blob: 3e7e910c7c0fd6f229a9ee67a10936f24db36a3f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The Internet Protocol (IP) output module.
7 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Richard Underwood
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
17 *
18 * See ip_input.c for original log
19 *
20 * Fixes:
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090023 * Bradford Johnson: Fix faulty handling of some frames when
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * no route is found.
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090034 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
41 * datagrams.
42 * Hirokazu Takahashi: sendfile() on UDP works now.
43 */
44
45#include <asm/uaccess.h>
46#include <asm/system.h>
47#include <linux/module.h>
48#include <linux/types.h>
49#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/mm.h>
51#include <linux/string.h>
52#include <linux/errno.h>
Al Viroa1f8e7f2006-10-19 16:08:53 -040053#include <linux/highmem.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <linux/socket.h>
56#include <linux/sockios.h>
57#include <linux/in.h>
58#include <linux/inet.h>
59#include <linux/netdevice.h>
60#include <linux/etherdevice.h>
61#include <linux/proc_fs.h>
62#include <linux/stat.h>
63#include <linux/init.h>
64
65#include <net/snmp.h>
66#include <net/ip.h>
67#include <net/protocol.h>
68#include <net/route.h>
Patrick McHardycfacb052006-01-08 22:36:54 -080069#include <net/xfrm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/skbuff.h>
71#include <net/sock.h>
72#include <net/arp.h>
73#include <net/icmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <net/checksum.h>
75#include <net/inetpeer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/igmp.h>
77#include <linux/netfilter_ipv4.h>
78#include <linux/netfilter_bridge.h>
79#include <linux/mroute.h>
80#include <linux/netlink.h>
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -070081#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Brian Haleyab32ea52006-09-22 14:15:41 -070083int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85/* Generate a checksum for an outgoing IP datagram. */
86__inline__ void ip_send_check(struct iphdr *iph)
87{
88 iph->check = 0;
89 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
90}
91
Herbert Xuc439cb22008-01-11 19:14:00 -080092int __ip_local_out(struct sk_buff *skb)
93{
94 struct iphdr *iph = ip_hdr(skb);
95
96 iph->tot_len = htons(skb->len);
97 ip_send_check(iph);
Patrick McHardy6e23ae22007-11-19 18:53:30 -080098 return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb->dst->dev,
Herbert Xuc439cb22008-01-11 19:14:00 -080099 dst_output);
100}
101
102int ip_local_out(struct sk_buff *skb)
103{
104 int err;
105
106 err = __ip_local_out(skb);
107 if (likely(err == 1))
108 err = dst_output(skb);
109
110 return err;
111}
112EXPORT_SYMBOL_GPL(ip_local_out);
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/* dev_loopback_xmit for use with netfilter. */
115static int ip_dev_loopback_xmit(struct sk_buff *newskb)
116{
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -0700117 skb_reset_mac_header(newskb);
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -0300118 __skb_pull(newskb, skb_network_offset(newskb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 newskb->pkt_type = PACKET_LOOPBACK;
120 newskb->ip_summed = CHECKSUM_UNNECESSARY;
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700121 WARN_ON(!newskb->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 netif_rx(newskb);
123 return 0;
124}
125
126static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
127{
128 int ttl = inet->uc_ttl;
129
130 if (ttl < 0)
131 ttl = dst_metric(dst, RTAX_HOPLIMIT);
132 return ttl;
133}
134
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900135/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 * Add an ip header to a skbuff and send it out.
137 *
138 */
139int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
Al Viro13d8eaa2006-09-26 22:27:30 -0700140 __be32 saddr, __be32 daddr, struct ip_options *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetee6b9672008-03-05 18:30:47 -0800143 struct rtable *rt = skb->rtable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 struct iphdr *iph;
145
146 /* Build the IP header. */
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300147 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
148 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700149 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 iph->version = 4;
151 iph->ihl = 5;
152 iph->tos = inet->tos;
153 if (ip_dont_fragment(sk, &rt->u.dst))
154 iph->frag_off = htons(IP_DF);
155 else
156 iph->frag_off = 0;
157 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
158 iph->daddr = rt->rt_dst;
159 iph->saddr = rt->rt_src;
160 iph->protocol = sk->sk_protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 if (opt && opt->optlen) {
164 iph->ihl += opt->optlen>>2;
165 ip_options_build(skb, opt, daddr, rt, 0);
166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800169 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 /* Send it out. */
Herbert Xuc439cb22008-01-11 19:14:00 -0800172 return ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -0700175EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static inline int ip_finish_output2(struct sk_buff *skb)
178{
179 struct dst_entry *dst = skb->dst;
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700180 struct rtable *rt = (struct rtable *)dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 struct net_device *dev = dst->dev;
Chuck Leverc2636b42007-10-23 21:07:32 -0700182 unsigned int hh_len = LL_RESERVED_SPACE(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700184 if (rt->rt_type == RTN_MULTICAST)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700185 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTMCASTPKTS);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700186 else if (rt->rt_type == RTN_BROADCAST)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700187 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTBCASTPKTS);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 /* Be paranoid, rather than too clever. */
