blob: cd48801a8d6f465d546b61a2e9629bf6e2091714 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * IPv6 output functions
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09003 * Linux INET6 implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * Authors:
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09006 * Pedro Roque <roque@di.fc.ul.pt>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/errno.h>
Herbert Xuef76bc22008-01-11 19:15:08 -080030#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#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>
Herbert Xub59f45d2006-05-27 23:05:54 -070039#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#include <linux/netfilter.h>
42#include <linux/netfilter_ipv6.h>
43
44#include <net/sock.h>
45#include <net/snmp.h>
46
47#include <net/ipv6.h>
48#include <net/ndisc.h>
49#include <net/protocol.h>
50#include <net/ip6_route.h>
51#include <net/addrconf.h>
52#include <net/rawv6.h>
53#include <net/icmp.h>
54#include <net/xfrm.h>
55#include <net/checksum.h>
YOSHIFUJI Hideaki7bc570c2008-04-03 09:22:53 +090056#include <linux/mroute6.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
59
Herbert Xuef76bc22008-01-11 19:15:08 -080060int __ip6_local_out(struct sk_buff *skb)
61{
62 int len;
63
64 len = skb->len - sizeof(struct ipv6hdr);
65 if (len > IPV6_MAXPLEN)
66 len = 0;
67 ipv6_hdr(skb)->payload_len = htons(len);
68
Eric Dumazetadf30902009-06-02 05:19:30 +000069 return nf_hook(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
Herbert Xuef76bc22008-01-11 19:15:08 -080070 dst_output);
71}
72
73int ip6_local_out(struct sk_buff *skb)
74{
75 int err;
76
77 err = __ip6_local_out(skb);
78 if (likely(err == 1))
79 err = dst_output(skb);
80
81 return err;
82}
83EXPORT_SYMBOL_GPL(ip6_local_out);
84
Patrick McHardyad643a72007-10-15 01:51:38 -070085static int ip6_output_finish(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Eric Dumazetadf30902009-06-02 05:19:30 +000087 struct dst_entry *dst = skb_dst(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Stephen Hemminger3644f0c2006-12-07 15:08:17 -080089 if (dst->hh)
90 return neigh_hh_output(dst->hh, skb);
91 else if (dst->neighbour)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 return dst->neighbour->output(skb);
93
Denis V. Lunev483a47d2008-10-08 11:09:27 -070094 IP6_INC_STATS_BH(dev_net(dst->dev),
95 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 kfree_skb(skb);
97 return -EINVAL;
98
99}
100
101/* dev_loopback_xmit for use with netfilter. */
102static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
103{
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -0700104 skb_reset_mac_header(newskb);
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -0300105 __skb_pull(newskb, skb_network_offset(newskb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 newskb->pkt_type = PACKET_LOOPBACK;
107 newskb->ip_summed = CHECKSUM_UNNECESSARY;
Eric Dumazetadf30902009-06-02 05:19:30 +0000108 WARN_ON(!skb_dst(newskb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110 netif_rx(newskb);
111 return 0;
112}
113
114
115static int ip6_output2(struct sk_buff *skb)
116{
Eric Dumazetadf30902009-06-02 05:19:30 +0000117 struct dst_entry *dst = skb_dst(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 struct net_device *dev = dst->dev;
119
120 skb->protocol = htons(ETH_P_IPV6);
121 skb->dev = dev;
122
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700123 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
Eric Dumazetadf30902009-06-02 05:19:30 +0000125 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
Benjamin Therybd91b8b2008-12-10 16:07:08 -0800128 ((mroute6_socket(dev_net(dev)) &&
129 !(IP6CB(skb)->flags & IP6SKB_FORWARDED)) ||
YOSHIFUJI Hideaki7bc570c2008-04-03 09:22:53 +0900130 ipv6_chk_mcast_addr(dev, &ipv6_hdr(skb)->daddr,
131 &ipv6_hdr(skb)->saddr))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
133
134 /* Do not check for IFF_ALLMULTI; multicast routing
135 is not supported in any case.
136 */
137 if (newskb)
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800138 NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, newskb,
139 NULL, newskb->dev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 ip6_dev_loopback_xmit);
141
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700142 if (ipv6_hdr(skb)->hop_limit == 0) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700143 IP6_INC_STATS(dev_net(dev), idev,
144 IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 kfree_skb(skb);
146 return 0;
147 }
148 }
149
Neil Hormanedf391f2009-04-27 02:45:02 -0700150 IP6_UPD_PO_STATS(dev_net(dev), idev, IPSTATS_MIB_OUTMCAST,
151 skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 }
153
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800154 return NF_HOOK(PF_INET6, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
155 ip6_output_finish);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
John Heffner628a5c52007-04-20 15:53:27 -0700158static inline int ip6_skb_dst_mtu(struct sk_buff *skb)
159{
160 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
161
162 return (np && np->pmtudisc == IPV6_PMTUDISC_PROBE) ?
Eric Dumazetadf30902009-06-02 05:19:30 +0000163 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
John Heffner628a5c52007-04-20 15:53:27 -0700164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166int ip6_output(struct sk_buff *skb)
167{
Eric Dumazetadf30902009-06-02 05:19:30 +0000168 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
YOSHIFUJI Hideaki778d80b2008-06-28 14:17:11 +0900169 if (unlikely(idev->cnf.disable_ipv6)) {
Eric Dumazetadf30902009-06-02 05:19:30 +0000170 IP6_INC_STATS(dev_net(skb_dst(skb)->dev), idev,
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700171 IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideaki778d80b2008-06-28 14:17:11 +0900172 kfree_skb(skb);
173 return 0;
174 }
175
John Heffner628a5c52007-04-20 15:53:27 -0700176 if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
Eric Dumazetadf30902009-06-02 05:19:30 +0000177 dst_allfrag(skb_dst(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 return ip6_fragment(skb, ip6_output2);
179 else
180 return ip6_output2(skb);
181}
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/*
184 * xmit an sk_buff (used by TCP)
185 */
186
187int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
188 struct ipv6_txoptions *opt, int ipfragok)
189{
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700190 struct net *net = sock_net(sk);
Patrick McHardyb30bd282006-03-23 01:17:25 -0800191 struct ipv6_pinfo *np = inet6_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 struct in6_addr *first_hop = &fl->fl6_dst;
Eric Dumazetadf30902009-06-02 05:19:30 +0000193 struct dst_entry *dst = skb_dst(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 struct ipv6hdr *hdr;
195 u8 proto = fl->proto;
196 int seg_len = skb->len;
Gerrit Renkere651f032009-08-09 08:12:48 +0000197 int hlimit = -1;
198 int tclass = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 u32 mtu;
200
201 if (opt) {
Chuck Leverc2636b42007-10-23 21:07:32 -0700202 unsigned int head_room;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 /* First: exthdrs may take lots of space (~8K for now)
