blob: a5f4562b5d29c19697bc4407914d7797cc61162d [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 *
8 * $Id: ip6_output.c,v 1.34 2002/02/01 22:01:04 davem Exp $
9 *
10 * Based on linux/net/ipv4/ip_output.c
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 *
17 * Changes:
18 * A.N.Kuznetsov : airthmetics in fragmentation.
19 * extension headers are implemented.
20 * route changes now work.
21 * ip6_forward does not confuse sniffers.
22 * etc.
23 *
24 * H. von Brand : Added missing #include <linux/string.h>
25 * Imran Patel : frag id should be in NBO
26 * Kazunori MIYAZAWA @USAGI
27 * : add ip6_append_data and related functions
28 * for datagram xmit
29 */
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/errno.h>
32#include <linux/types.h>
33#include <linux/string.h>
34#include <linux/socket.h>
35#include <linux/net.h>
36#include <linux/netdevice.h>
37#include <linux/if_arp.h>
38#include <linux/in6.h>
39#include <linux/tcp.h>
40#include <linux/route.h>
Herbert Xub59f45d2006-05-27 23:05:54 -070041#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#include <linux/netfilter.h>
44#include <linux/netfilter_ipv6.h>
45
46#include <net/sock.h>
47#include <net/snmp.h>
48
49#include <net/ipv6.h>
50#include <net/ndisc.h>
51#include <net/protocol.h>
52#include <net/ip6_route.h>
53#include <net/addrconf.h>
54#include <net/rawv6.h>
55#include <net/icmp.h>
56#include <net/xfrm.h>
57#include <net/checksum.h>
58
59static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *));
60
61static __inline__ void ipv6_select_ident(struct sk_buff *skb, struct frag_hdr *fhdr)
62{
63 static u32 ipv6_fragmentation_id = 1;
64 static DEFINE_SPINLOCK(ip6_id_lock);
65
66 spin_lock_bh(&ip6_id_lock);
67 fhdr->identification = htonl(ipv6_fragmentation_id);
68 if (++ipv6_fragmentation_id == 0)
69 ipv6_fragmentation_id = 1;
70 spin_unlock_bh(&ip6_id_lock);
71}
72
73static inline int ip6_output_finish(struct sk_buff *skb)
74{
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 struct dst_entry *dst = skb->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Stephen Hemminger3644f0c2006-12-07 15:08:17 -080077 if (dst->hh)
78 return neigh_hh_output(dst->hh, skb);
79 else if (dst->neighbour)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 return dst->neighbour->output(skb);
81
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +090082 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 kfree_skb(skb);
84 return -EINVAL;
85
86}
87
88/* dev_loopback_xmit for use with netfilter. */
89static int ip6_dev_loopback_xmit(struct sk_buff *newskb)
90{
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -070091 skb_reset_mac_header(newskb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 __skb_pull(newskb, newskb->nh.raw - newskb->data);
93 newskb->pkt_type = PACKET_LOOPBACK;
94 newskb->ip_summed = CHECKSUM_UNNECESSARY;
95 BUG_TRAP(newskb->dst);
96
97 netif_rx(newskb);
98 return 0;
99}
100
101
102static int ip6_output2(struct sk_buff *skb)
103{
104 struct dst_entry *dst = skb->dst;
105 struct net_device *dev = dst->dev;
106
107 skb->protocol = htons(ETH_P_IPV6);
108 skb->dev = dev;
109
110 if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr)) {
111 struct ipv6_pinfo* np = skb->sk ? inet6_sk(skb->sk) : NULL;
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900112 struct inet6_dev *idev = ip6_dst_idev(skb->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 if (!(dev->flags & IFF_LOOPBACK) && (!np || np->mc_loop) &&
115 ipv6_chk_mcast_addr(dev, &skb->nh.ipv6h->daddr,
116 &skb->nh.ipv6h->saddr)) {
117 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
118
119 /* Do not check for IFF_ALLMULTI; multicast routing
120 is not supported in any case.
121 */
122 if (newskb)
123 NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, newskb, NULL,
124 newskb->dev,
125 ip6_dev_loopback_xmit);
126
127 if (skb->nh.ipv6h->hop_limit == 0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900128 IP6_INC_STATS(idev, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 kfree_skb(skb);
130 return 0;
131 }
132 }
133
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900134 IP6_INC_STATS(idev, IPSTATS_MIB_OUTMCASTPKTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
136
137 return NF_HOOK(PF_INET6, NF_IP6_POST_ROUTING, skb,NULL, skb->dev,ip6_output_finish);
138}
139
140int ip6_output(struct sk_buff *skb)
141{
Herbert Xu89114af2006-07-08 13:34:32 -0700142 if ((skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) ||
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700143 dst_allfrag(skb->dst))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 return ip6_fragment(skb, ip6_output2);
145 else
146 return ip6_output2(skb);
147}
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149/*
150 * xmit an sk_buff (used by TCP)
151 */
152
153int ip6_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
154 struct ipv6_txoptions *opt, int ipfragok)
155{
Patrick McHardyb30bd282006-03-23 01:17:25 -0800156 struct ipv6_pinfo *np = inet6_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 struct in6_addr *first_hop = &fl->fl6_dst;
158 struct dst_entry *dst = skb->dst;
159 struct ipv6hdr *hdr;
160 u8 proto = fl->proto;
161 int seg_len = skb->len;
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +0900162 int hlimit, tclass;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 u32 mtu;
164
165 if (opt) {
166 int head_room;
167
168 /* First: exthdrs may take lots of space (~8K for now)