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -0700190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 struct sk_buff *skb2;
192
193 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
194 if (skb2 == NULL) {
195 kfree_skb(skb);
196 return -ENOMEM;
197 }
198 if (skb->sk)
199 skb_set_owner_w(skb2, skb->sk);
200 kfree_skb(skb);
201 skb = skb2;
202 }
203
Stephen Hemminger3644f0c2006-12-07 15:08:17 -0800204 if (dst->hh)
205 return neigh_hh_output(dst->hh, skb);
206 else if (dst->neighbour)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 return dst->neighbour->output(skb);
208
209 if (net_ratelimit())
210 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
211 kfree_skb(skb);
212 return -EINVAL;
213}
214
John Heffner628a5c52007-04-20 15:53:27 -0700215static inline int ip_skb_dst_mtu(struct sk_buff *skb)
216{
217 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
218
219 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
220 skb->dst->dev->mtu : dst_mtu(skb->dst);
221}
222
Patrick McHardy861d0482007-10-15 01:48:39 -0700223static int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800225#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
226 /* Policy lookup after SNAT yielded a new policy */
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800227 if (skb->dst->xfrm != NULL) {
228 IPCB(skb)->flags |= IPSKB_REROUTED;
229 return dst_output(skb);
230 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800231#endif
John Heffner628a5c52007-04-20 15:53:27 -0700232 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800233 return ip_fragment(skb, ip_finish_output2);
234 else
235 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
238int ip_mc_output(struct sk_buff *skb)
239{
240 struct sock *sk = skb->sk;
Eric Dumazetee6b9672008-03-05 18:30:47 -0800241 struct rtable *rt = skb->rtable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 struct net_device *dev = rt->u.dst.dev;
243
244 /*
245 * If the indicated interface is up and running, send the packet.
246 */
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700247 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTREQUESTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249 skb->dev = dev;
250 skb->protocol = htons(ETH_P_IP);
251
252 /*
253 * Multicasts are looped back for other local users
254 */
255
256 if (rt->rt_flags&RTCF_MULTICAST) {
257 if ((!sk || inet_sk(sk)->mc_loop)
258#ifdef CONFIG_IP_MROUTE
259 /* Small optimization: do not loopback not local frames,
260 which returned after forwarding; they will be dropped
261 by ip_mr_input in any case.
262 Note, that local frames are looped back to be delivered
263 to local recipients.
264
265 This check is duplicated in ip_mr_input at the moment.
266 */
267 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
268#endif
269 ) {
270 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
271 if (newskb)
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800272 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb,
273 NULL, newskb->dev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 ip_dev_loopback_xmit);
275 }
276
277 /* Multicasts with ttl 0 must not go beyond the host */
278
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700279 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 kfree_skb(skb);
281 return 0;
282 }
283 }
284
285 if (rt->rt_flags&RTCF_BROADCAST) {
286 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
287 if (newskb)
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800288 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 newskb->dev, ip_dev_loopback_xmit);
290 }
291
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800292 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800293 ip_finish_output,
294 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
297int ip_output(struct sk_buff *skb)
298{
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800299 struct net_device *dev = skb->dst->dev;
300
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700301 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_OUTREQUESTS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800303 skb->dev = dev;
304 skb->protocol = htons(ETH_P_IP);
305
Patrick McHardy6e23ae22007-11-19 18:53:30 -0800306 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900307 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800308 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
David S. Millere89862f2007-01-26 01:04:55 -0800311int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
David S. Millere89862f2007-01-26 01:04:55 -0800313 struct sock *sk = skb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 struct inet_sock *inet = inet_sk(sk);
315 struct ip_options *opt = inet->opt;
316 struct rtable *rt;
317 struct iphdr *iph;
318
319 /* Skip all of this if the packet is already routed,
320 * f.e. by something like SCTP.
321 */
Eric Dumazetee6b9672008-03-05 18:30:47 -0800322 rt = skb->rtable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 if (rt != NULL)
324 goto packet_routed;
325
326 /* Make sure we can route this packet. */
327 rt = (struct rtable *)__sk_dst_check(sk, 0);
328 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700329 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331 /* Use correct destination address if we have options. */
332 daddr = inet->daddr;
333 if(opt && opt->srr)
334 daddr = opt->faddr;
335
336 {
337 struct flowi fl = { .oif = sk->sk_bound_dev_if,
338 .nl_u = { .ip4_u =
339 { .daddr = daddr,
340 .saddr = inet->saddr,
341 .tos = RT_CONN_FLAGS(sk) } },
342 .proto = sk->sk_protocol,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700343 .flags = inet_sk_flowi_flags(sk),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 .uli_u = { .ports =
345 { .sport = inet->sport,
346 .dport = inet->dport } } };
347
348 /* If this fails, retransmit mechanism of transport layer will
349 * keep trying until route appears or the connection times
350 * itself out.
