blob: 4bcabf3ab4cad3bdc43f5b9ed33eba9c1357557d [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/module.h>
47#include <linux/types.h>
48#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mm.h>
50#include <linux/string.h>
51#include <linux/errno.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040052#include <linux/highmem.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090053#include <linux/slab.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;
David S. Miller323e1262010-12-12 21:55:08 -080084EXPORT_SYMBOL(sysctl_ip_default_ttl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86/* Generate a checksum for an outgoing IP datagram. */
Denis Efremov2fbd9672013-05-08 23:19:42 +000087void ip_send_check(struct iphdr *iph)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 iph->check = 0;
90 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
91}
Eric Dumazet4bc2f182010-07-09 21:22:10 +000092EXPORT_SYMBOL(ip_send_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Herbert Xuc439cb22008-01-11 19:14:00 -080094int __ip_local_out(struct sk_buff *skb)
95{
96 struct iphdr *iph = ip_hdr(skb);
97
98 iph->tot_len = htons(skb->len);
99 ip_send_check(iph);
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100100 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
101 skb_dst(skb)->dev, dst_output);
Herbert Xuc439cb22008-01-11 19:14:00 -0800102}
103
104int ip_local_out(struct sk_buff *skb)
105{
106 int err;
107
108 err = __ip_local_out(skb);
109 if (likely(err == 1))
110 err = dst_output(skb);
111
112 return err;
113}
114EXPORT_SYMBOL_GPL(ip_local_out);
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
117{
118 int ttl = inet->uc_ttl;
119
120 if (ttl < 0)
David S. Miller323e1262010-12-12 21:55:08 -0800121 ttl = ip4_dst_hoplimit(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return ttl;
123}
124
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * Add an ip header to a skbuff and send it out.
127 *
128 */
129int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000130 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
132 struct inet_sock *inet = inet_sk(sk);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000133 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 struct iphdr *iph;
135
136 /* Build the IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000137 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300138 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700139 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 iph->version = 4;
141 iph->ihl = 5;
142 iph->tos = inet->tos;
Changli Gaod8d1f302010-06-10 23:31:35 -0700143 if (ip_dont_fragment(sk, &rt->dst))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 iph->frag_off = htons(IP_DF);
145 else
146 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700147 iph->ttl = ip_select_ttl(inet, &rt->dst);
David S. Millerdd927a22011-05-04 12:03:30 -0700148 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
149 iph->saddr = saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 iph->protocol = sk->sk_protocol;
Changli Gaod8d1f302010-06-10 23:31:35 -0700151 ip_select_ident(iph, &rt->dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000153 if (opt && opt->opt.optlen) {
154 iph->ihl += opt->opt.optlen>>2;
155 ip_options_build(skb, &opt->opt, daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800159 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
161 /* Send it out. */
Herbert Xuc439cb22008-01-11 19:14:00 -0800162 return ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -0700164EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166static inline int ip_finish_output2(struct sk_buff *skb)
167{
Eric Dumazetadf30902009-06-02 05:19:30 +0000168 struct dst_entry *dst = skb_dst(skb);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700169 struct rtable *rt = (struct rtable *)dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 struct net_device *dev = dst->dev;
Chuck Leverc2636b42007-10-23 21:07:32 -0700171 unsigned int hh_len = LL_RESERVED_SPACE(dev);
David S. Millerf6b72b62011-07-14 07:53:20 -0700172 struct neighbour *neigh;