169 MAX_HEADER is not enough.
170 */
171 head_room = opt->opt_nflen + opt->opt_flen;
172 seg_len += head_room;
173 head_room += sizeof(struct ipv6hdr) + LL_RESERVED_SPACE(dst->dev);
174
175 if (skb_headroom(skb) < head_room) {
176 struct sk_buff *skb2 = skb_realloc_headroom(skb, head_room);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900177 if (skb2 == NULL) {
178 IP6_INC_STATS(ip6_dst_idev(skb->dst),
179 IPSTATS_MIB_OUTDISCARDS);
180 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 return -ENOBUFS;
182 }
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900183 kfree_skb(skb);
184 skb = skb2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 if (sk)
186 skb_set_owner_w(skb, sk);
187 }
188 if (opt->opt_flen)
189 ipv6_push_frag_opts(skb, opt, &proto);
190 if (opt->opt_nflen)
191 ipv6_push_nfrag_opts(skb, opt, &proto, &first_hop);
192 }
193
194 hdr = skb->nh.ipv6h = (struct ipv6hdr*)skb_push(skb, sizeof(struct ipv6hdr));
195
196 /*
197 * Fill in the IPv6 header
198 */
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 hlimit = -1;
201 if (np)
202 hlimit = np->hop_limit;
203 if (hlimit < 0)
204 hlimit = dst_metric(dst, RTAX_HOPLIMIT);
205 if (hlimit < 0)
206 hlimit = ipv6_get_hoplimit(dst->dev);
207
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +0900208 tclass = -1;
209 if (np)
210 tclass = np->tclass;
211 if (tclass < 0)
212 tclass = 0;
213
Al Viro90bcaf72006-11-08 00:25:17 -0800214 *(__be32 *)hdr = htonl(0x60000000 | (tclass << 20)) | fl->fl6_flowlabel;
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +0900215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 hdr->payload_len = htons(seg_len);
217 hdr->nexthdr = proto;
218 hdr->hop_limit = hlimit;
219
220 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
221 ipv6_addr_copy(&hdr->daddr, first_hop);
222
Patrick McHardya2c20642006-01-08 22:37:26 -0800223 skb->priority = sk->sk_priority;
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 mtu = dst_mtu(dst);
Herbert Xu89114af2006-07-08 13:34:32 -0700226 if ((skb->len <= mtu) || ipfragok || skb_is_gso(skb)) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900227 IP6_INC_STATS(ip6_dst_idev(skb->dst),
228 IPSTATS_MIB_OUTREQUESTS);
Harald Welte6869c4d2005-08-09 19:24:19 -0700229 return NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, dst->dev,
230 dst_output);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 }
232
233 if (net_ratelimit())
234 printk(KERN_DEBUG "IPv6: sending pkt_too_big to self\n");
235 skb->dev = dst->dev;
236 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, skb->dev);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900237 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 kfree_skb(skb);
239 return -EMSGSIZE;
240}
241
YOSHIFUJI Hideaki71590392007-02-22 22:05:40 +0900242EXPORT_SYMBOL(ip6_xmit);
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244/*
245 * To avoid extra problems ND packets are send through this
246 * routine. It's code duplication but I really want to avoid
247 * extra checks since ipv6_build_header is used by TCP (which
248 * is for us performance critical)
249 */
250
251int ip6_nd_hdr(struct sock *sk, struct sk_buff *skb, struct net_device *dev,
252 struct in6_addr *saddr, struct in6_addr *daddr,
253 int proto, int len)
254{
255 struct ipv6_pinfo *np = inet6_sk(sk);
256 struct ipv6hdr *hdr;
257 int totlen;
258
259 skb->protocol = htons(ETH_P_IPV6);
260 skb->dev = dev;
261
262 totlen = len + sizeof(struct ipv6hdr);
263
264 hdr = (struct ipv6hdr *) skb_put(skb, sizeof(struct ipv6hdr));
265 skb->nh.ipv6h = hdr;
266
Al Viroae08e1f2006-11-08 00:27:11 -0800267 *(__be32*)hdr = htonl(0x60000000);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
269 hdr->payload_len = htons(len);
270 hdr->nexthdr = proto;
271 hdr->hop_limit = np->hop_limit;
272
273 ipv6_addr_copy(&hdr->saddr, saddr);
274 ipv6_addr_copy(&hdr->daddr, daddr);
275
276 return 0;
277}
278
279static int ip6_call_ra_chain(struct sk_buff *skb, int sel)
280{
281 struct ip6_ra_chain *ra;
282 struct sock *last = NULL;
283
284 read_lock(&ip6_ra_lock);
285 for (ra = ip6_ra_chain; ra; ra = ra->next) {
286 struct sock *sk = ra->sk;
Andrew McDonald0bd1b592005-08-09 19:44:42 -0700287 if (sk && ra->sel == sel &&
288 (!sk->sk_bound_dev_if ||
289 sk->sk_bound_dev_if == skb->dev->ifindex)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 if (last) {
291 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
292 if (skb2)
293 rawv6_rcv(last, skb2);
294 }
295 last = sk;
296 }
297 }
298
299 if (last) {
300 rawv6_rcv(last, skb);
301 read_unlock(&ip6_ra_lock);
302 return 1;
303 }
304 read_unlock(&ip6_ra_lock);
305 return 0;
306}
307
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700308static int ip6_forward_proxy_check(struct sk_buff *skb)
309{
310 struct ipv6hdr *hdr = skb->nh.ipv6h;
311 u8 nexthdr = hdr->nexthdr;
312 int offset;
313
314 if (ipv6_ext_hdr(nexthdr)) {
315 offset = ipv6_skip_exthdr(skb, sizeof(*hdr), &nexthdr);
316 if (offset < 0)
317 return 0;
318 } else
319 offset = sizeof(struct ipv6hdr);
320
321 if (nexthdr == IPPROTO_ICMPV6) {
322 struct icmp6hdr *icmp6;
323
324 if (!pskb_may_pull(skb, skb->nh.raw + offset + 1 - skb->data))
325 return 0;
326
327 icmp6 = (struct icmp6hdr *)(skb->nh.raw + offset);
328
329 switch (icmp6->icmp6_type) {
330 case NDISC_ROUTER_SOLICITATION:
331 case NDISC_ROUTER_ADVERTISEMENT:
332 case NDISC_NEIGHBOUR_SOLICITATION:
333 case NDISC_NEIGHBOUR_ADVERTISEMENT:
334 case NDISC_REDIRECT:
335 /* For reaction involving unicast neighbor discovery
336 * message destined to the proxied address, pass it to
337 * input function.
338 */
339 return 1;
340 default:
341 break;
342 }
343 }
344
Ville Nuorvala74553b02006-09-22 14:42:18 -0700345 /*
346 * The proxying router can't forward traffic sent to a link-local
347 * address, so signal the sender and discard the packet. This
348 * behavior is clarified by the MIPv6 specification.
349 */
350 if (ipv6_addr_type(&hdr->daddr) & IPV6_ADDR_LINKLOCAL) {
351 dst_link_failure(skb);
352 return -1;
353 }
354
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700355 return 0;
356}
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358static inline int ip6_forward_finish(struct sk_buff *skb)
359{
360 return dst_output(skb);
361}
362
363int ip6_forward(struct sk_buff *skb)
364{
365 struct dst_entry *dst = skb->dst;
366 struct ipv6hdr *hdr = skb->nh.ipv6h;
367 struct inet6_skb_parm *opt = IP6CB(skb);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 if (ipv6_devconf.forwarding == 0)
370 goto error;
371
372 if (!xfrm6_policy_check(NULL, XFRM_POLICY_FWD, skb)) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900373 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 goto drop;
375 }
376
377 skb->ip_summed = CHECKSUM_NONE;
378
379 /*
380 * We DO NOT make any processing on
381 * RA packets, pushing them to user level AS IS
382 * without ane WARRANTY that application will be able
383 * to interpret them. The reason is that we
384 * cannot make anything clever here.
