_append_data(sk, fl6, &sk->sk_write_queue, &inet->cork.base,
1595 &np->cork, sk_page_frag(sk), getfrag,
1596 from, length, transhdrlen, flags, dontfrag);
1597}
1598EXPORT_SYMBOL_GPL(ip6_append_data);
1599
1600static void ip6_cork_release(struct inet_cork_full *cork,
1601 struct inet6_cork *v6_cork)
1602{
1603 if (v6_cork->opt) {
1604 kfree(v6_cork->opt->dst0opt);
1605 kfree(v6_cork->opt->dst1opt);
1606 kfree(v6_cork->opt->hopopt);
1607 kfree(v6_cork->opt->srcrt);
1608 kfree(v6_cork->opt);
1609 v6_cork->opt = NULL;
1610 }
1611
1612 if (cork->base.dst) {
1613 dst_release(cork->base.dst);
1614 cork->base.dst = NULL;
1615 cork->base.flags &= ~IPCORK_ALLFRAG;
1616 }
1617 memset(&cork->fl, 0, sizeof(cork->fl));
1618}
1619
1620struct sk_buff *__ip6_make_skb(struct sock *sk,
1621 struct sk_buff_head *queue,
1622 struct inet_cork_full *cork,
1623 struct inet6_cork *v6_cork)
1624{
1625 struct sk_buff *skb, *tmp_skb;
1626 struct sk_buff **tail_skb;
1627 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1628 struct ipv6_pinfo *np = inet6_sk(sk);
1629 struct net *net = sock_net(sk);
1630 struct ipv6hdr *hdr;
1631 struct ipv6_txoptions *opt = v6_cork->opt;
1632 struct rt6_info *rt = (struct rt6_info *)cork->base.dst;
1633 struct flowi6 *fl6 = &cork->fl.u.ip6;
1634 unsigned char proto = fl6->flowi6_proto;
1635
1636 skb = __skb_dequeue(queue);
1637 if (!skb)
1638 goto out;
1639 tail_skb = &(skb_shinfo(skb)->frag_list);
1640
1641 /* move skb->data to ip header from ext header */
1642 if (skb->data < skb_network_header(skb))
1643 __skb_pull(skb, skb_network_offset(skb));
1644 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1645 __skb_pull(tmp_skb, skb_network_header_len(skb));
1646 *tail_skb = tmp_skb;
1647 tail_skb = &(tmp_skb->next);
1648 skb->len += tmp_skb->len;
1649 skb->data_len += tmp_skb->len;
1650 skb->truesize += tmp_skb->truesize;
1651 tmp_skb->destructor = NULL;
1652 tmp_skb->sk = NULL;
1653 }
1654
1655 /* Allow local fragmentation. */
1656 skb->ignore_df = ip6_sk_ignore_df(sk);
1657
1658 *final_dst = fl6->daddr;
1659 __skb_pull(skb, skb_network_header_len(skb));
1660 if (opt && opt->opt_flen)
1661 ipv6_push_frag_opts(skb, opt, &proto);
1662 if (opt && opt->opt_nflen)
1663 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1664
1665 skb_push(skb, sizeof(struct ipv6hdr));
1666 skb_reset_network_header(skb);
1667 hdr = ipv6_hdr(skb);
1668
1669 ip6_flow_hdr(hdr, v6_cork->tclass,
1670 ip6_make_flowlabel(net, skb, fl6->flowlabel,
1671 np->autoflowlabel, fl6));
1672 hdr->hop_limit = v6_cork->hop_limit;
1673 hdr->nexthdr = proto;
1674 hdr->saddr = fl6->saddr;
1675 hdr->daddr = *final_dst;
1676
1677 skb->priority = sk->sk_priority;
1678 skb->mark = sk->sk_mark;
1679
1680 skb_dst_set(skb, dst_clone(&rt->dst));
1681 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
1682 if (proto == IPPROTO_ICMPV6) {