351 */
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -0700352 security_sk_classify_flow(sk, &fl);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900353 if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 goto no_route;
355 }
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -0700356 sk_setup_caps(sk, &rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 }
358 skb->dst = dst_clone(&rt->u.dst);
359
360packet_routed:
361 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
362 goto no_route;
363
364 /* OK, we know where to send it, allocate and build IP header. */
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300365 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
366 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700367 iph = ip_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800368 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
370 iph->frag_off = htons(IP_DF);
371 else
372 iph->frag_off = 0;
373 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
374 iph->protocol = sk->sk_protocol;
375 iph->saddr = rt->rt_src;
376 iph->daddr = rt->rt_dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 /* Transport layer set skb->h.foo itself. */
378
379 if (opt && opt->optlen) {
380 iph->ihl += opt->optlen >> 2;
381 ip_options_build(skb, opt, inet->daddr, rt, 0);
382 }
383
Herbert Xu89f5f0a2005-11-08 09:41:56 -0800384 ip_select_ident_more(iph, &rt->u.dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700385 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800388 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Herbert Xuc439cb22008-01-11 19:14:00 -0800390 return ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392no_route:
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700393 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 kfree_skb(skb);
395 return -EHOSTUNREACH;
396}
397
398
399static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
400{
401 to->pkt_type = from->pkt_type;
402 to->priority = from->priority;
403 to->protocol = from->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 dst_release(to->dst);
405 to->dst = dst_clone(from->dst);
406 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800407 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409 /* Copy the flags to each fragment. */
410 IPCB(to)->flags = IPCB(from)->flags;
411
412#ifdef CONFIG_NET_SCHED
413 to->tc_index = from->tc_index;
414#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700415 nf_copy(to, from);
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700416#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
417 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
418 to->nf_trace = from->nf_trace;
419#endif
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300420#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
421 to->ipvs_property = from->ipvs_property;
422#endif
James Morris984bc162006-06-09 00:29:17 -0700423 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426/*
427 * This IP datagram is too large to be sent in one piece. Break it up into
428 * smaller pieces (each of size equal to IP header plus
429 * a block of the data of the original IP data part) that will yet fit in a
430 * single device frame, and queue such a frame for sending.
431 */
432
Jianjun Kongd9319102008-11-03 00:23:42 -0800433int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 struct iphdr *iph;
436 int raw = 0;
437 int ptr;
438 struct net_device *dev;
439 struct sk_buff *skb2;
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700440 unsigned int mtu, hlen, left, len, ll_rs, pad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800442 __be16 not_last_frag;
Eric Dumazetee6b9672008-03-05 18:30:47 -0800443 struct rtable *rt = skb->rtable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 int err = 0;
445
446 dev = rt->u.dst.dev;
447
448 /*
449 * Point into the IP datagram header.
450 */
451
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700452 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700455 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
John Heffner628a5c52007-04-20 15:53:27 -0700457 htonl(ip_skb_dst_mtu(skb)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 kfree_skb(skb);
459 return -EMSGSIZE;
460 }
461
462 /*
463 * Setup starting values.
464 */
465
466 hlen = iph->ihl * 4;
467 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800468 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 /* When frag_list is given, use it. First, check its validity:
471 * some transformers could create wrong frag_list or break existing
472 * one, it is not prohibited. In this case fall back to copying.
473 *
474 * LATER: this step can be merged to real generation of fragments,
475 * we can switch to copy when see the first bad fragment.
476 */
477 if (skb_shinfo(skb)->frag_list) {
478 struct sk_buff *frag;
479 int first_len = skb_pagelen(skb);
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800480 int truesizes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 if (first_len - hlen > mtu ||
483 ((first_len - hlen) & 7) ||
484 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
485 skb_cloned(skb))
486 goto slow_path;
487
488 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
489 /* Correct geometry. */
490 if (frag->len > mtu ||
491 ((frag->len & 7) && frag->next) ||
492 skb_headroom(frag) < hlen)
493 goto slow_path;
494
495 /* Partially cloned skb? */
496 if (skb_shared(frag))
497 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700498
499 BUG_ON(frag->sk);
500 if (skb->sk) {
501 sock_hold(skb->sk);
502 frag->sk = skb->sk;
503 frag->destructor = sock_wfree;
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800504 truesizes += frag->truesize;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 }
507
508 /* Everything is OK. Generate! */
509
510 err = 0;
511 offset = 0;
512 frag = skb_shinfo(skb)->frag_list;
513 skb_shinfo(skb)->frag_list = NULL;
514 skb->data_len = first_len - skb_headlen(skb);
Herbert Xu29ffe1a2008-01-28 20:45:20 -0800515 skb->truesize -= truesizes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 skb->len = first_len;
517 iph->tot_len = htons(first_len);
518 iph->frag_off = htons(IP_MF);
519 ip_send_check(iph);
520
521 for (;;) {
522 /* Prepare header of the next frame,
523 * before previous one went down. */
524 if (frag) {
525 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300526 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700527 __skb_push(frag, hlen);
528 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700529 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700530 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 iph->tot_len = htons(frag->len);
532 ip_copy_metadata(frag, skb);
533 if (offset == 0)
534 ip_options_fragment(frag);
535 offset += skb->len - hlen;
536 iph->frag_off = htons(offset>>3);
537 if (frag->next != NULL)
538 iph->frag_off |= htons(IP_MF);
539 /* Ready, complete checksum */
540 ip_send_check(iph);
541 }
542
543 err = output(skb);
544
Wei Dongdafee492006-08-02 13:41:21 -0700545 if (!err)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700546 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (err || !frag)
548 break;
549
550 skb = frag;
551 frag = skb->next;
552 skb->next = NULL;
553 }
554
555 if (err == 0) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700556 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return 0;
558 }
559
560 while (frag) {
561 skb = frag->next;
562 kfree_skb(frag);
563 frag = skb;
564 }
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700565 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 return err;
567 }
568
569slow_path:
570 left = skb->len - hlen; /* Space per frame */
571 ptr = raw + hlen; /* Where to start from */
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700574 * we need to make room for the encapsulating header
575 */
576 pad = nf_bridge_pad(skb);
577 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
578 mtu -= pad;
579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 /*
581 * Fragment the datagram.