385 *
386 * We are not end-node, so that if packet contains
387 * AH/ESP, we cannot make anything.
388 * Defragmentation also would be mistake, RA packets
389 * cannot be fragmented, because there is no warranty
390 * that different fragments will go along one path. --ANK
391 */
392 if (opt->ra) {
393 u8 *ptr = skb->nh.raw + opt->ra;
394 if (ip6_call_ra_chain(skb, (ptr[2]<<8) + ptr[3]))
395 return 0;
396 }
397
398 /*
399 * check and decrement ttl
400 */
401 if (hdr->hop_limit <= 1) {
402 /* Force OUTPUT device used as source address */
403 skb->dev = dst->dev;
404 icmpv6_send(skb, ICMPV6_TIME_EXCEED, ICMPV6_EXC_HOPLIMIT,
405 0, skb->dev);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900406 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INHDRERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
408 kfree_skb(skb);
409 return -ETIMEDOUT;
410 }
411
YOSHIFUJI Hideakifbea49e2006-09-22 14:43:49 -0700412 /* XXX: idev->cnf.proxy_ndp? */
413 if (ipv6_devconf.proxy_ndp &&
414 pneigh_lookup(&nd_tbl, &hdr->daddr, skb->dev, 0)) {
Ville Nuorvala74553b02006-09-22 14:42:18 -0700415 int proxied = ip6_forward_proxy_check(skb);
416 if (proxied > 0)
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700417 return ip6_input(skb);
Ville Nuorvala74553b02006-09-22 14:42:18 -0700418 else if (proxied < 0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900419 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
Ville Nuorvala74553b02006-09-22 14:42:18 -0700420 goto drop;
421 }
Ville Nuorvalae21e0b52006-09-22 14:41:44 -0700422 }
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 if (!xfrm6_route_forward(skb)) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900425 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_INDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 goto drop;
427 }
428 dst = skb->dst;
429
430 /* IPv6 specs say nothing about it, but it is clear that we cannot
431 send redirects to source routed frames.
432 */
433 if (skb->dev == dst->dev && dst->neighbour && opt->srcrt == 0) {
434 struct in6_addr *target = NULL;
435 struct rt6_info *rt;
436 struct neighbour *n = dst->neighbour;
437
438 /*
439 * incoming and outgoing devices are the same
440 * send a redirect.
441 */
442
443 rt = (struct rt6_info *) dst;
444 if ((rt->rt6i_flags & RTF_GATEWAY))
445 target = (struct in6_addr*)&n->primary_key;
446 else
447 target = &hdr->daddr;
448
449 /* Limit redirects both by destination (here)
450 and by source (inside ndisc_send_redirect)
451 */
452 if (xrlim_allow(dst, 1*HZ))
453 ndisc_send_redirect(skb, n, target);
454 } else if (ipv6_addr_type(&hdr->saddr)&(IPV6_ADDR_MULTICAST|IPV6_ADDR_LOOPBACK
455 |IPV6_ADDR_LINKLOCAL)) {
456 /* This check is security critical. */
457 goto error;
458 }
459
460 if (skb->len > dst_mtu(dst)) {
461 /* Again, force OUTPUT device used as source address */
462 skb->dev = dst->dev;
463 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, dst_mtu(dst), skb->dev);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900464 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INTOOBIGERRORS);
465 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 kfree_skb(skb);
467 return -EMSGSIZE;
468 }
469
470 if (skb_cow(skb, dst->dev->hard_header_len)) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900471 IP6_INC_STATS(ip6_dst_idev(dst), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 goto drop;
473 }
474
475 hdr = skb->nh.ipv6h;
476
477 /* Mangling hops number delayed to point after skb COW */
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 hdr->hop_limit--;
480
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900481 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_OUTFORWDATAGRAMS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return NF_HOOK(PF_INET6,NF_IP6_FORWARD, skb, skb->dev, dst->dev, ip6_forward_finish);
483
484error:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900485 IP6_INC_STATS_BH(ip6_dst_idev(dst), IPSTATS_MIB_INADDRERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486drop:
487 kfree_skb(skb);
488 return -EINVAL;
489}
490
491static void ip6_copy_metadata(struct sk_buff *to, struct sk_buff *from)
492{
493 to->pkt_type = from->pkt_type;
494 to->priority = from->priority;
495 to->protocol = from->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 dst_release(to->dst);
497 to->dst = dst_clone(from->dst);
498 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800499 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
501#ifdef CONFIG_NET_SCHED
502 to->tc_index = from->tc_index;
503#endif
504#ifdef CONFIG_NETFILTER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 /* Connection association is same as pre-frag packet */
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800506 nf_conntrack_put(to->nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 to->nfct = from->nfct;
508 nf_conntrack_get(to->nfct);
509 to->nfctinfo = from->nfctinfo;
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800510#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
511 nf_conntrack_put_reasm(to->nfct_reasm);
512 to->nfct_reasm = from->nfct_reasm;
513 nf_conntrack_get_reasm(to->nfct_reasm);
514#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515#ifdef CONFIG_BRIDGE_NETFILTER
516 nf_bridge_put(to->nf_bridge);
517 to->nf_bridge = from->nf_bridge;
518 nf_bridge_get(to->nf_bridge);
519#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520#endif
James Morris984bc162006-06-09 00:29:17 -0700521 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524int ip6_find_1stfragopt(struct sk_buff *skb, u8 **nexthdr)
525{
526 u16 offset = sizeof(struct ipv6hdr);
527 struct ipv6_opt_hdr *exthdr = (struct ipv6_opt_hdr*)(skb->nh.ipv6h + 1);
528 unsigned int packet_len = skb->tail - skb->nh.raw;
529 int found_rhdr = 0;
530 *nexthdr = &skb->nh.ipv6h->nexthdr;
531
532 while (offset + 1 <= packet_len) {
533
534 switch (**nexthdr) {
535
536 case NEXTHDR_HOP:
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700537 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 case NEXTHDR_ROUTING:
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700539 found_rhdr = 1;