205 MAX_HEADER is not enough.
206 */
207 head_room = opt->opt_nflen + opt->opt_flen;
208 seg_len += head_room;
209 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
210
211 if (skb_headroom(skb) < head_room) {
212 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900213 if (skb2 == NULL) {
Eric Dumazetadf30902009-06-02 05:19:30 +0000214 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900215 IPSTATS_MIB_OUTDISCARDS);
216 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 return -ENOBUFS;
218 }
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900219 kfree_skb(skb);
220 skb = skb2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 if (sk)
222 skb_set_owner_w(skb, sk);
223 }
224 if (opt->opt_flen)
225 ipv6_push_frag_opts(skb, opt, &proto);
226 if (opt->opt_nflen)
227 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
228 }
229
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700230 skb_push(skb, sizeof(struct ipv6hdr));
231 skb_reset_network_header(skb);
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700232 hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Wei Yongjun77e2f142008-07-31 20:46:47 -0700234 /* Allow local fragmentation. */
235 if (ipfragok)
236 skb->local_df = 1;
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 /*
239 * Fill in the IPv6 header
240 */
Gerrit Renkere651f032009-08-09 08:12:48 +0000241 if (np) {
242 tclass = np->tclass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 hlimit = np->hop_limit;
Gerrit Renkere651f032009-08-09 08:12:48 +0000244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 if (hlimit < 0)
YOSHIFUJI Hideaki6b75d092008-03-10 06:00:30 -0400246 hlimit = ip6_dst_hoplimit(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Al Viro90bcaf72006-11-08 00:25:17 -0800248 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +0900249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 hdr->payload_len = htons(seg_len);
251 hdr->nexthdr = proto;
252 hdr->hop_limit = hlimit;
253
254 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
255 ipv6_addr_copy(&hdr->daddr, first_hop);
256
Patrick McHardya2c20642006-01-08 22:37:26 -0800257 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800258 skb->mark = sk->sk_mark;
Patrick McHardya2c20642006-01-08 22:37:26 -0800259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 mtu = dst_mtu(dst);
Wei Yongjun283d07ac2008-08-03 21:15:59 -0700261 if ((skb->len <= mtu) || skb->local_df || skb_is_gso(skb)) {
Eric Dumazetadf30902009-06-02 05:19:30 +0000262 IP6_UPD_PO_STATS(net, ip6_dst_idev(skb_dst(skb)),
Neil Hormanedf391f2009-04-27 02:45:02 -0700263 IPSTATS_MIB_OUT, skb->len);
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800264 return NF_HOOK(PF_INET6, NF_INET_LOCAL_OUT, skb, NULL, dst->dev,
Harald Welte6869c4d2005-08-09 19:24:19 -0700265 dst_output);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
267
268 if (net_ratelimit())
269 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
270 skb->dev = dst->dev;
271 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
Eric Dumazetadf30902009-06-02 05:19:30 +0000272 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 kfree_skb(skb);
274 return -EMSGSIZE;
275}
276
YOSHIFUJI Hideaki71590392007-02-22 22:05:40 +0900277EXPORT_SYMBOL(ip6_xmit);
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279/*
280 * To avoid extra problems ND packets are send through this
281 * routine. It's code duplication but I really want to avoid
282 * extra checks since ipv6_build_header is used by TCP (which
283 * is for us performance critical)
284 */
285
286int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
YOSHIFUJI Hideaki9acd9f32008-04-10 15:42:10 +0900287 const struct in6_addr *saddr, const struct in6_addr *daddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 int proto, int len)
289{
290 struct ipv6_pinfo *np = inet6_sk(sk);
291 struct ipv6hdr *hdr;
292 int totlen;
293
294 skb->protocol = htons(ETH_P_IPV6);
295 skb->dev = dev;
296
297 totlen = len + sizeof(struct ipv6hdr);
298
Arnaldo Carvalho de Melo55f79cc2007-03-14 21:05:03 -0300299 skb_reset_network_header(skb);
300 skb_put(skb, sizeof(struct ipv6hdr));
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700301 hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Al Viroae08e1f2006-11-08 00:27:11 -0800303 *(__be32*)hdr = htonl(0x60000000);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
305 hdr->payload_len = htons(len);
306 hdr->nexthdr = proto;
307 hdr->hop_limit = np->hop_limit;
308
309 ipv6_addr_copy(&hdr->saddr, saddr);
310 ipv6_addr_copy(&hdr->daddr, daddr);
311
312 return 0;
313}
314
315static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
316{
317 struct ip6_ra_chain *ra;
318 struct sock *last = NULL;
319
320 read_lock(&ip6_ra_lock);
321 for (ra = ip6_ra_chain; ra; ra = ra->next) {
322 struct sock *sk = ra->sk;
Andrew McDonald0bd1b592005-08-09 19:44:42 -0700323 if (sk && ra->sel == sel &&
324 (!sk->sk_bound_dev_if ||
325 sk->sk_bound_dev_if == skb->dev->ifindex)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 if (last) {
327 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
328 if (skb2)
329 rawv6_rcv(last, skb2);
330 }
331 last = sk;
332 }
333 }
334
335 if (last) {
336 rawv6_rcv(last, skb);
337 read_unlock(&ip6_ra_lock);
338 return 1;
339 }
340 read_unlock(&ip6_ra_lock);
341 return 0;
342}
343
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700344static int ip6_forward_proxy_check(struct sk_buff *skb)
345{
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700346 struct ipv6hdr *hdr = ipv6_hdr(skb);
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700347 u8 nexthdr = hdr->nexthdr;
348 int offset;
349
350 if (ipv6_ext_hdr(nexthdr)) {
351 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
352 if (offset < 0)
353 return 0;
354 } else
355 offset = sizeof(struct ipv6hdr);
356
357 if (nexthdr == IPPROTO_ICMPV6) {
358 struct icmp6hdr *icmp6;
359
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700360 if (!pskb_may_pull(skb, (skb_network_header(skb) +
361 offset + 1 - skb->data)))
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700362 return 0;
363
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700364 icmp6 = (struct icmp6hdr *)(skb_network_header(skb) + offset);
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700365
366 switch (icmp6->icmp6_type) {
367 case NDISC_ROUTER_SOLICITATION:
368 case NDISC_ROUTER_ADVERTISEMENT:
369 case NDISC_NEIGHBOUR_SOLICITATION:
370 case NDISC_NEIGHBOUR_ADVERTISEMENT:
371 case NDISC_REDIRECT:
372 /* For reaction involving unicast neighbor discovery
373 * message destined to the proxied address, pass it to
374 * input function.
375 */
376 return 1;
377 default:
378 break;
379 }
380 }
381
Ville Nuorvala74553b02006-09-22 14:42:18 -0700382 /*
383 * The proxying router can't forward traffic sent to a link-local
384 * address, so signal the sender and discard the packet. This
385 * behavior is clarified by the MIPv6 specification.