582 */
583
584 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
585 not_last_frag = iph->frag_off & htons(IP_MF);
586
587 /*
588 * Keep copying data until we run out.
589 */
590
Stephen Hemminger132adf52007-03-08 20:44:43 -0800591 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 len = left;
593 /* IF: it doesn't fit, use 'mtu' - the data space left */
594 if (len > mtu)
595 len = mtu;
596 /* IF: we are not sending upto and including the packet end
597 then align the next start on an eight byte boundary */
598 if (len < left) {
599 len &= ~7;
600 }
601 /*
602 * Allocate buffer.
603 */
604
605 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700606 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 err = -ENOMEM;
608 goto fail;
609 }
610
611 /*
612 * Set up data on packet
613 */
614
615 ip_copy_metadata(skb2, skb);
616 skb_reserve(skb2, ll_rs);
617 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700618 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700619 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 /*
622 * Charge the memory for the fragment to any owner
623 * it might possess
624 */
625
626 if (skb->sk)
627 skb_set_owner_w(skb2, skb->sk);
628
629 /*
630 * Copy the packet header into the new buffer.
631 */
632
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300633 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635 /*
636 * Copy a block of the IP datagram.
637 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300638 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 BUG();
640 left -= len;
641
642 /*
643 * Fill in the new header fields.
644 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700645 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 iph->frag_off = htons((offset >> 3));
647
648 /* ANK: dirty, but effective trick. Upgrade options only if
649 * the segment to be fragmented was THE FIRST (otherwise,
650 * options are already fixed) and make it ONCE
651 * on the initial skb, so that all the following fragments
652 * will inherit fixed options.
653 */
654 if (offset == 0)
655 ip_options_fragment(skb);
656
657 /*
658 * Added AC : If we are fragmenting a fragment that's not the
659 * last fragment then keep MF on each bit
660 */
661 if (left > 0 || not_last_frag)
662 iph->frag_off |= htons(IP_MF);
663 ptr += len;
664 offset += len;
665
666 /*
667 * Put this fragment into the sending queue.
668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 iph->tot_len = htons(len + hlen);
670
671 ip_send_check(iph);
672
673 err = output(skb2);
674 if (err)
675 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700676
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700677 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700680 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return err;
682
683fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900684 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700685 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return err;
687}
688
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700689EXPORT_SYMBOL(ip_fragment);
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691int
692ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
693{
694 struct iovec *iov = from;
695
Patrick McHardy84fa7932006-08-29 16:44:56 -0700696 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
698 return -EFAULT;
699 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800700 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
702 return -EFAULT;
703 skb->csum = csum_block_add(skb->csum, csum, odd);
704 }
705 return 0;
706}
707
Al Viro44bb9362006-11-14 21:36:14 -0800708static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709csum_page(struct page *page, int offset, int copy)
710{
711 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800712 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 kaddr = kmap(page);
714 csum = csum_partial(kaddr + offset, copy, 0);
715 kunmap(page);
716 return csum;
717}
718
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800719static inline int ip_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700720 int getfrag(void *from, char *to, int offset, int len,
721 int odd, struct sk_buff *skb),
722 void *from, int length, int hh_len, int fragheaderlen,
Jianjun Kongd9319102008-11-03 00:23:42 -0800723 int transhdrlen, int mtu, unsigned int flags)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700724{
725 struct sk_buff *skb;
726 int err;
727
728 /* There is support for UDP fragmentation offload by network
729 * device, so create one single skb packet containing complete
730 * udp datagram
731 */
732 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
733 skb = sock_alloc_send_skb(sk,
734 hh_len + fragheaderlen + transhdrlen + 20,
735 (flags & MSG_DONTWAIT), &err);
736
737 if (skb == NULL)
738 return err;
739
740 /* reserve space for Hardware header */
741 skb_reserve(skb, hh_len);
742
743 /* create space for UDP/IP header */
Jianjun Kongd9319102008-11-03 00:23:42 -0800744 skb_put(skb, fragheaderlen + transhdrlen);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700745
746 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700747 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700748
749 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700750 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700751
Patrick McHardy84fa7932006-08-29 16:44:56 -0700752 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700753 skb->csum = 0;
754 sk->sk_sndmsg_off = 0;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700755
Kostya Bbe9164e2008-04-29 22:36:30 -0700756 /* specify the length of each IP datagram fragment */
Herbert Xu79671682006-06-22 02:40:14 -0700757 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700758 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700759 __skb_queue_tail(&sk->sk_write_queue, skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700760 }
Kostya Bbe9164e2008-04-29 22:36:30 -0700761
762 return skb_append_datato_frags(sk, skb, getfrag, from,
763 (length - transhdrlen));
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700764}
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766/*
767 * ip_append_data() and ip_append_page() can make one large IP datagram
768 * from many pieces of data. Each pieces will be holded on the socket
769 * until ip_push_pending_frames() is called. Each piece can be a page
770 * or non-page data.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900771 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 * Not only UDP, other transport protocols - e.g. raw sockets - can use
773 * this interface potentially.