540 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 case NEXTHDR_DEST:
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700542#ifdef CONFIG_IPV6_MIP6
543 if (ipv6_find_tlv(skb, offset, IPV6_TLV_HAO) >= 0)
544 break;
545#endif
546 if (found_rhdr)
547 return offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 break;
549 default :
550 return offset;
551 }
Masahide NAKAMURA27637df2006-08-23 19:29:47 -0700552
553 offset += ipv6_optlen(exthdr);
554 *nexthdr = &exthdr->nexthdr;
555 exthdr = (struct ipv6_opt_hdr*)(skb->nh.raw + offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
557
558 return offset;
559}
Herbert Xub59f45d2006-05-27 23:05:54 -0700560EXPORT_SYMBOL_GPL(ip6_find_1stfragopt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562static int ip6_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
563{
564 struct net_device *dev;
565 struct sk_buff *frag;
566 struct rt6_info *rt = (struct rt6_info*)skb->dst;
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -0800567 struct ipv6_pinfo *np = skb->sk ? inet6_sk(skb->sk) : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 struct ipv6hdr *tmp_hdr;
569 struct frag_hdr *fh;
570 unsigned int mtu, hlen, left, len;
Al Viroae08e1f2006-11-08 00:27:11 -0800571 __be32 frag_id = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 int ptr, offset = 0, err=0;
573 u8 *prevhdr, nexthdr = 0;
574
575 dev = rt->u.dst.dev;
576 hlen = ip6_find_1stfragopt(skb, &prevhdr);
577 nexthdr = *prevhdr;
578
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -0800579 mtu = dst_mtu(&rt->u.dst);
580 if (np && np->frag_size < mtu) {
581 if (np->frag_size)
582 mtu = np->frag_size;
583 }
584 mtu -= hlen + sizeof(struct frag_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 if (skb_shinfo(skb)->frag_list) {
587 int first_len = skb_pagelen(skb);
588
589 if (first_len - hlen > mtu ||
590 ((first_len - hlen) & 7) ||
591 skb_cloned(skb))
592 goto slow_path;
593
594 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
595 /* Correct geometry. */
596 if (frag->len > mtu ||
597 ((frag->len & 7) && frag->next) ||
598 skb_headroom(frag) < hlen)
599 goto slow_path;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /* Partially cloned skb? */
602 if (skb_shared(frag))
603 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700604
605 BUG_ON(frag->sk);
606 if (skb->sk) {
607 sock_hold(skb->sk);
608 frag->sk = skb->sk;
609 frag->destructor = sock_wfree;
610 skb->truesize -= frag->truesize;
611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 }
613
614 err = 0;
615 offset = 0;
616 frag = skb_shinfo(skb)->frag_list;
617 skb_shinfo(skb)->frag_list = NULL;
618 /* BUILD HEADER */
619
YOSHIFUJI Hideaki9a217a12006-12-05 13:47:21 -0800620 *prevhdr = NEXTHDR_FRAGMENT;
Arnaldo Carvalho de Meloaf879cc2006-11-17 12:14:37 -0200621 tmp_hdr = kmemdup(skb->nh.raw, hlen, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (!tmp_hdr) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900623 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return -ENOMEM;
625 }
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 __skb_pull(skb, hlen);
628 fh = (struct frag_hdr*)__skb_push(skb, sizeof(struct frag_hdr));
629 skb->nh.raw = __skb_push(skb, hlen);
630 memcpy(skb->nh.raw, tmp_hdr, hlen);
631
632 ipv6_select_ident(skb, fh);
633 fh->nexthdr = nexthdr;
634 fh->reserved = 0;
635 fh->frag_off = htons(IP6_MF);
636 frag_id = fh->identification;
637
638 first_len = skb_pagelen(skb);
639 skb->data_len = first_len - skb_headlen(skb);
640 skb->len = first_len;
641 skb->nh.ipv6h->payload_len = htons(first_len - sizeof(struct ipv6hdr));
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900642
643 dst_hold(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
645 for (;;) {
646 /* Prepare header of the next frame,
647 * before previous one went down. */
648 if (frag) {
649 frag->ip_summed = CHECKSUM_NONE;
650 frag->h.raw = frag->data;
651 fh = (struct frag_hdr*)__skb_push(frag, sizeof(struct frag_hdr));
652 frag->nh.raw = __skb_push(frag, hlen);
653 memcpy(frag->nh.raw, tmp_hdr, hlen);
654 offset += skb->len - hlen - sizeof(struct frag_hdr);
655 fh->nexthdr = nexthdr;
656 fh->reserved = 0;
657 fh->frag_off = htons(offset);
658 if (frag->next != NULL)
659 fh->frag_off |= htons(IP6_MF);
660 fh->identification = frag_id;
661 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
662 ip6_copy_metadata(frag, skb);
663 }
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 err = output(skb);
Wei Dongdafee492006-08-02 13:41:21 -0700666 if(!err)
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900667 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGCREATES);
Wei Dongdafee492006-08-02 13:41:21 -0700668
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (err || !frag)
670 break;
671
672 skb = frag;
673 frag = skb->next;
674 skb->next = NULL;
675 }
676
Jesper Juhla51482b2005-11-08 09:41:34 -0800677 kfree(tmp_hdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679 if (err == 0) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900680 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGOKS);
681 dst_release(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 return 0;
683 }
684
685 while (frag) {
686 skb = frag->next;
687 kfree_skb(frag);
688 frag = skb;
689 }
690
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900691 IP6_INC_STATS(ip6_dst_idev(&rt->u.dst), IPSTATS_MIB_FRAGFAILS);
692 dst_release(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return err;
694 }
695
696slow_path:
697 left = skb->len - hlen; /* Space per frame */
698 ptr = hlen; /* Where to start from */
699
700 /*
701 * Fragment the datagram.
702 */
703
704 *prevhdr = NEXTHDR_FRAGMENT;
705
706 /*
707 * Keep copying data until we run out.
708 */
709 while(left > 0) {
710 len = left;
711 /* IF: it doesn't fit, use 'mtu' - the data space left */
712 if (len > mtu)
713 len = mtu;
714 /* IF: we are not sending upto and including the packet end
715 then align the next start on an eight byte boundary */
716 if (len < left) {
717 len &= ~7;
718 }
719 /*
720 * Allocate buffer.