386 */
387 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
388 dst_link_failure(skb);
389 return -1;
390 }
391
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700392 return 0;
393}
394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395static inline int ip6_forward_finish(struct sk_buff *skb)
396{
397 return dst_output(skb);
398}
399
400int ip6_forward(struct sk_buff *skb)
401{
Eric Dumazetadf30902009-06-02 05:19:30 +0000402 struct dst_entry *dst = skb_dst(skb);
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700403 struct ipv6hdr *hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 struct inet6_skb_parm *opt = IP6CB(skb);
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +0900405 struct net *net = dev_net(dst->dev);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900406
YOSHIFUJI Hideaki53b79972008-07-19 22:35:03 -0700407 if (net->ipv6.devconf_all->forwarding == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 goto error;
409
Ben Hutchings4497b072008-06-19 16:22:28 -0700410 if (skb_warn_if_lro(skb))
411 goto drop;
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700414 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 goto drop;
416 }
417
Herbert Xu35fc92a2007-03-26 23:22:20 -0700418 skb_forward_csum(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 /*
421 * We DO NOT make any processing on
422 * RA packets, pushing them to user level AS IS
423 * without ane WARRANTY that application will be able
424 * to interpret them. The reason is that we
425 * cannot make anything clever here.
426 *
427 * We are not end-node, so that if packet contains
428 * AH/ESP, we cannot make anything.
429 * Defragmentation also would be mistake, RA packets
430 * cannot be fragmented, because there is no warranty
431 * that different fragments will go along one path. --ANK
432 */
433 if (opt->ra) {
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700434 u8 *ptr = skb_network_header(skb) + opt->ra;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
436 return 0;
437 }
438
439 /*
440 * check and decrement ttl
441 */
442 if (hdr->hop_limit <= 1) {
443 /* Force OUTPUT device used as source address */
444 skb->dev = dst->dev;
445 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
446 0, skb->dev);
Denis V. Lunev483a47d2008-10-08 11:09:27 -0700447 IP6_INC_STATS_BH(net,
448 ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450 kfree_skb(skb);
451 return -ETIMEDOUT;
452 }
453
YOSHIFUJI Hideakifbea49e2006-09-22 14:43:49 -0700454 /* XXX: idev->cnf.proxy_ndp? */
YOSHIFUJI Hideaki53b79972008-07-19 22:35:03 -0700455 if (net->ipv6.devconf_all->proxy_ndp &&
Daniel Lezcano8a3edd82008-03-07 11:14:16 -0800456 pneigh_lookup(&nd_tbl, net, &hdr->daddr, skb->dev, 0)) {
Ville Nuorvala74553b02006-09-22 14:42:18 -0700457 int proxied = ip6_forward_proxy_check(skb);
458 if (proxied > 0)
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700459 return ip6_input(skb);
Ville Nuorvala74553b02006-09-22 14:42:18 -0700460 else if (proxied < 0) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700461 IP6_INC_STATS(net, ip6_dst_idev(dst),
462 IPSTATS_MIB_INDISCARDS);
Ville Nuorvala74553b02006-09-22 14:42:18 -0700463 goto drop;
464 }
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700465 }
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 if (!xfrm6_route_forward(skb)) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700468 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 goto drop;
470 }
Eric Dumazetadf30902009-06-02 05:19:30 +0000471 dst = skb_dst(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 /* IPv6 specs say nothing about it, but it is clear that we cannot
474 send redirects to source routed frames.
Masahide NAKAMURA1e5dc142007-08-24 19:08:55 +0900475 We don't send redirects to frames decapsulated from IPsec.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 */
Masahide NAKAMURA1e5dc142007-08-24 19:08:55 +0900477 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0 &&
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -0700478 !skb_sec_path(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 struct in6_addr *target = NULL;
480 struct rt6_info *rt;
481 struct neighbour *n = dst->neighbour;
482
483 /*
484 * incoming and outgoing devices are the same
485 * send a redirect.
486 */
487
488 rt = (struct rt6_info *) dst;
489 if ((rt->rt6i_flags & RTF_GATEWAY))
490 target = (struct in6_addr*)&n->primary_key;
491 else
492 target = &hdr->daddr;
493
494 /* Limit redirects both by destination (here)
495 and by source (inside ndisc_send_redirect)
496 */
497 if (xrlim_allow(dst, 1*HZ))
498 ndisc_send_redirect(skb, n, target);
David L Stevens5bb1ab02007-05-09 13:53:44 -0700499 } else {
500 int addrtype = ipv6_addr_type(&hdr->saddr);
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 /* This check is security critical. */
YOSHIFUJI Hideakif81b2e72008-06-25 16:55:26 +0900503 if (addrtype == IPV6_ADDR_ANY ||
504 addrtype & (IPV6_ADDR_MULTICAST | IPV6_ADDR_LOOPBACK))
David L Stevens5bb1ab02007-05-09 13:53:44 -0700505 goto error;
506 if (addrtype & IPV6_ADDR_LINKLOCAL) {
507 icmpv6_send(skb, ICMPV6_DEST_UNREACH,
508 ICMPV6_NOT_NEIGHBOUR, 0, skb->dev);
509 goto error;
510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512
513 if (skb->len > dst_mtu(dst)) {
514 /* Again, force OUTPUT device used as source address */
515 skb->dev = dst->dev;
516 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
Denis V. Lunev483a47d2008-10-08 11:09:27 -0700517 IP6_INC_STATS_BH(net,
518 ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
519 IP6_INC_STATS_BH(net,
520 ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 kfree_skb(skb);
522 return -EMSGSIZE;
523 }
524
525 if (skb_cow(skb, dst->dev->hard_header_len)) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700526 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 goto drop;
528 }
529
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700530 hdr = ipv6_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
532 /* Mangling hops number delayed to point after skb COW */
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 hdr->hop_limit--;
535
Denis V. Lunev483a47d2008-10-08 11:09:27 -0700536 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800537 return NF_HOOK(PF_INET6, NF_INET_FORWARD, skb, skb->dev, dst->dev,
538 ip6_forward_finish);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540error:
Denis V. Lunev483a47d2008-10-08 11:09:27 -0700541 IP6_INC_STATS_BH(net, ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542drop:
543 kfree_skb(skb);
544 return -EINVAL;
545}
546
547static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
548{
549 to->pkt_type = from->pkt_type;
550 to->priority = from->priority;
551 to->protocol = from->protocol;
Eric Dumazetadf30902009-06-02 05:19:30 +0000552 skb_dst_drop(to);
553 skb_dst_set(to, dst_clone(skb_dst(from)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800555 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557#ifdef CONFIG_NET_SCHED
558 to->tc_index = from->tc_index;
559#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700560 nf_copy(to, from);
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700561#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
562 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
563 to->nf_trace = from->nf_trace;
564#endif
James Morris984bc162006-06-09 00:29:17 -0700565 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
568int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
569{
570 u16 offset = sizeof(struct ipv6hdr);
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700571 struct ipv6_opt_hdr *exthdr =
572 (struct ipv6_opt_hdr *)(ipv6_hdr(skb) + 1);
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700573 unsigned int packet_len = skb->tail - skb->network_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 int found_rhdr = 0;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700575 *nexthdr = &ipv6_hdr(skb)->nexthdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 while (offset + 1 <= packet_len) {
578
579 switch (**nexthdr) {
580
581 case NEXTHDR_HOP:
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700582 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 case NEXTHDR_ROUTING:
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700584 found_rhdr = 1;
585 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 case NEXTHDR_DEST:
Masahide NAKAMURA59fbb3a62007-06-26 23:56:32 -0700587#if defined(CONFIG_IPV6_MIP6) || defined(CONFIG_IPV6_MIP6_MODULE)
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700588 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
589 break;
590#endif
591 if (found_rhdr)
592 return offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 break;
594 default :
595 return offset;
596 }
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700597
598 offset += ipv6_optlen(exthdr);
599 *nexthdr = &exthdr->nexthdr;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700600 exthdr = (struct ipv6_opt_hdr *)(skb_network_header(skb) +
601 offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 }
603
604 return offset;
605}
606
607static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
608{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 struct sk_buff *frag;
Eric Dumazetadf30902009-06-02 05:19:30 +0000610 struct rt6_info *rt = (struct rt6_info*)skb_dst(skb);
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -0800611 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 struct ipv6hdr *tmp_hdr;
613 struct frag_hdr *fh;
614 unsigned int mtu, hlen, left, len;
Al Viroae08e1f2006-11-08 00:27:11 -0800615 __be32 frag_id = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 int ptr, offset = 0, err=0;
617 u8 *prevhdr, nexthdr = 0;
Eric Dumazetadf30902009-06-02 05:19:30 +0000618 struct net *net = dev_net(skb_dst(skb)->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 hlen = ip6_find_1stfragopt(skb, &prevhdr);
621 nexthdr = *prevhdr;
622
John Heffner628a5c52007-04-20 15:53:27 -0700623 mtu = ip6_skb_dst_mtu(skb);
John Heffnerb881ef72007-04-20 15:52:39 -0700624
625 /* We must not fragment if the socket is set to force MTU discovery
626 * or if the skb it not generated by a local socket. (This last
627 * check should be redundant, but it's free.)