David S. Millera263b302012-07-02 02:02:15 -0700173 u32 nexthop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Neil Hormanedf391f2009-04-27 02:45:02 -0700175 if (rt->rt_type == RTN_MULTICAST) {
176 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
177 } else if (rt->rt_type == RTN_BROADCAST)
178 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 /* Be paranoid, rather than too clever. */
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -0700181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 struct sk_buff *skb2;
183
184 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
185 if (skb2 == NULL) {
186 kfree_skb(skb);
187 return -ENOMEM;
188 }
189 if (skb->sk)
190 skb_set_owner_w(skb2, skb->sk);
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000191 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 skb = skb2;
193 }
194
David S. Millera263b302012-07-02 02:02:15 -0700195 rcu_read_lock_bh();
Julian Anastasov155e8332012-10-08 11:41:18 +0000196 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
David S. Millera263b302012-07-02 02:02:15 -0700197 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
198 if (unlikely(!neigh))
199 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
Vasiliy Kulikov9871f1a2012-08-06 03:55:29 +0000200 if (!IS_ERR(neigh)) {
David S. Miller5110effe2012-07-02 02:21:03 -0700201 int res = dst_neigh_output(dst, neigh, skb);
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000202
David S. Millera263b302012-07-02 02:02:15 -0700203 rcu_read_unlock_bh();
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000204 return res;
205 }
David S. Millera263b302012-07-02 02:02:15 -0700206 rcu_read_unlock_bh();
David S. Miller05e3aa02011-07-16 17:26:00 -0700207
Joe Perchese87cc472012-05-13 21:56:26 +0000208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
209 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 kfree_skb(skb);
211 return -EINVAL;
212}
213
John Heffner628a5c52007-04-20 15:53:27 -0700214static inline int ip_skb_dst_mtu(struct sk_buff *skb)
215{
216 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
217
218 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
Eric Dumazetadf30902009-06-02 05:19:30 +0000219 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
John Heffner628a5c52007-04-20 15:53:27 -0700220}
221
Patrick McHardy861d0482007-10-15 01:48:39 -0700222static int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800224#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
225 /* Policy lookup after SNAT yielded a new policy */
Eric Dumazetadf30902009-06-02 05:19:30 +0000226 if (skb_dst(skb)->xfrm != NULL) {
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800227 IPCB(skb)->flags |= IPSKB_REROUTED;
228 return dst_output(skb);
229 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800230#endif
John Heffner628a5c52007-04-20 15:53:27 -0700231 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800232 return ip_fragment(skb, ip_finish_output2);
233 else
234 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
237int ip_mc_output(struct sk_buff *skb)
238{
239 struct sock *sk = skb->sk;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000240 struct rtable *rt = skb_rtable(skb);
Changli Gaod8d1f302010-06-10 23:31:35 -0700241 struct net_device *dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 /*
244 * If the indicated interface is up and running, send the packet.
245 */
Neil Hormanedf391f2009-04-27 02:45:02 -0700246 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
248 skb->dev = dev;
249 skb->protocol = htons(ETH_P_IP);
250
251 /*
252 * Multicasts are looped back for other local users
253 */
254
255 if (rt->rt_flags&RTCF_MULTICAST) {
Octavian Purdila7ad68482010-01-06 20:37:01 -0800256 if (sk_mc_loop(sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257#ifdef CONFIG_IP_MROUTE
258 /* Small optimization: do not loopback not local frames,
259 which returned after forwarding; they will be dropped
260 by ip_mr_input in any case.
261 Note, that local frames are looped back to be delivered
262 to local recipients.