721 */
722
723 if ((frag = alloc_skb(len+hlen+sizeof(struct frag_hdr)+LL_RESERVED_SPACE(rt->u.dst.dev), GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700724 NETDEBUG(KERN_INFO "IPv6: frag: no memory for new fragment!\n");
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900725 IP6_INC_STATS(ip6_dst_idev(skb->dst),
726 IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 err = -ENOMEM;
728 goto fail;
729 }
730
731 /*
732 * Set up data on packet
733 */
734
735 ip6_copy_metadata(frag, skb);
736 skb_reserve(frag, LL_RESERVED_SPACE(rt->u.dst.dev));
737 skb_put(frag, len + hlen + sizeof(struct frag_hdr));
738 frag->nh.raw = frag->data;
739 fh = (struct frag_hdr*)(frag->data + hlen);
740 frag->h.raw = frag->data + hlen + sizeof(struct frag_hdr);
741
742 /*
743 * Charge the memory for the fragment to any owner
744 * it might possess
745 */
746 if (skb->sk)
747 skb_set_owner_w(frag, skb->sk);
748
749 /*
750 * Copy the packet header into the new buffer.
751 */
752 memcpy(frag->nh.raw, skb->data, hlen);
753
754 /*
755 * Build fragment header.
756 */
757 fh->nexthdr = nexthdr;
758 fh->reserved = 0;
Yan Zhengf36d6ab2005-10-03 14:19:15 -0700759 if (!frag_id) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 ipv6_select_ident(skb, fh);
761 frag_id = fh->identification;
762 } else
763 fh->identification = frag_id;
764
765 /*
766 * Copy a block of the IP datagram.
767 */
768 if (skb_copy_bits(skb, ptr, frag->h.raw, len))
769 BUG();
770 left -= len;
771
772 fh->frag_off = htons(offset);
773 if (left > 0)
774 fh->frag_off |= htons(IP6_MF);
775 frag->nh.ipv6h->payload_len = htons(frag->len - sizeof(struct ipv6hdr));
776
777 ptr += len;
778 offset += len;
779
780 /*
781 * Put this fragment into the sending queue.
782 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 err = output(frag);
784 if (err)
785 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700786
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900787 IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900789 IP6_INC_STATS(ip6_dst_idev(skb->dst),
790 IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 return err;
793
794fail:
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +0900795 IP6_INC_STATS(ip6_dst_idev(skb->dst),
796 IPSTATS_MIB_FRAGFAILS);
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900797 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 return err;
799}
800
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700801static inline int ip6_rt_check(struct rt6key *rt_key,
802 struct in6_addr *fl_addr,
803 struct in6_addr *addr_cache)
804{
805 return ((rt_key->plen != 128 || !ipv6_addr_equal(fl_addr, &rt_key->addr)) &&
806 (addr_cache == NULL || !ipv6_addr_equal(fl_addr, addr_cache)));
807}
808
Herbert Xu497c6152006-07-30 20:19:33 -0700809static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
810 struct dst_entry *dst,
811 struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Herbert Xu497c6152006-07-30 20:19:33 -0700813 struct ipv6_pinfo *np = inet6_sk(sk);
814 struct rt6_info *rt = (struct rt6_info *)dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Herbert Xu497c6152006-07-30 20:19:33 -0700816 if (!dst)
817 goto out;
818
819 /* Yes, checking route validity in not connected
820 * case is not very simple. Take into account,
821 * that we do not support routing by source, TOS,
822 * and MSG_DONTROUTE --ANK (980726)
823 *
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700824 * 1. ip6_rt_check(): If route was host route,
825 * check that cached destination is current.
Herbert Xu497c6152006-07-30 20:19:33 -0700826 * If it is network route, we still may
827 * check its validity using saved pointer
828 * to the last used address: daddr_cache.
829 * We do not want to save whole address now,
830 * (because main consumer of this service
831 * is tcp, which has not this problem),
832 * so that the last trick works only on connected
833 * sockets.
834 * 2. oif also should be the same.
835 */
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700836 if (ip6_rt_check(&rt->rt6i_dst, &fl->fl6_dst, np->daddr_cache) ||
YOSHIFUJI Hideaki8e1ef0a2006-08-29 17:15:09 -0700837#ifdef CONFIG_IPV6_SUBTREES
838 ip6_rt_check(&rt->rt6i_src, &fl->fl6_src, np->saddr_cache) ||
839#endif
YOSHIFUJI Hideakicf6b1982006-08-23 17:19:18 -0700840 (fl->oif && fl->oif != dst->dev->ifindex)) {
Herbert Xu497c6152006-07-30 20:19:33 -0700841 dst_release(dst);
842 dst = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 }
844
Herbert Xu497c6152006-07-30 20:19:33 -0700845out:
846 return dst;
847}
848
849static int ip6_dst_lookup_tail(struct sock *sk,
850 struct dst_entry **dst, struct flowi *fl)
851{
852 int err;
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (*dst == NULL)
855 *dst = ip6_route_output(sk, fl);
856
857 if ((err = (*dst)->error))
858 goto out_err_release;
859
860 if (ipv6_addr_any(&fl->fl6_src)) {
861 err = ipv6_get_saddr(*dst, &fl->fl6_dst, &fl->fl6_src);
Olaf Hering44456d32005-07-27 11:45:17 -0700862 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 goto out_err_release;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 }
865
Neil Horman95c385b2007-04-25 17:08:10 -0700866#ifdef CONFIG_IPV6_OPTIMISTIC_DAD
867 /*
868 * Here if the dst entry we've looked up
869 * has a neighbour entry that is in the INCOMPLETE
870 * state and the src address from the flow is
871 * marked as OPTIMISTIC, we release the found
872 * dst entry and replace it instead with the
873 * dst entry of the nexthop router
874 */
875 if (!((*dst)->neighbour->nud_state & NUD_VALID)) {
876 struct inet6_ifaddr *ifp;
877 struct flowi fl_gw;
878 int redirect;
879
880 ifp = ipv6_get_ifaddr(&fl->fl6_src, (*dst)->dev, 1);
881
882 redirect = (ifp && ifp->flags & IFA_F_OPTIMISTIC);
883 if (ifp)
884 in6_ifa_put(ifp);
885
886 if (redirect) {
887 /*
888 * We need to get the dst entry for the
889 * default router instead
890 */
891 dst_release(*dst);
892 memcpy(&fl_gw, fl, sizeof(struct flowi));
893 memset(&fl_gw.fl6_dst, 0, sizeof(struct in6_addr));
894 *dst = ip6_route_output(sk, &fl_gw);
895 if ((err = (*dst)->error))
896 goto out_err_release;
897 }
898 }
899#endif
900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return 0;
902
903out_err_release:
904 dst_release(*dst);
905 *dst = NULL;
906 return err;
907}
Adrian Bunk34a0b3c2005-11-29 16:28:56 -0800908
Herbert Xu497c6152006-07-30 20:19:33 -0700909/**
910 * ip6_dst_lookup - perform route lookup on flow
911 * @sk: socket which provides route info
912 * @dst: pointer to dst_entry * for result
913 * @fl: flow to lookup
914 *
915 * This function performs a route lookup on the given flow.