628 */
Herbert Xub5c15fc2008-02-14 23:49:37 -0800629 if (!skb->local_df) {
Eric Dumazetadf30902009-06-02 05:19:30 +0000630 skb->dev = skb_dst(skb)->dev;
John Heffnerb881ef72007-04-20 15:52:39 -0700631 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
Eric Dumazetadf30902009-06-02 05:19:30 +0000632 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700633 IPSTATS_MIB_FRAGFAILS);
John Heffnerb881ef72007-04-20 15:52:39 -0700634 kfree_skb(skb);
635 return -EMSGSIZE;
636 }
637
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -0800638 if (np && np->frag_size < mtu) {
639 if (np->frag_size)
640 mtu = np->frag_size;
641 }
642 mtu -= hlen + sizeof(struct frag_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
David S. Miller4d9092b2009-06-09 00:20:05 -0700644 if (skb_has_frags(skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 int first_len = skb_pagelen(skb);
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800646 int truesizes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 if (first_len - hlen > mtu ||
649 ((first_len - hlen) & 7) ||
650 skb_cloned(skb))
651 goto slow_path;
652
David S. Miller4d9092b2009-06-09 00:20:05 -0700653 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 /* Correct geometry. */
655 if (frag->len > mtu ||
656 ((frag->len & 7) && frag->next) ||
657 skb_headroom(frag) < hlen)
658 goto slow_path;
659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 /* Partially cloned skb? */
661 if (skb_shared(frag))
662 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700663
664 BUG_ON(frag->sk);
665 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700666 frag->sk = skb->sk;
667 frag->destructor = sock_wfree;
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800668 truesizes += frag->truesize;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700669 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
671
672 err = 0;
673 offset = 0;
674 frag = skb_shinfo(skb)->frag_list;
David S. Miller4d9092b2009-06-09 00:20:05 -0700675 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 /* BUILD HEADER */
677
YOSHIFUJI Hideaki9a217a12006-12-05 13:47:21 -0800678 *prevhdr = NEXTHDR_FRAGMENT;
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700679 tmp_hdr = kmemdup(skb_network_header(skb), hlen, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 if (!tmp_hdr) {
Eric Dumazetadf30902009-06-02 05:19:30 +0000681 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700682 IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 return -ENOMEM;
684 }
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 __skb_pull(skb, hlen);
687 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700688 __skb_push(skb, hlen);
689 skb_reset_network_header(skb);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700690 memcpy(skb_network_header(skb), tmp_hdr, hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Sridhar Samudrala7ea2f2c2009-07-09 08:10:01 +0000692 ipv6_select_ident(fh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 fh->nexthdr = nexthdr;
694 fh->reserved = 0;
695 fh->frag_off = htons(IP6_MF);
696 frag_id = fh->identification;
697
698 first_len = skb_pagelen(skb);
699 skb->data_len = first_len - skb_headlen(skb);
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800700 skb->truesize -= truesizes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 skb->len = first_len;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700702 ipv6_hdr(skb)->payload_len = htons(first_len -
703 sizeof(struct ipv6hdr));
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900704
705 dst_hold(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
707 for (;;) {
708 /* Prepare header of the next frame,
709 * before previous one went down. */
710 if (frag) {
711 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300712 skb_reset_transport_header(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700714 __skb_push(frag, hlen);
715 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700716 memcpy(skb_network_header(frag), tmp_hdr,
717 hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 offset += skb->len - hlen - sizeof(struct frag_hdr);
719 fh->nexthdr = nexthdr;
720 fh->reserved = 0;
721 fh->frag_off = htons(offset);
722 if (frag->next != NULL)
723 fh->frag_off |= htons(IP6_MF);
724 fh->identification = frag_id;
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700725 ipv6_hdr(frag)->payload_len =
726 htons(frag->len -
727 sizeof(struct ipv6hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 ip6_copy_metadata(frag, skb);
729 }
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 err = output(skb);
Wei Dongdafee492006-08-02 13:41:21 -0700732 if(!err)
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700733 IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
734 IPSTATS_MIB_FRAGCREATES);
Wei Dongdafee492006-08-02 13:41:21 -0700735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (err || !frag)
737 break;
738
739 skb = frag;
740 frag = skb->next;
741 skb->next = NULL;
742 }
743
Jesper Juhla51482b2005-11-08 09:41:34 -0800744 kfree(tmp_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 if (err == 0) {
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700747 IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
748 IPSTATS_MIB_FRAGOKS);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900749 dst_release(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return 0;
751 }
752
753 while (frag) {
754 skb = frag->next;
755 kfree_skb(frag);
756 frag = skb;
757 }
758
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700759 IP6_INC_STATS(net, ip6_dst_idev(&rt->u.dst),
760 IPSTATS_MIB_FRAGFAILS);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900761 dst_release(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return err;
763 }
764
765slow_path:
766 left = skb->len - hlen; /* Space per frame */
767 ptr = hlen; /* Where to start from */
768
769 /*
770 * Fragment the datagram.
771 */
772
773 *prevhdr = NEXTHDR_FRAGMENT;
774
775 /*
776 * Keep copying data until we run out.
777 */
778 while(left > 0) {
779 len = left;
780 /* IF: it doesn't fit, use 'mtu' - the data space left */
781 if (len > mtu)
782 len = mtu;
783 /* IF: we are not sending upto and including the packet end
784 then align the next start on an eight byte boundary */
785 if (len < left) {
786 len &= ~7;
787 }
788 /*
789 * Allocate buffer.