774 *
775 * LATER: length must be adjusted by pad at tail, when it is required.
776 */
777int ip_append_data(struct sock *sk,
778 int getfrag(void *from, char *to, int offset, int len,
779 int odd, struct sk_buff *skb),
780 void *from, int length, int transhdrlen,
Eric Dumazet2e77d892008-11-24 15:52:46 -0800781 struct ipcm_cookie *ipc, struct rtable **rtp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 unsigned int flags)
783{
784 struct inet_sock *inet = inet_sk(sk);
785 struct sk_buff *skb;
786
787 struct ip_options *opt = NULL;
788 int hh_len;
789 int exthdrlen;
790 int mtu;
791 int copy;
792 int err;
793 int offset = 0;
794 unsigned int maxfraglen, fragheaderlen;
795 int csummode = CHECKSUM_NONE;
Eric Dumazet2e77d892008-11-24 15:52:46 -0800796 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 if (flags&MSG_PROBE)
799 return 0;
800
801 if (skb_queue_empty(&sk->sk_write_queue)) {
802 /*
803 * setup for corking.
804 */
805 opt = ipc->opt;
806 if (opt) {
807 if (inet->cork.opt == NULL) {
808 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
809 if (unlikely(inet->cork.opt == NULL))
810 return -ENOBUFS;
811 }
812 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
813 inet->cork.flags |= IPCORK_OPT;
814 inet->cork.addr = ipc->addr;
815 }
Eric Dumazet2e77d892008-11-24 15:52:46 -0800816 rt = *rtp;
817 /*
818 * We steal reference to this route, caller should not release it
819 */
820 *rtp = NULL;
John Heffner628a5c52007-04-20 15:53:27 -0700821 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
822 rt->u.dst.dev->mtu :
823 dst_mtu(rt->u.dst.path);
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -0400824 inet->cork.dst = &rt->u.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 inet->cork.length = 0;
826 sk->sk_sndmsg_page = NULL;
827 sk->sk_sndmsg_off = 0;
828 if ((exthdrlen = rt->u.dst.header_len) != 0) {
829 length += exthdrlen;
830 transhdrlen += exthdrlen;
831 }
832 } else {
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -0400833 rt = (struct rtable *)inet->cork.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (inet->cork.flags & IPCORK_OPT)
835 opt = inet->cork.opt;
836
837 transhdrlen = 0;
838 exthdrlen = 0;
839 mtu = inet->cork.fragsize;
840 }
841 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
842
843 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
844 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
845
846 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
847 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
848 return -EMSGSIZE;
849 }
850
851 /*
852 * transhdrlen > 0 means that this is the first fragment and we wish
853 * it won't be fragmented in the future.
854 */
855 if (transhdrlen &&
856 length + fragheaderlen <= mtu &&
Stephen Hemmingerd212f872007-06-27 00:47:37 -0700857 rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700859 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861 inet->cork.length += length;
Kostya Bbe9164e2008-04-29 22:36:30 -0700862 if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) &&
863 (sk->sk_protocol == IPPROTO_UDP) &&
864 (rt->u.dst.dev->features & NETIF_F_UFO)) {
Patrick McHardybaa829d2006-03-12 20:35:12 -0800865 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
866 fragheaderlen, transhdrlen, mtu,
867 flags);
868 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700869 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700870 return 0;
871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
873 /* So, what's going on in the loop below?
874 *
875 * We use calculated fragment length to generate chained skb,
876 * each of segments is IP fragment ready for sending to network after
877 * adding appropriate IP header.
878 */
879
880 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
881 goto alloc_new_skb;
882
883 while (length > 0) {
884 /* Check if the remaining data fits into current packet. */
885 copy = mtu - skb->len;
886 if (copy < length)
887 copy = maxfraglen - skb->len;
888 if (copy <= 0) {
889 char *data;
890 unsigned int datalen;
891 unsigned int fraglen;
892 unsigned int fraggap;
893 unsigned int alloclen;
894 struct sk_buff *skb_prev;
895alloc_new_skb:
896 skb_prev = skb;
897 if (skb_prev)
898 fraggap = skb_prev->len - maxfraglen;
899 else
900 fraggap = 0;
901
902 /*
903 * If remaining data exceeds the mtu,
904 * we know we need more fragment(s).
905 */
906 datalen = length + fraggap;
907 if (datalen > mtu - fragheaderlen)
908 datalen = maxfraglen - fragheaderlen;
909 fraglen = datalen + fragheaderlen;
910
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900911 if ((flags & MSG_MORE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 !(rt->u.dst.dev->features&NETIF_F_SG))
913 alloclen = mtu;
914 else
915 alloclen = datalen + fragheaderlen;
916
917 /* The last fragment gets additional space at tail.
918 * Note, with MSG_MORE we overallocate on fragments,
919 * because we have no idea what fragment will be
920 * the last.