1683 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1684
1685 ICMP6MSGOUT_INC_STATS(net, idev, icmp6_hdr(skb)->icmp6_type);
1686 ICMP6_INC_STATS(net, idev, ICMP6_MIB_OUTMSGS);
1687 }
1688
1689 ip6_cork_release(cork, v6_cork);
1690out:
1691 return skb;
1692}
1693
1694int ip6_send_skb(struct sk_buff *skb)
1695{
1696 struct net *net = sock_net(skb->sk);
1697 struct rt6_info *rt = (struct rt6_info *)skb_dst(skb);
1698 int err;
1699
1700 err = ip6_local_out(net, skb->sk, skb);
1701 if (err) {
1702 if (err > 0)
1703 err = net_xmit_errno(err);
1704 if (err)
1705 IP6_INC_STATS(net, rt->rt6i_idev,
1706 IPSTATS_MIB_OUTDISCARDS);
1707 }
1708
1709 return err;
1710}
1711
1712int ip6_push_pending_frames(struct sock *sk)
1713{
1714 struct sk_buff *skb;
1715
1716 skb = ip6_finish_skb(sk);
1717 if (!skb)
1718 return 0;
1719
1720 return ip6_send_skb(skb);
1721}
1722EXPORT_SYMBOL_GPL(ip6_push_pending_frames);
1723
1724static void __ip6_flush_pending_frames(struct sock *sk,
1725 struct sk_buff_head *queue,
1726 struct inet_cork_full *cork,
1727 struct inet6_cork *v6_cork)
1728{
1729 struct sk_buff *skb;
1730
1731 while ((skb = __skb_dequeue_tail(queue)) != NULL) {
1732 if (skb_dst(skb))
1733 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
1734 IPSTATS_MIB_OUTDISCARDS);
1735 kfree_skb(skb);
1736 }
1737
1738 ip6_cork_release(cork, v6_cork);
1739}
1740
1741void ip6_flush_pending_frames(struct sock *sk)
1742{
1743 __ip6_flush_pending_frames(sk, &sk->sk_write_queue,
1744 &inet_sk(sk)->cork, &inet6_sk(sk)->cork);
1745}
1746EXPORT_SYMBOL_GPL(ip6_flush_pending_frames);
1747
1748struct sk_buff *ip6_make_skb(struct sock *sk,
1749 int getfrag(void *from, char *to, int offset,
1750 int len, int odd, struct sk_buff *skb),
1751 void *from, int length, int transhdrlen,
1752 int hlimit, int tclass,
1753 struct ipv6_txoptions *opt, struct flowi6 *fl6,
1754 struct rt6_info *rt, unsigned int flags,
1755 int dontfrag)
1756{
1757 struct inet_cork_full cork;
1758 struct inet6_cork v6_cork;
1759 struct sk_buff_head queue;
1760 int exthdrlen = (opt ? opt->opt_flen : 0);
1761 int err;
1762
1763 if (flags & MSG_PROBE)
1764 return NULL;
1765
1766 __skb_queue_head_init(&queue);
1767
1768 cork.base.flags = 0;
1769 cork.base.addr = 0;
1770 cork.base.opt = NULL;
1771 v6_cork.opt = NULL;
1772 err = ip6_setup_cork(sk, &cork, &v6_cork, hlimit, tclass, opt, rt, fl6);
1773 if (err)
1774 return ERR_PTR(err);
1775
1776 if (dontfrag < 0)
1777 dontfrag = inet6_sk(sk)->dontfrag;
1778
1779 err = __ip6_append_data(sk, fl6, &queue, &cork.base, &v6_cork,
1780 ¤t->task_frag, getfrag, from,
1781 length + exthdrlen, transhdrlen + exthdrlen,
1782 flags, dontfrag);
1783 if (err) {
1784 __ip6_flush_pending_frames(sk, &queue, &cork, &v6_cork);
1785 return ERR_PTR(err);
1786 }
1787
1788 return __ip6_make_skb(sk, &queue, &cork, &v6_cork);
1789}
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}