790 */
791
Johannes Bergf5184d22008-05-12 20:48:31 -0700792 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_ALLOCATED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700793 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
Eric Dumazetadf30902009-06-02 05:19:30 +0000794 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900795 IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 err = -ENOMEM;
797 goto fail;
798 }
799
800 /*
801 * Set up data on packet
802 */
803
804 ip6_copy_metadata(frag, skb);
805 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
806 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700807 skb_reset_network_header(frag);
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300808 fh = (struct frag_hdr *)(skb_network_header(frag) + hlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700809 frag->transport_header = (frag->network_header + hlen +
810 sizeof(struct frag_hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812 /*
813 * Charge the memory for the fragment to any owner
814 * it might possess
815 */
816 if (skb->sk)
817 skb_set_owner_w(frag, skb->sk);
818
819 /*
820 * Copy the packet header into the new buffer.
821 */
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300822 skb_copy_from_linear_data(skb, skb_network_header(frag), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
824 /*
825 * Build fragment header.
826 */
827 fh->nexthdr = nexthdr;
828 fh->reserved = 0;
Yan Zhengf36d6ab2005-10-03 14:19:15 -0700829 if (!frag_id) {
Sridhar Samudrala7ea2f2c2009-07-09 08:10:01 +0000830 ipv6_select_ident(fh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 frag_id = fh->identification;
832 } else
833 fh->identification = frag_id;
834
835 /*
836 * Copy a block of the IP datagram.
837 */
Wei Yongjun8984e412007-08-21 20:59:08 -0700838 if (skb_copy_bits(skb, ptr, skb_transport_header(frag), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 BUG();
840 left -= len;
841
842 fh->frag_off = htons(offset);
843 if (left > 0)
844 fh->frag_off |= htons(IP6_MF);
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -0700845 ipv6_hdr(frag)->payload_len = htons(frag->len -
846 sizeof(struct ipv6hdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 ptr += len;
849 offset += len;
850
851 /*
852 * Put this fragment into the sending queue.
853 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 err = output(frag);
855 if (err)
856 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700857
Eric Dumazetadf30902009-06-02 05:19:30 +0000858 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
Denis V. Lunev3bd653c2008-10-08 10:54:51 -0700859 IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 }
Eric Dumazetadf30902009-06-02 05:19:30 +0000861 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900862 IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 return err;
865
866fail:
Eric Dumazetadf30902009-06-02 05:19:30 +0000867 IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900868 IPSTATS_MIB_FRAGFAILS);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900869 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return err;
871}
872
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700873static inline int ip6_rt_check(struct rt6key *rt_key,
874 struct in6_addr *fl_addr,
875 struct in6_addr *addr_cache)
876{
877 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
878 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
879}
880
Herbert Xu497c6152006-07-30 20:19:33 -0700881static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
882 struct dst_entry *dst,
883 struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
Herbert Xu497c6152006-07-30 20:19:33 -0700885 struct ipv6_pinfo *np = inet6_sk(sk);
886 struct rt6_info *rt = (struct rt6_info *)dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Herbert Xu497c6152006-07-30 20:19:33 -0700888 if (!dst)
889 goto out;
890
891 /* Yes, checking route validity in not connected
892 * case is not very simple. Take into account,
893 * that we do not support routing by source, TOS,
894 * and MSG_DONTROUTE --ANK (980726)
895 *
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700896 * 1. ip6_rt_check(): If route was host route,
897 * check that cached destination is current.
Herbert Xu497c6152006-07-30 20:19:33 -0700898 * If it is network route, we still may
899 * check its validity using saved pointer
900 * to the last used address: daddr_cache.
901 * We do not want to save whole address now,
902 * (because main consumer of this service
903 * is tcp, which has not this problem),
904 * so that the last trick works only on connected
905 * sockets.
906 * 2. oif also should be the same.
907 */
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700908 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
YOSHIFUJI Hideaki8e1ef0a2006-08-29 17:15:09 -0700909#ifdef CONFIG_IPV6_SUBTREES
910 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
911#endif
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700912 (fl->oif && fl->oif != dst->dev->ifindex)) {
Herbert Xu497c6152006-07-30 20:19:33 -0700913 dst_release(dst);
914 dst = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 }
916
Herbert Xu497c6152006-07-30 20:19:33 -0700917out:
918 return dst;
919}
920
921static int ip6_dst_lookup_tail(struct sock *sk,
922 struct dst_entry **dst, struct flowi *fl)
923{
924 int err;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900925 struct net *net = sock_net(sk);
Herbert Xu497c6152006-07-30 20:19:33 -0700926
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (*dst == NULL)
Daniel Lezcano8a3edd82008-03-07 11:14:16 -0800928 *dst = ip6_route_output(net, sk, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 if ((err = (*dst)->error))
931 goto out_err_release;
932
933 if (ipv6_addr_any(&fl->fl6_src)) {
Brian Haley191cd582008-08-14 15:33:21 -0700934 err = ipv6_dev_get_saddr(net, ip6_dst_idev(*dst)->dev,
YOSHIFUJI Hideaki7cbca672008-03-25 09:37:42 +0900935 &fl->fl6_dst,
936 sk ? inet6_sk(sk)->srcprefs : 0,
937 &fl->fl6_src);
Olaf Hering44456d32005-07-27 11:45:17 -0700938 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 goto out_err_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 }
941
Neil Horman95c385b2007-04-25 17:08:10 -0700942#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
Neil Hormane550dfb2008-09-09 13:51:35 -0700943 /*
944 * Here if the dst entry we've looked up
945 * has a neighbour entry that is in the INCOMPLETE
946 * state and the src address from the flow is
947 * marked as OPTIMISTIC, we release the found
948 * dst entry and replace it instead with the
949 * dst entry of the nexthop router
950 */
951 if ((*dst)->neighbour && !((*dst)->neighbour->nud_state & NUD_VALID)) {
952 struct inet6_ifaddr *ifp;
953 struct flowi fl_gw;
954 int redirect;
Neil Horman95c385b2007-04-25 17:08:10 -0700955
Neil Hormane550dfb2008-09-09 13:51:35 -0700956 ifp = ipv6_get_ifaddr(net, &fl->fl6_src,
957 (*dst)->dev, 1);
Neil Horman95c385b2007-04-25 17:08:10 -0700958
Neil Hormane550dfb2008-09-09 13:51:35 -0700959 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
960 if (ifp)
961 in6_ifa_put(ifp);
Neil Horman95c385b2007-04-25 17:08:10 -0700962
Neil Hormane550dfb2008-09-09 13:51:35 -0700963 if (redirect) {
964 /*
965 * We need to get the dst entry for the
966 * default router instead
967 */
968 dst_release(*dst);
969 memcpy(&fl_gw, fl, sizeof(struct flowi));
970 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
971 *dst = ip6_route_output(net, sk, &fl_gw);
972 if ((err = (*dst)->error))
973 goto out_err_release;
Neil Horman95c385b2007-04-25 17:08:10 -0700974 }
Neil Hormane550dfb2008-09-09 13:51:35 -0700975 }
Neil Horman95c385b2007-04-25 17:08:10 -0700976#endif
977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return 0;
979
980out_err_release:
Mitsuru Chinenca46f9c2007-12-05 22:31:47 -0800981 if (err == -ENETUNREACH)
Denis V. Lunev483a47d2008-10-08 11:09:27 -0700982 IP6_INC_STATS_BH(net, NULL, IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 dst_release(*dst);
984 *dst = NULL;
985 return err;
986}
Adrian Bunk34a0b3c2005-11-29 16:28:56 -0800987
Herbert Xu497c6152006-07-30 20:19:33 -0700988/**
989 * ip6_dst_lookup - perform route lookup on flow
990 * @sk: socket which provides route info
991 * @dst: pointer to dst_entry * for result
992 * @fl: flow to lookup
993 *
994 * This function performs a route lookup on the given flow.