921 */
Zach Brown3d9dd752006-04-14 16:04:18 -0700922 if (datalen == length + fraggap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 alloclen += rt->u.dst.trailer_len;
924
925 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900926 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 alloclen + hh_len + 15,
928 (flags & MSG_DONTWAIT), &err);
929 } else {
930 skb = NULL;
931 if (atomic_read(&sk->sk_wmem_alloc) <=
932 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900933 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 alloclen + hh_len + 15, 1,
935 sk->sk_allocation);
936 if (unlikely(skb == NULL))
937 err = -ENOBUFS;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000938 else
939 /* only the initial fragment is
940 time stamped */
941 ipc->shtx.flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 }
943 if (skb == NULL)
944 goto error;
945
946 /*
947 * Fill in the control structures
948 */
949 skb->ip_summed = csummode;
950 skb->csum = 0;
951 skb_reserve(skb, hh_len);
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000952 *skb_tx(skb) = ipc->shtx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 /*
955 * Find where to start putting bytes.
956 */
957 data = skb_put(skb, fraglen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -0300958 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700959 skb->transport_header = (skb->network_header +
960 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 data += fragheaderlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 if (fraggap) {
964 skb->csum = skb_copy_and_csum_bits(
965 skb_prev, maxfraglen,
966 data + transhdrlen, fraggap, 0);
967 skb_prev->csum = csum_sub(skb_prev->csum,
968 skb->csum);
969 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700970 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
972
973 copy = datalen - transhdrlen - fraggap;
974 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
975 err = -EFAULT;
976 kfree_skb(skb);
977 goto error;
978 }
979
980 offset += copy;
981 length -= datalen - fraggap;
982 transhdrlen = 0;
983 exthdrlen = 0;
984 csummode = CHECKSUM_NONE;
985
986 /*
987 * Put the packet on the pending queue.
988 */
989 __skb_queue_tail(&sk->sk_write_queue, skb);
990 continue;
991 }
992
993 if (copy > length)
994 copy = length;
995
996 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
997 unsigned int off;
998
999 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001000 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 offset, copy, off, skb) < 0) {
1002 __skb_trim(skb, off);
1003 err = -EFAULT;
1004 goto error;
1005 }
1006 } else {
1007 int i = skb_shinfo(skb)->nr_frags;
1008 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1009 struct page *page = sk->sk_sndmsg_page;
1010 int off = sk->sk_sndmsg_off;
1011 unsigned int left;
1012
1013 if (page && (left = PAGE_SIZE - off) > 0) {
1014 if (copy >= left)
1015 copy = left;
1016 if (page != frag->page) {
1017 if (i == MAX_SKB_FRAGS) {
1018 err = -EMSGSIZE;
1019 goto error;
1020 }
1021 get_page(page);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001022 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 frag = &skb_shinfo(skb)->frags[i];
1024 }
1025 } else if (i < MAX_SKB_FRAGS) {
1026 if (copy > PAGE_SIZE)
1027 copy = PAGE_SIZE;
1028 page = alloc_pages(sk->sk_allocation, 0);
1029 if (page == NULL) {
1030 err = -ENOMEM;
1031 goto error;
1032 }
1033 sk->sk_sndmsg_page = page;
1034 sk->sk_sndmsg_off = 0;
1035
1036 skb_fill_page_desc(skb, i, page, 0, 0);
1037 frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 } else {
1039 err = -EMSGSIZE;
1040 goto error;
1041 }
1042 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1043 err = -EFAULT;
1044 goto error;
1045 }
1046 sk->sk_sndmsg_off += copy;
1047 frag->size += copy;
1048 skb->len += copy;
1049 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001050 skb->truesize += copy;
1051 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 }
1053 offset += copy;
1054 length -= copy;
1055 }
1056
1057 return 0;
1058
1059error:
1060 inet->cork.length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001061 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001062 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}
1064
1065ssize_t ip_append_page(struct sock *sk, struct page *page,
1066 int offset, size_t size, int flags)
1067{
1068 struct inet_sock *inet = inet_sk(sk);
1069 struct sk_buff *skb;
1070 struct rtable *rt;
1071 struct ip_options *opt = NULL;
1072 int hh_len;
1073 int mtu;
1074 int len;
1075 int err;
1076 unsigned int maxfraglen, fragheaderlen, fraggap;
1077
1078 if (inet->hdrincl)
1079 return -EPERM;
1080
1081 if (flags&MSG_PROBE)
1082 return 0;
1083
1084 if (skb_queue_empty(&sk->sk_write_queue))
1085 return -EINVAL;
1086
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001087 rt = (struct rtable *)inet->cork.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if (inet->cork.flags & IPCORK_OPT)
1089 opt = inet->cork.opt;
1090
1091 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1092 return -EOPNOTSUPP;
1093
1094 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1095 mtu = inet->cork.fragsize;
1096
1097 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1098 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1099
1100 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1101 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1102 return -EMSGSIZE;
1103 }
1104
1105 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1106 return -EINVAL;
1107
1108 inet->cork.length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001109 if ((sk->sk_protocol == IPPROTO_UDP) &&
Herbert Xu79671682006-06-22 02:40:14 -07001110 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1111 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001112 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001113 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 while (size > 0) {
1117 int i;
1118
Herbert Xu89114af2006-07-08 13:34:32 -07001119 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001120 len = size;
1121 else {
1122
1123 /* Check if the remaining data fits into current packet. */
1124 len = mtu - skb->len;
1125 if (len < size)
1126 len = maxfraglen - skb->len;
1127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (len <= 0) {
1129 struct sk_buff *skb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 int alloclen;
1131
1132 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001133 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135 alloclen = fragheaderlen + hh_len + fraggap + 15;
1136 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1137 if (unlikely(!skb)) {
1138 err = -ENOBUFS;
1139 goto error;
1140 }
1141
1142 /*
1143 * Fill in the control structures
1144 */
1145 skb->ip_summed = CHECKSUM_NONE;
1146 skb->csum = 0;
1147 skb_reserve(skb, hh_len);
1148
1149 /*
1150 * Find where to start putting bytes.