916 *
917 * It returns zero on success, or a standard errno code on error.
918 */
919int ip6_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
920{
921 *dst = NULL;
922 return ip6_dst_lookup_tail(sk, dst, fl);
923}
Arnaldo Carvalho de Melo3cf3dc62005-12-13 23:23:20 -0800924EXPORT_SYMBOL_GPL(ip6_dst_lookup);
925
Herbert Xu497c6152006-07-30 20:19:33 -0700926/**
927 * ip6_sk_dst_lookup - perform socket cached route lookup on flow
928 * @sk: socket which provides the dst cache and route info
929 * @dst: pointer to dst_entry * for result
930 * @fl: flow to lookup
931 *
932 * This function performs a route lookup on the given flow with the
933 * possibility of using the cached route in the socket if it is valid.
934 * It will take the socket dst lock when operating on the dst cache.
935 * As a result, this function can only be used in process context.
936 *
937 * It returns zero on success, or a standard errno code on error.
938 */
939int ip6_sk_dst_lookup(struct sock *sk, struct dst_entry **dst, struct flowi *fl)
940{
941 *dst = NULL;
942 if (sk) {
943 *dst = sk_dst_check(sk, inet6_sk(sk)->dst_cookie);
944 *dst = ip6_sk_dst_check(sk, *dst, fl);
945 }
946
947 return ip6_dst_lookup_tail(sk, dst, fl);
948}
949EXPORT_SYMBOL_GPL(ip6_sk_dst_lookup);
950
Adrian Bunk34a0b3c2005-11-29 16:28:56 -0800951static inline int ip6_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700952 int getfrag(void *from, char *to, int offset, int len,
953 int odd, struct sk_buff *skb),
954 void *from, int length, int hh_len, int fragheaderlen,
955 int transhdrlen, int mtu,unsigned int flags)
956
957{
958 struct sk_buff *skb;
959 int err;
960
961 /* There is support for UDP large send offload by network
962 * device, so create one single skb packet containing complete
963 * udp datagram
964 */
965 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
966 skb = sock_alloc_send_skb(sk,
967 hh_len + fragheaderlen + transhdrlen + 20,
968 (flags & MSG_DONTWAIT), &err);
969 if (skb == NULL)
970 return -ENOMEM;
971
972 /* reserve space for Hardware header */
973 skb_reserve(skb, hh_len);
974
975 /* create space for UDP/IP header */
976 skb_put(skb,fragheaderlen + transhdrlen);
977
978 /* initialize network header pointer */
979 skb->nh.raw = skb->data;
980
981 /* initialize protocol header pointer */
982 skb->h.raw = skb->data + fragheaderlen;
983
Patrick McHardy84fa7932006-08-29 16:44:56 -0700984 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700985 skb->csum = 0;
986 sk->sk_sndmsg_off = 0;
987 }
988
989 err = skb_append_datato_frags(sk,skb, getfrag, from,
990 (length - transhdrlen));
991 if (!err) {
992 struct frag_hdr fhdr;
993
994 /* specify the length of each IP datagram fragment*/
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +0900995 skb_shinfo(skb)->gso_size = mtu - fragheaderlen -
Herbert Xu79671682006-06-22 02:40:14 -0700996 sizeof(struct frag_hdr);
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700997 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700998 ipv6_select_ident(skb, &fhdr);
999 skb_shinfo(skb)->ip6_frag_id = fhdr.identification;
1000 __skb_queue_tail(&sk->sk_write_queue, skb);
1001
1002 return 0;
1003 }
1004 /* There is not enough support do UPD LSO,
1005 * so follow normal path
1006 */
1007 kfree_skb(skb);
1008
1009 return err;
1010}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001012int ip6_append_data(struct sock *sk, int getfrag(void *from, char *to,
1013 int offset, int len, int odd, struct sk_buff *skb),
1014 void *from, int length, int transhdrlen,
1015 int hlimit, int tclass, struct ipv6_txoptions *opt, struct flowi *fl,
1016 struct rt6_info *rt, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 struct inet_sock *inet = inet_sk(sk);
1019 struct ipv6_pinfo *np = inet6_sk(sk);
1020 struct sk_buff *skb;
1021 unsigned int maxfraglen, fragheaderlen;
1022 int exthdrlen;
1023 int hh_len;
1024 int mtu;
1025 int copy;
1026 int err;
1027 int offset = 0;
1028 int csummode = CHECKSUM_NONE;
1029
1030 if (flags&MSG_PROBE)
1031 return 0;
1032 if (skb_queue_empty(&sk->sk_write_queue)) {
1033 /*
1034 * setup for corking
1035 */
1036 if (opt) {
1037 if (np->cork.opt == NULL) {
1038 np->cork.opt = kmalloc(opt->tot_len,
1039 sk->sk_allocation);
1040 if (unlikely(np->cork.opt == NULL))
1041 return -ENOBUFS;
1042 } else if (np->cork.opt->tot_len < opt->tot_len) {
1043 printk(KERN_DEBUG "ip6_append_data: invalid option length\n");
1044 return -EINVAL;
1045 }
1046 memcpy(np->cork.opt, opt, opt->tot_len);
1047 inet->cork.flags |= IPCORK_OPT;
1048 /* need source address above miyazawa*/
1049 }
1050 dst_hold(&rt->u.dst);
1051 np->cork.rt = rt;
1052 inet->cork.fl = *fl;
1053 np->cork.hop_limit = hlimit;
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001054 np->cork.tclass = tclass;
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -08001055 mtu = dst_mtu(rt->u.dst.path);
Dave Jonesc7503602006-03-20 22:44:52 -08001056 if (np->frag_size < mtu) {
YOSHIFUJI Hideakid91675f2006-02-24 13:18:33 -08001057 if (np->frag_size)
1058 mtu = np->frag_size;
1059 }
1060 inet->cork.fragsize = mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (dst_allfrag(rt->u.dst.path))
1062 inet->cork.flags |= IPCORK_ALLFRAG;
1063 inet->cork.length = 0;
1064 sk->sk_sndmsg_page = NULL;
1065 sk->sk_sndmsg_off = 0;
1066 exthdrlen = rt->u.dst.header_len + (opt ? opt->opt_flen : 0);
1067 length += exthdrlen;
1068 transhdrlen += exthdrlen;
1069 } else {
1070 rt = np->cork.rt;
1071 fl = &inet->cork.fl;
1072 if (inet->cork.flags & IPCORK_OPT)
1073 opt = np->cork.opt;
1074 transhdrlen = 0;
1075 exthdrlen = 0;
1076 mtu = inet->cork.fragsize;
1077 }
1078
1079 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1080
Masahide NAKAMURA1b5c2292006-08-23 18:11:50 -07001081 fragheaderlen = sizeof(struct ipv6hdr) + rt->u.dst.nfheader_len + (opt ? opt->opt_nflen : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen - sizeof(struct frag_hdr);
1083
1084 if (mtu <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN) {
1085 if (inet->cork.length + length > sizeof(struct ipv6hdr) + IPV6_MAXPLEN - fragheaderlen) {
1086 ipv6_local_error(sk, EMSGSIZE, fl, mtu-exthdrlen);
1087 return -EMSGSIZE;
1088 }
1089 }
1090
1091 /*
1092 * Let's try using as much space as possible.