995 *
996 * It returns zero on success, or a standard errno code on error.
997 */
998int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
999{
1000 *dst = NULL;
1001 return ip6_dst_lookup_tail(sk, dst, fl);
1002}
Arnaldo Carvalho de Melo3cf3dc62005-12-13 23:23:20 -08001003EXPORT_SYMBOL_GPL(ip6_dst_lookup);
1004
Herbert Xu497c6152006-07-30 20:19:33 -07001005/**
1006 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
1007 * @sk: socket which provides the dst cache and route info
1008 * @dst: pointer to dst_entry * for result
1009 * @fl: flow to lookup
1010 *
1011 * This function performs a route lookup on the given flow with the
1012 * possibility of using the cached route in the socket if it is valid.
1013 * It will take the socket dst lock when operating on the dst cache.
1014 * As a result, this function can only be used in process context.
1015 *
1016 * It returns zero on success, or a standard errno code on error.
1017 */
1018int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
1019{
1020 *dst = NULL;
1021 if (sk) {
1022 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
1023 *dst = ip6_sk_dst_check(sk, *dst, fl);
1024 }
1025
1026 return ip6_dst_lookup_tail(sk, dst, fl);
1027}
1028EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
1029
Adrian Bunk34a0b3c2005-11-29 16:28:56 -08001030static inline int ip6_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001031 int getfrag(void *from, char *to, int offset, int len,
1032 int odd, struct sk_buff *skb),
1033 void *from, int length, int hh_len, int fragheaderlen,
1034 int transhdrlen, int mtu,unsigned int flags)
1035
1036{
1037 struct sk_buff *skb;
1038 int err;
1039
1040 /* There is support for UDP large send offload by network
1041 * device, so create one single skb packet containing complete
1042 * udp datagram
1043 */
1044 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
1045 skb = sock_alloc_send_skb(sk,
1046 hh_len + fragheaderlen + transhdrlen + 20,
1047 (flags & MSG_DONTWAIT), &err);
1048 if (skb == NULL)
1049 return -ENOMEM;
1050
1051 /* reserve space for Hardware header */
1052 skb_reserve(skb, hh_len);
1053
1054 /* create space for UDP/IP header */
1055 skb_put(skb,fragheaderlen + transhdrlen);
1056
1057 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -07001058 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001059
1060 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001061 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001062
Patrick McHardy84fa7932006-08-29 16:44:56 -07001063 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001064 skb->csum = 0;
1065 sk->sk_sndmsg_off = 0;
1066 }
1067
1068 err = skb_append_datato_frags(sk,skb, getfrag, from,
1069 (length - transhdrlen));
1070 if (!err) {
1071 struct frag_hdr fhdr;
1072
Sridhar Samudralac31d5322009-07-09 08:09:58 +00001073 /* Specify the length of each IPv6 datagram fragment.
1074 * It has to be a multiple of 8.
1075 */
1076 skb_shinfo(skb)->gso_size = (mtu - fragheaderlen -
1077 sizeof(struct frag_hdr)) & ~7;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001078 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Sridhar Samudrala7ea2f2c2009-07-09 08:10:01 +00001079 ipv6_select_ident(&fhdr);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001080 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1081 __skb_queue_tail(&sk->sk_write_queue, skb);
1082
1083 return 0;
1084 }
1085 /* There is not enough support do UPD LSO,
1086 * so follow normal path
1087 */
1088 kfree_skb(skb);
1089
1090 return err;
1091}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Herbert Xu0178b692009-02-05 15:15:50 -08001093static inline struct ipv6_opt_hdr *ip6_opt_dup(struct ipv6_opt_hdr *src,
1094 gfp_t gfp)
1095{
1096 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1097}
1098
1099static inline struct ipv6_rt_hdr *ip6_rthdr_dup(struct ipv6_rt_hdr *src,
1100 gfp_t gfp)
1101{
1102 return src ? kmemdup(src, (src->hdrlen + 1) * 8, gfp) : NULL;
1103}
1104
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001105int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1106 int offset, int len, int odd, struct sk_buff *skb),
1107 void *from, int length, int transhdrlen,
1108 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1109 struct rt6_info *rt, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110{
1111 struct inet_sock *inet = inet_sk(sk);
1112 struct ipv6_pinfo *np = inet6_sk(sk);
1113 struct sk_buff *skb;
1114 unsigned int maxfraglen, fragheaderlen;
1115 int exthdrlen;
1116 int hh_len;
1117 int mtu;
1118 int copy;
1119 int err;
1120 int offset = 0;
1121 int csummode = CHECKSUM_NONE;
1122
1123 if (flags&MSG_PROBE)
1124 return 0;
1125 if (skb_queue_empty(&sk->sk_write_queue)) {
1126 /*
1127 * setup for corking
1128 */
1129 if (opt) {
Herbert Xu0178b692009-02-05 15:15:50 -08001130 if (WARN_ON(np->cork.opt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return -EINVAL;
Herbert Xu0178b692009-02-05 15:15:50 -08001132
1133 np->cork.opt = kmalloc(opt->tot_len, sk->sk_allocation);
1134 if (unlikely(np->cork.opt == NULL))
1135 return -ENOBUFS;
1136
1137 np->cork.opt->tot_len = opt->tot_len;
1138 np->cork.opt->opt_flen = opt->opt_flen;
1139 np->cork.opt->opt_nflen = opt->opt_nflen;
1140
1141 np->cork.opt->dst0opt = ip6_opt_dup(opt->dst0opt,
1142 sk->sk_allocation);
1143 if (opt->dst0opt && !np->cork.opt->dst0opt)
1144 return -ENOBUFS;
1145
1146 np->cork.opt->dst1opt = ip6_opt_dup(opt->dst1opt,
1147 sk->sk_allocation);
1148 if (opt->dst1opt && !np->cork.opt->dst1opt)
1149 return -ENOBUFS;
1150
1151 np->cork.opt->hopopt = ip6_opt_dup(opt->hopopt,
1152 sk->sk_allocation);
1153 if (opt->hopopt && !np->cork.opt->hopopt)
1154 return -ENOBUFS;
1155
1156 np->cork.opt->srcrt = ip6_rthdr_dup(opt->srcrt,
1157 sk->sk_allocation);
1158 if (opt->srcrt && !np->cork.opt->srcrt)
1159 return -ENOBUFS;
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 /* need source address above miyazawa*/
1162 }
1163 dst_hold(&rt->u.dst);
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001164 inet->cork.dst = &rt->u.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 inet->cork.fl = *fl;
1166 np->cork.hop_limit = hlimit;
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001167 np->cork.tclass = tclass;
John Heffner628a5c52007-04-20 15:53:27 -07001168 mtu = np->pmtudisc == IPV6_PMTUDISC_PROBE ?