1151 */
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001152 skb_put(skb, fragheaderlen + fraggap);
Arnaldo Carvalho de Melo2ca9e6f2007-03-10 19:15:25 -03001153 skb_reset_network_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001154 skb->transport_header = (skb->network_header +
1155 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (fraggap) {
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001157 skb->csum = skb_copy_and_csum_bits(skb_prev,
1158 maxfraglen,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001159 skb_transport_header(skb),
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001160 fraggap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 skb_prev->csum = csum_sub(skb_prev->csum,
1162 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001163 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 }
1165
1166 /*
1167 * Put the packet on the pending queue.
1168 */
1169 __skb_queue_tail(&sk->sk_write_queue, skb);
1170 continue;
1171 }
1172
1173 i = skb_shinfo(skb)->nr_frags;
1174 if (len > size)
1175 len = size;
1176 if (skb_can_coalesce(skb, i, page, offset)) {
1177 skb_shinfo(skb)->frags[i-1].size += len;
1178 } else if (i < MAX_SKB_FRAGS) {
1179 get_page(page);
1180 skb_fill_page_desc(skb, i, page, offset, len);
1181 } else {
1182 err = -EMSGSIZE;
1183 goto error;
1184 }
1185
1186 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001187 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 csum = csum_page(page, offset, len);
1189 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1190 }
1191
1192 skb->len += len;
1193 skb->data_len += len;
David S. Miller1e34a112008-01-22 23:44:31 -08001194 skb->truesize += len;
1195 atomic_add(len, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 offset += len;
1197 size -= len;
1198 }
1199 return 0;
1200
1201error:
1202 inet->cork.length -= size;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001203 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 return err;
1205}
1206
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001207static void ip_cork_release(struct inet_sock *inet)
1208{
1209 inet->cork.flags &= ~IPCORK_OPT;
1210 kfree(inet->cork.opt);
1211 inet->cork.opt = NULL;
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001212 dst_release(inet->cork.dst);
1213 inet->cork.dst = NULL;
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001214}
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216/*
1217 * Combined all pending IP fragments on the socket as one IP datagram
1218 * and push them out.
1219 */
1220int ip_push_pending_frames(struct sock *sk)
1221{
1222 struct sk_buff *skb, *tmp_skb;
1223 struct sk_buff **tail_skb;
1224 struct inet_sock *inet = inet_sk(sk);
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001225 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 struct ip_options *opt = NULL;
YOSHIFUJI Hideakic8cdaf92008-03-10 04:30:37 -04001227 struct rtable *rt = (struct rtable *)inet->cork.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001229 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 __u8 ttl;
1231 int err = 0;
1232
1233 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1234 goto out;
1235 tail_skb = &(skb_shinfo(skb)->frag_list);
1236
1237 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001238 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001239 __skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001241 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 *tail_skb = tmp_skb;
1243 tail_skb = &(tmp_skb->next);
1244 skb->len += tmp_skb->len;
1245 skb->data_len += tmp_skb->len;
1246 skb->truesize += tmp_skb->truesize;
1247 __sock_put(tmp_skb->sk);
1248 tmp_skb->destructor = NULL;
1249 tmp_skb->sk = NULL;
1250 }
1251
1252 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1253 * to fragment the frame generated here. No matter, what transforms
1254 * how transforms change size of the packet, it will come out.
1255 */
John Heffner628a5c52007-04-20 15:53:27 -07001256 if (inet->pmtudisc < IP_PMTUDISC_DO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 skb->local_df = 1;
1258
1259 /* DF bit is set when we want to see DF on outgoing frames.