1093 * Use MTU if total length of the message fits into the MTU.
1094 * Otherwise, we need to reserve fragment header and
1095 * fragment alignment (= 8-15 octects, in total).
1096 *
1097 * Note that we may need to "move" the data from the tail of
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001098 * of the buffer to the new fragment when we split
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 * the message.
1100 *
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001101 * FIXME: It may be fragmented into multiple chunks
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 * at once if non-fragmentable extension headers
1103 * are too large.
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001104 * --yoshfuji
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 */
1106
1107 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001108 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1109 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1110
Patrick McHardybaa829d2006-03-12 20:35:12 -08001111 err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1112 fragheaderlen, transhdrlen, mtu,
1113 flags);
1114 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001115 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001116 return 0;
1117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1120 goto alloc_new_skb;
1121
1122 while (length > 0) {
1123 /* Check if the remaining data fits into current packet. */
1124 copy = (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - skb->len;
1125 if (copy < length)
1126 copy = maxfraglen - skb->len;
1127
1128 if (copy <= 0) {
1129 char *data;
1130 unsigned int datalen;
1131 unsigned int fraglen;
1132 unsigned int fraggap;
1133 unsigned int alloclen;
1134 struct sk_buff *skb_prev;
1135alloc_new_skb:
1136 skb_prev = skb;
1137
1138 /* There's no room in the current skb */
1139 if (skb_prev)
1140 fraggap = skb_prev->len - maxfraglen;
1141 else
1142 fraggap = 0;
1143
1144 /*
1145 * If remaining data exceeds the mtu,
1146 * we know we need more fragment(s).
1147 */
1148 datalen = length + fraggap;
1149 if (datalen > (inet->cork.length <= mtu && !(inet->cork.flags & IPCORK_ALLFRAG) ? mtu : maxfraglen) - fragheaderlen)
1150 datalen = maxfraglen - fragheaderlen;
1151
1152 fraglen = datalen + fragheaderlen;
1153 if ((flags & MSG_MORE) &&
1154 !(rt->u.dst.dev->features&NETIF_F_SG))
1155 alloclen = mtu;
1156 else
1157 alloclen = datalen + fragheaderlen;
1158
1159 /*
1160 * The last fragment gets additional space at tail.
1161 * Note: we overallocate on fragments with MSG_MODE
1162 * because we have no idea if we're the last one.
1163 */
1164 if (datalen == length + fraggap)
1165 alloclen += rt->u.dst.trailer_len;
1166
1167 /*
1168 * We just reserve space for fragment header.
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001169 * Note: this may be overallocation if the message
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 * (without MSG_MORE) fits into the MTU.
1171 */
1172 alloclen += sizeof(struct frag_hdr);
1173
1174 if (transhdrlen) {
1175 skb = sock_alloc_send_skb(sk,
1176 alloclen + hh_len,
1177 (flags & MSG_DONTWAIT), &err);
1178 } else {
1179 skb = NULL;
1180 if (atomic_read(&sk->sk_wmem_alloc) <=
1181 2 * sk->sk_sndbuf)
1182 skb = sock_wmalloc(sk,
1183 alloclen + hh_len, 1,
1184 sk->sk_allocation);
1185 if (unlikely(skb == NULL))
1186 err = -ENOBUFS;
1187 }
1188 if (skb == NULL)
1189 goto error;
1190 /*
1191 * Fill in the control structures
1192 */
1193 skb->ip_summed = csummode;
1194 skb->csum = 0;
1195 /* reserve for fragmentation */
1196 skb_reserve(skb, hh_len+sizeof(struct frag_hdr));
1197
1198 /*
1199 * Find where to start putting bytes
1200 */
1201 data = skb_put(skb, fraglen);
1202 skb->nh.raw = data + exthdrlen;
1203 data += fragheaderlen;
1204 skb->h.raw = data + exthdrlen;
1205
1206 if (fraggap) {
1207 skb->csum = skb_copy_and_csum_bits(
1208 skb_prev, maxfraglen,
1209 data + transhdrlen, fraggap, 0);
1210 skb_prev->csum = csum_sub(skb_prev->csum,
1211 skb->csum);
1212 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -07001213 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215 copy = datalen - transhdrlen - fraggap;
1216 if (copy < 0) {
1217 err = -EINVAL;
1218 kfree_skb(skb);
1219 goto error;
1220 } else if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1221 err = -EFAULT;
1222 kfree_skb(skb);
1223 goto error;
1224 }
1225
1226 offset += copy;
1227 length -= datalen - fraggap;
1228 transhdrlen = 0;
1229 exthdrlen = 0;
1230 csummode = CHECKSUM_NONE;
1231
1232 /*
1233 * Put the packet on the pending queue
1234 */
1235 __skb_queue_tail(&sk->sk_write_queue, skb);
1236 continue;
1237 }
1238
1239 if (copy > length)
1240 copy = length;
1241
1242 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1243 unsigned int off;
1244
1245 off = skb->len;
1246 if (getfrag(from, skb_put(skb, copy),
1247 offset, copy, off, skb) < 0) {
1248 __skb_trim(skb, off);
1249 err = -EFAULT;
1250 goto error;
1251 }
1252 } else {
1253 int i = skb_shinfo(skb)->nr_frags;
1254 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1255 struct page *page = sk->sk_sndmsg_page;
1256 int off = sk->sk_sndmsg_off;
1257 unsigned int left;
1258
1259 if (page && (left = PAGE_SIZE - off) > 0) {