1169 rt->u.dst.dev->mtu : dst_mtu(rt->u.dst.path);
Dave Jonesc7503602006-03-20 22:44:52 -08001170 if (np->frag_size < mtu) {
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -08001171 if (np->frag_size)
1172 mtu = np->frag_size;
1173 }
1174 inet->cork.fragsize = mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (dst_allfrag(rt->u.dst.path))
1176 inet->cork.flags |= IPCORK_ALLFRAG;
1177 inet->cork.length = 0;
1178 sk->sk_sndmsg_page = NULL;
1179 sk->sk_sndmsg_off = 0;
Herbert Xu014889422007-11-13 21:32:26 -08001180 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0) -
Masahide NAKAMURAa1b05142007-12-20 20:41:12 -08001181 rt->rt6i_nfheader_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 length += exthdrlen;
1183 transhdrlen += exthdrlen;
1184 } else {
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001185 rt = (struct rt6_info *)inet->cork.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 fl = &inet->cork.fl;
Herbert Xu0178b692009-02-05 15:15:50 -08001187 opt = np->cork.opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 transhdrlen = 0;
1189 exthdrlen = 0;
1190 mtu = inet->cork.fragsize;
1191 }
1192
1193 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1194
Masahide NAKAMURAa1b05142007-12-20 20:41:12 -08001195 fragheaderlen = sizeof(struct ipv6hdr) + rt->rt6i_nfheader_len +
Herbert Xub4ce9272007-11-13 21:33:32 -08001196 (opt ? opt->opt_nflen : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1198
1199 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1200 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1201 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1202 return -EMSGSIZE;
1203 }
1204 }
1205
1206 /*
1207 * Let's try using as much space as possible.
1208 * Use MTU if total length of the message fits into the MTU.
1209 * Otherwise, we need to reserve fragment header and
1210 * fragment alignment (= 8-15 octects, in total).
1211 *
1212 * Note that we may need to "move" the data from the tail of
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001213 * of the buffer to the new fragment when we split
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 * the message.
1215 *
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001216 * FIXME: It may be fragmented into multiple chunks
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 * at once if non-fragmentable extension headers
1218 * are too large.
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001219 * --yoshfuji
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 */
1221
1222 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001223 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1224 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1225
Patrick McHardybaa829d2006-03-12 20:35:12 -08001226 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1227 fragheaderlen, transhdrlen, mtu,
1228 flags);
1229 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001230 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001231 return 0;
1232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1235 goto alloc_new_skb;
1236
1237 while (length > 0) {
1238 /* Check if the remaining data fits into current packet. */
1239 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1240 if (copy < length)
1241 copy = maxfraglen - skb->len;
1242
1243 if (copy <= 0) {
1244 char *data;
1245 unsigned int datalen;
1246 unsigned int fraglen;
1247 unsigned int fraggap;
1248 unsigned int alloclen;
1249 struct sk_buff *skb_prev;
1250alloc_new_skb:
1251 skb_prev = skb;
1252
1253 /* There's no room in the current skb */
1254 if (skb_prev)
1255 fraggap = skb_prev->len - maxfraglen;
1256 else
1257 fraggap = 0;
1258
1259 /*
1260 * If remaining data exceeds the mtu,
1261 * we know we need more fragment(s).
1262 */
1263 datalen = length + fraggap;
1264 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1265 datalen = maxfraglen - fragheaderlen;
1266
1267 fraglen = datalen + fragheaderlen;
1268 if ((flags & MSG_MORE) &&
1269 !(rt->u.dst.dev->features&NETIF_F_SG))
1270 alloclen = mtu;
1271 else
1272 alloclen = datalen + fragheaderlen;
1273
1274 /*
1275 * The last fragment gets additional space at tail.
1276 * Note: we overallocate on fragments with MSG_MODE
1277 * because we have no idea if we're the last one.
1278 */
1279 if (datalen == length + fraggap)
1280 alloclen += rt->u.dst.trailer_len;
1281
1282 /*
1283 * We just reserve space for fragment header.
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001284 * Note: this may be overallocation if the message
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 * (without MSG_MORE) fits into the MTU.
1286 */
1287 alloclen += sizeof(struct frag_hdr);
1288
1289 if (transhdrlen) {
1290 skb = sock_alloc_send_skb(sk,
1291 alloclen + hh_len,
1292 (flags & MSG_DONTWAIT), &err);
1293 } else {
1294 skb = NULL;
1295 if (atomic_read(&sk->sk_wmem_alloc) <=
1296 2 * sk->sk_sndbuf)
1297 skb = sock_wmalloc(sk,
1298 alloclen + hh_len, 1,
1299 sk->sk_allocation);
1300 if (unlikely(skb == NULL))
1301 err = -ENOBUFS;
1302 }
1303 if (skb == NULL)
1304 goto error;
1305 /*
1306 * Fill in the control structures
1307 */
1308 skb->ip_summed = csummode;
1309 skb->csum = 0;
1310 /* reserve for fragmentation */
1311 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1312
1313 /*
1314 * Find where to start putting bytes
1315 */
1316 data = skb_put(skb, fraglen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -03001317 skb_set_network_header(skb, exthdrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 data += fragheaderlen;
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001319 skb->transport_header = (skb->network_header +
1320 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (fraggap) {
1322 skb->csum = skb_copy_and_csum_bits(
1323 skb_prev, maxfraglen,
1324 data + transhdrlen, fraggap, 0);
1325 skb_prev->csum = csum_sub(skb_prev->csum,
1326 skb->csum);
1327 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -07001328 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
1330 copy = datalen - transhdrlen - fraggap;
1331 if (copy < 0) {
1332 err = -EINVAL;
1333 kfree_skb(skb);
1334 goto error;
1335 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1336 err = -EFAULT;
1337 kfree_skb(skb);
1338 goto error;
1339 }
1340
1341 offset += copy;
1342 length -= datalen - fraggap;
1343 transhdrlen = 0;
1344 exthdrlen = 0;
1345 csummode = CHECKSUM_NONE;
1346
1347 /*
1348 * Put the packet on the pending queue
1349 */
1350 __skb_queue_tail(&sk->sk_write_queue, skb);
1351 continue;
1352 }
1353
1354 if (copy > length)
1355 copy = length;
1356
1357 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1358 unsigned int off;
1359
1360 off = skb->len;
1361 if (getfrag(from, skb_put(skb, copy),
1362 offset, copy, off, skb) < 0) {
1363 __skb_trim(skb, off);
1364 err = -EFAULT;
1365 goto error;
1366 }
1367 } else {
1368 int i = skb_shinfo(skb)->nr_frags;
1369 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1370 struct page *page = sk->sk_sndmsg_page;