1260 * If local_df is set too, we still allow to fragment this frame
1261 * locally. */
John Heffner628a5c52007-04-20 15:53:27 -07001262 if (inet->pmtudisc >= IP_PMTUDISC_DO ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 (skb->len <= dst_mtu(&rt->u.dst) &&
1264 ip_dont_fragment(sk, &rt->u.dst)))
1265 df = htons(IP_DF);
1266
1267 if (inet->cork.flags & IPCORK_OPT)
1268 opt = inet->cork.opt;
1269
1270 if (rt->rt_type == RTN_MULTICAST)
1271 ttl = inet->mc_ttl;
1272 else
1273 ttl = ip_select_ttl(inet, &rt->u.dst);
1274
1275 iph = (struct iphdr *)skb->data;
1276 iph->version = 4;
1277 iph->ihl = 5;
1278 if (opt) {
1279 iph->ihl += opt->optlen>>2;
1280 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1281 }
1282 iph->tos = inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 iph->frag_off = df;
Alexey Kuznetsov1a55d572006-03-22 14:27:59 -08001284 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 iph->ttl = ttl;
1286 iph->protocol = sk->sk_protocol;
1287 iph->saddr = rt->rt_src;
1288 iph->daddr = rt->rt_dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001291 skb->mark = sk->sk_mark;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001292 /*
1293 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1294 * on dst refcount
1295 */
1296 inet->cork.dst = NULL;
1297 skb->dst = &rt->u.dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
David L Stevens96793b42007-09-17 09:57:33 -07001299 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001300 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001301 skb_transport_header(skb))->type);
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 /* Netfilter gets whole the not fragmented skb. */
Herbert Xuc439cb22008-01-11 19:14:00 -08001304 err = ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (err) {
1306 if (err > 0)
1307 err = inet->recverr ? net_xmit_errno(err) : 0;
1308 if (err)
1309 goto error;
1310 }
1311
1312out:
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001313 ip_cork_release(inet);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return err;
1315
1316error:
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001317 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 goto out;
1319}
1320
1321/*
1322 * Throw away all pending data on the socket.
1323 */
1324void ip_flush_pending_frames(struct sock *sk)
1325{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 struct sk_buff *skb;
1327
1328 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1329 kfree_skb(skb);
1330
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001331 ip_cork_release(inet_sk(sk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332}
1333
1334
1335/*
1336 * Fetch data from kernel space and fill in checksum if needed.
1337 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001338static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 int len, int odd, struct sk_buff *skb)
1340{
Al Viro50842052006-11-14 21:36:34 -08001341 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1344 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001345 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
1347
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001348/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * Generic function to send a packet as reply to another packet.
1350 * Used to send TCP resets so far. ICMP should use this function too.
1351 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001352 * Should run single threaded per socket because it uses the sock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 * structure to pass arguments.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 */
1355void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1356 unsigned int len)
1357{
1358 struct inet_sock *inet = inet_sk(sk);
1359 struct {
1360 struct ip_options opt;
1361 char data[40];
1362 } replyopts;
1363 struct ipcm_cookie ipc;
Al Viro3ca3c682006-09-27 18:28:07 -07001364 __be32 daddr;
Eric Dumazetee6b9672008-03-05 18:30:47 -08001365 struct rtable *rt = skb->rtable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 if (ip_options_echo(&replyopts.opt, skb))
1368 return;
1369
1370 daddr = ipc.addr = rt->rt_src;
1371 ipc.opt = NULL;
Patrick Ohly51f31ca2009-02-12 05:03:39 +00001372 ipc.shtx.flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 if (replyopts.opt.optlen) {
1375 ipc.opt = &replyopts.opt;
1376
1377 if (ipc.opt->srr)
1378 daddr = replyopts.opt.faddr;
1379 }
1380
1381 {
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001382 struct flowi fl = { .oif = arg->bound_dev_if,
1383 .nl_u = { .ip4_u =
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 { .daddr = daddr,
1385 .saddr = rt->rt_spec_dst,
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001386 .tos = RT_TOS(ip_hdr(skb)->tos) } },
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 /* Not quite clean, but right. */
1388 .uli_u = { .ports =
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -07001389 { .sport = tcp_hdr(skb)->dest,
1390 .dport = tcp_hdr(skb)->source } },
KOVACS Krisztian86b08d82008-10-01 07:44:42 -07001391 .proto = sk->sk_protocol,
1392 .flags = ip_reply_arg_flowi_flags(arg) };
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001393 security_skb_classify_flow(skb, &fl);
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001394 if (ip_route_output_key(sock_net(sk), &rt, &fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return;
1396 }
1397
1398 /* And let IP do all the hard work.
1399
1400 This chunk is not reenterable, hence spinlock.
1401 Note that it uses the fact, that this function is called
1402 with locally disabled BH and that sk cannot be already spinlocked.
1403 */
1404 bh_lock_sock(sk);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001405 inet->tos = ip_hdr(skb)->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001407 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001408 sk->sk_bound_dev_if = arg->bound_dev_if;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
Eric Dumazet2e77d892008-11-24 15:52:46 -08001410 &ipc, &rt, MSG_DONTWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1412 if (arg->csumoffset >= 0)
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001413 *((__sum16 *)skb_transport_header(skb) +
1414 arg->csumoffset) = csum_fold(csum_add(skb->csum,
1415 arg->csum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 skb->ip_summed = CHECKSUM_NONE;
1417 ip_push_pending_frames(sk);
1418 }
1419
1420 bh_unlock_sock(sk);
1421
1422 ip_rt_put(rt);
1423}
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425void __init ip_init(void)
1426{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 ip_rt_init();
1428 inet_initpeers();
1429
1430#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1431 igmp_mc_proc_init();
1432#endif
1433}
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435EXPORT_SYMBOL(ip_generic_getfrag);
1436EXPORT_SYMBOL(ip_queue_xmit);
1437EXPORT_SYMBOL(ip_send_check);