1260 if (copy >= left)
1261 copy = left;
1262 if (page != frag->page) {
1263 if (i == MAX_SKB_FRAGS) {
1264 err = -EMSGSIZE;
1265 goto error;
1266 }
1267 get_page(page);
1268 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1269 frag = &skb_shinfo(skb)->frags[i];
1270 }
1271 } else if(i < MAX_SKB_FRAGS) {
1272 if (copy > PAGE_SIZE)
1273 copy = PAGE_SIZE;
1274 page = alloc_pages(sk->sk_allocation, 0);
1275 if (page == NULL) {
1276 err = -ENOMEM;
1277 goto error;
1278 }
1279 sk->sk_sndmsg_page = page;
1280 sk->sk_sndmsg_off = 0;
1281
1282 skb_fill_page_desc(skb, i, page, 0, 0);
1283 frag = &skb_shinfo(skb)->frags[i];
1284 skb->truesize += PAGE_SIZE;
1285 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1286 } else {
1287 err = -EMSGSIZE;
1288 goto error;
1289 }
1290 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1291 err = -EFAULT;
1292 goto error;
1293 }
1294 sk->sk_sndmsg_off += copy;
1295 frag->size += copy;
1296 skb->len += copy;
1297 skb->data_len += copy;
1298 }
1299 offset += copy;
1300 length -= copy;
1301 }
1302 return 0;
1303error:
1304 inet->cork.length -= length;
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +09001305 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 return err;
1307}
1308
1309int ip6_push_pending_frames(struct sock *sk)
1310{
1311 struct sk_buff *skb, *tmp_skb;
1312 struct sk_buff **tail_skb;
1313 struct in6_addr final_dst_buf, *final_dst = &final_dst_buf;
1314 struct inet_sock *inet = inet_sk(sk);
1315 struct ipv6_pinfo *np = inet6_sk(sk);
1316 struct ipv6hdr *hdr;
1317 struct ipv6_txoptions *opt = np->cork.opt;
1318 struct rt6_info *rt = np->cork.rt;
1319 struct flowi *fl = &inet->cork.fl;
1320 unsigned char proto = fl->proto;
1321 int err = 0;
1322
1323 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1324 goto out;
1325 tail_skb = &(skb_shinfo(skb)->frag_list);
1326
1327 /* move skb->data to ip header from ext header */
1328 if (skb->data < skb->nh.raw)
1329 __skb_pull(skb, skb->nh.raw - skb->data);
1330 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1331 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1332 *tail_skb = tmp_skb;
1333 tail_skb = &(tmp_skb->next);
1334 skb->len += tmp_skb->len;
1335 skb->data_len += tmp_skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 skb->truesize += tmp_skb->truesize;
1337 __sock_put(tmp_skb->sk);
1338 tmp_skb->destructor = NULL;
1339 tmp_skb->sk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
1341
1342 ipv6_addr_copy(final_dst, &fl->fl6_dst);
1343 __skb_pull(skb, skb->h.raw - skb->nh.raw);
1344 if (opt && opt->opt_flen)
1345 ipv6_push_frag_opts(skb, opt, &proto);
1346 if (opt && opt->opt_nflen)
1347 ipv6_push_nfrag_opts(skb, opt, &proto, &final_dst);
1348
1349 skb->nh.ipv6h = hdr = (struct ipv6hdr*) skb_push(skb, sizeof(struct ipv6hdr));
YOSHIFUJI Hideaki1ab14572007-02-09 23:24:49 +09001350
Al Viro90bcaf72006-11-08 00:25:17 -08001351 *(__be32*)hdr = fl->fl6_flowlabel |
YOSHIFUJI Hideaki41a1f8e2005-09-08 10:19:03 +09001352 htonl(0x60000000 | ((int)np->cork.tclass << 20));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
1354 if (skb->len <= sizeof(struct ipv6hdr) + IPV6_MAXPLEN)
1355 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
1356 else
1357 hdr->payload_len = 0;
1358 hdr->hop_limit = np->cork.hop_limit;
1359 hdr->nexthdr = proto;
1360 ipv6_addr_copy(&hdr->saddr, &fl->fl6_src);
1361 ipv6_addr_copy(&hdr->daddr, final_dst);
1362
Patrick McHardya2c20642006-01-08 22:37:26 -08001363 skb->priority = sk->sk_priority;
1364
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 skb->dst = dst_clone(&rt->u.dst);
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +09001366 IP6_INC_STATS(rt->rt6i_idev, IPSTATS_MIB_OUTREQUESTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 err = NF_HOOK(PF_INET6, NF_IP6_LOCAL_OUT, skb, NULL, skb->dst->dev, dst_output);
1368 if (err) {
1369 if (err > 0)
Herbert Xu3320da82005-04-19 22:32:22 -07001370 err = np->recverr ? net_xmit_errno(err) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (err)
1372 goto error;
1373 }
1374
1375out:
1376 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001377 kfree(np->cork.opt);
1378 np->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 if (np->cork.rt) {
1380 dst_release(&np->cork.rt->u.dst);
1381 np->cork.rt = NULL;
1382 inet->cork.flags &= ~IPCORK_ALLFRAG;
1383 }
1384 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1385 return err;
1386error:
1387 goto out;
1388}
1389
1390void ip6_flush_pending_frames(struct sock *sk)
1391{
1392 struct inet_sock *inet = inet_sk(sk);
1393 struct ipv6_pinfo *np = inet6_sk(sk);
1394 struct sk_buff *skb;
1395
1396 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) {
YOSHIFUJI Hideakia11d2062006-11-04 20:11:37 +09001397 IP6_INC_STATS(ip6_dst_idev(skb->dst),
1398 IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 kfree_skb(skb);
1400 }
1401
1402 inet->cork.flags &= ~IPCORK_OPT;
1403
Jesper Juhla51482b2005-11-08 09:41:34 -08001404 kfree(np->cork.opt);
1405 np->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 if (np->cork.rt) {
1407 dst_release(&np->cork.rt->u.dst);
1408 np->cork.rt = NULL;
1409 inet->cork.flags &= ~IPCORK_ALLFRAG;
1410 }
1411 memset(&inet->cork.fl, 0, sizeof(inet->cork.fl));
1412}