1371 int off = sk->sk_sndmsg_off;
1372 unsigned int left;
1373
1374 if (page && (left = PAGE_SIZE - off) > 0) {
1375 if (copy >= left)
1376 copy = left;
1377 if (page != frag->page) {
1378 if (i == MAX_SKB_FRAGS) {
1379 err = -EMSGSIZE;
1380 goto error;
1381 }
1382 get_page(page);
1383 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1384 frag = &skb_shinfo(skb)->frags[i];
1385 }
1386 } else if(i < MAX_SKB_FRAGS) {
1387 if (copy > PAGE_SIZE)
1388 copy = PAGE_SIZE;
1389 page = alloc_pages(sk->sk_allocation, 0);
1390 if (page == NULL) {
1391 err = -ENOMEM;
1392 goto error;
1393 }
1394 sk->sk_sndmsg_page = page;
1395 sk->sk_sndmsg_off = 0;
1396
1397 skb_fill_page_desc(skb, i, page, 0, 0);
1398 frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 } else {
1400 err = -EMSGSIZE;
1401 goto error;
1402 }
1403 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1404 err = -EFAULT;
1405 goto error;
1406 }
1407 sk->sk_sndmsg_off += copy;
1408 frag->size += copy;
1409 skb->len += copy;
1410 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001411 skb->truesize += copy;
1412 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 }
1414 offset += copy;
1415 length -= copy;
1416 }
1417 return 0;
1418error:
1419 inet->cork.length -= length;
Denis V. Lunev3bd653c2008-10-08 10:54:51 -07001420 IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return err;
1422}
1423
Pavel Emelyanovbf138862007-11-05 21:04:31 -08001424static void ip6_cork_release(struct inet_sock *inet, struct ipv6_pinfo *np)
1425{
Herbert Xu0178b692009-02-05 15:15:50 -08001426 if (np->cork.opt) {
1427 kfree(np->cork.opt->dst0opt);
1428 kfree(np->cork.opt->dst1opt);
1429 kfree(np->cork.opt->hopopt);
1430 kfree(np->cork.opt->srcrt);
1431 kfree(np->cork.opt);
1432 np->cork.opt = NULL;
1433 }
1434
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001435 if (inet->cork.dst) {
1436 dst_release(inet->cork.dst);
1437 inet->cork.dst = NULL;
Pavel Emelyanovbf138862007-11-05 21:04:31 -08001438 inet->cork.flags &= ~IPCORK_ALLFRAG;
1439 }
1440 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1441}
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443int ip6_push_pending_frames(struct sock *sk)
1444{
1445 struct sk_buff *skb, *tmp_skb;
1446 struct sk_buff **tail_skb;
1447 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1448 struct inet_sock *inet = inet_sk(sk);
1449 struct ipv6_pinfo *np = inet6_sk(sk);
Denis V. Lunev3bd653c2008-10-08 10:54:51 -07001450 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 struct ipv6hdr *hdr;
1452 struct ipv6_txoptions *opt = np->cork.opt;
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001453 struct rt6_info *rt = (struct rt6_info *)inet->cork.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 struct flowi *fl = &inet->cork.fl;
1455 unsigned char proto = fl->proto;
1456 int err = 0;
1457
1458 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1459 goto out;
1460 tail_skb = &(skb_shinfo(skb)->frag_list);
1461
1462 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001463 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001464 __skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001466 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 *tail_skb = tmp_skb;
1468 tail_skb = &(tmp_skb->next);
1469 skb->len += tmp_skb->len;
1470 skb->data_len += tmp_skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 skb->truesize += tmp_skb->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 tmp_skb->destructor = NULL;
1473 tmp_skb->sk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
1475
Herbert Xu28a894532008-02-12 18:07:27 -08001476 /* Allow local fragmentation. */
Herbert Xub5c15fc2008-02-14 23:49:37 -08001477 if (np->pmtudisc < IPV6_PMTUDISC_DO)
Herbert Xu28a894532008-02-12 18:07:27 -08001478 skb->local_df = 1;
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 ipv6_addr_copy(final_dst, &fl->fl6_dst);
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001481 __skb_pull(skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (opt && opt->opt_flen)
1483 ipv6_push_frag_opts(skb, opt, &proto);
1484 if (opt && opt->opt_nflen)
1485 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1486
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -07001487 skb_push(skb, sizeof(struct ipv6hdr));
1488 skb_reset_network_header(skb);
Arnaldo Carvalho de Melo0660e032007-04-25 17:54:47 -07001489 hdr = ipv6_hdr(skb);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001490
Al Viro90bcaf72006-11-08 00:25:17 -08001491 *(__be32*)hdr = fl->fl6_flowlabel |
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001492 htonl(0x60000000 | ((int)np->cork.tclass << 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 hdr->hop_limit = np->cork.hop_limit;
1495 hdr->nexthdr = proto;
1496 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1497 ipv6_addr_copy(&hdr->daddr, final_dst);
1498
Patrick McHardya2c20642006-01-08 22:37:26 -08001499 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001500 skb->mark = sk->sk_mark;
Patrick McHardya2c20642006-01-08 22:37:26 -08001501
Eric Dumazetadf30902009-06-02 05:19:30 +00001502 skb_dst_set(skb, dst_clone(&rt->u.dst));
Neil Hormanedf391f2009-04-27 02:45:02 -07001503 IP6_UPD_PO_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUT, skb->len);
David L Stevens14878f72007-09-16 16:52:35 -07001504 if (proto == IPPROTO_ICMPV6) {
Eric Dumazetadf30902009-06-02 05:19:30 +00001505 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
David L Stevens14878f72007-09-16 16:52:35 -07001506
Denis V. Lunev5a57d4c2008-10-08 10:34:14 -07001507 ICMP6MSGOUT_INC_STATS_BH(net, idev, icmp6_hdr(skb)->icmp6_type);
Denis V. Luneve41b5362008-10-08 10:33:26 -07001508 ICMP6_INC_STATS_BH(net, idev, ICMP6_MIB_OUTMSGS);
David L Stevens14878f72007-09-16 16:52:35 -07001509 }
1510
Herbert Xuef76bc22008-01-11 19:15:08 -08001511 err = ip6_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 if (err) {
1513 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001514 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 if (err)
1516 goto error;
1517 }
1518
1519out:
Pavel Emelyanovbf138862007-11-05 21:04:31 -08001520 ip6_cork_release(inet, np);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return err;
1522error:
Eric Dumazet06254912009-09-01 18:37:16 -07001523 IP6_INC_STATS(net, rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 goto out;
1525}
1526
1527void ip6_flush_pending_frames(struct sock *sk)
1528{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 struct sk_buff *skb;
1530
1531 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
Eric Dumazetadf30902009-06-02 05:19:30 +00001532 if (skb_dst(skb))
1533 IP6_INC_STATS(sock_net(sk), ip6_dst_idev(skb_dst(skb)),
YOSHIFUJI Hideakie1f522082007-09-11 11:31:43 +02001534 IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 kfree_skb(skb);
1536 }
1537
Pavel Emelyanovbf138862007-11-05 21:04:31 -08001538 ip6_cork_release(inet_sk(sk), inet6_sk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539}