263
264 This check is duplicated in ip_mr_input at the moment.
265 */
Joe Perches9d4fb272009-11-23 10:41:23 -0800266 &&
267 ((rt->rt_flags & RTCF_LOCAL) ||
268 !(IPCB(skb)->flags & IPSKB_FORWARDED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269#endif
Joe Perches9d4fb272009-11-23 10:41:23 -0800270 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
272 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100273 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
274 newskb, NULL, newskb->dev,
Michel Machado95603e22012-06-12 10:16:35 +0000275 dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277
278 /* Multicasts with ttl 0 must not go beyond the host */
279
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700280 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 kfree_skb(skb);
282 return 0;
283 }
284 }
285
286 if (rt->rt_flags&RTCF_BROADCAST) {
287 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
288 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100289 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
Michel Machado95603e22012-06-12 10:16:35 +0000290 NULL, newskb->dev, dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
292
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100293 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
294 skb->dev, ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800295 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
298int ip_output(struct sk_buff *skb)
299{
Eric Dumazetadf30902009-06-02 05:19:30 +0000300 struct net_device *dev = skb_dst(skb)->dev;
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800301
Neil Hormanedf391f2009-04-27 02:45:02 -0700302 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800304 skb->dev = dev;
305 skb->protocol = htons(ETH_P_IP);
306
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100307 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900308 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800309 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Eric Dumazet84f93072011-11-30 19:00:53 +0000312/*
313 * copy saddr and daddr, possibly using 64bit load/stores
314 * Equivalent to :
315 * iph->saddr = fl4->saddr;
316 * iph->daddr = fl4->daddr;
317 */
318static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
319{
320 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
321 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
322 memcpy(&iph->saddr, &fl4->saddr,
323 sizeof(fl4->saddr) + sizeof(fl4->daddr));
324}
325
David S. Millerd9d8da82011-05-06 22:23:20 -0700326int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
David S. Millere89862f2007-01-26 01:04:55 -0800328 struct sock *sk = skb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000330 struct ip_options_rcu *inet_opt;
David S. Millerb57ae012011-05-06 16:24:06 -0700331 struct flowi4 *fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 struct rtable *rt;
333 struct iphdr *iph;
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000334 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /* Skip all of this if the packet is already routed,
337 * f.e. by something like SCTP.
338 */
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000339 rcu_read_lock();
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000340 inet_opt = rcu_dereference(inet->inet_opt);
David S. Millerea4fc0d2011-05-06 22:30:20 -0700341 fl4 = &fl->u.ip4;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000342 rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (rt != NULL)
344 goto packet_routed;
345
346 /* Make sure we can route this packet. */
347 rt = (struct rtable *)__sk_dst_check(sk, 0);
348 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700349 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
351 /* Use correct destination address if we have options. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000352 daddr = inet->inet_daddr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000353 if (inet_opt && inet_opt->opt.srr)
354 daddr = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
David S. Miller78fbfd82011-03-12 00:00:52 -0500356 /* If this fails, retransmit mechanism of transport layer will
357 * keep trying until route appears or the connection times
358 * itself out.
359 */
David S. Millerb57ae012011-05-06 16:24:06 -0700360 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500361 daddr, inet->inet_saddr,
362 inet->inet_dport,
363 inet->inet_sport,
364 sk->sk_protocol,
365 RT_CONN_FLAGS(sk),
366 sk->sk_bound_dev_if);
367 if (IS_ERR(rt))
368 goto no_route;
Changli Gaod8d1f302010-06-10 23:31:35 -0700369 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
Changli Gaod8d1f302010-06-10 23:31:35 -0700371 skb_dst_set_noref(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373packet_routed:
Julian Anastasov155e8332012-10-08 11:41:18 +0000374 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 goto no_route;
376
377 /* OK, we know where to send it, allocate and build IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000378 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300379 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700380 iph = ip_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800381 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
Changli Gaod8d1f302010-06-10 23:31:35 -0700382 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 iph->frag_off = htons(IP_DF);
384 else
385 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700386 iph->ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +0000388 ip_copy_addrs(iph, fl4);
389
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 /* Transport layer set skb->h.foo itself. */
391
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000392 if (inet_opt && inet_opt->opt.optlen) {
393 iph->ihl += inet_opt->opt.optlen >> 2;
394 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396
Changli Gaod8d1f302010-06-10 23:31:35 -0700397 ip_select_ident_more(iph, &rt->dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700398 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800401 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000403 res = ip_local_out(skb);
404 rcu_read_unlock();
405 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407no_route:
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000408 rcu_read_unlock();
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700409 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 kfree_skb(skb);
411 return -EHOSTUNREACH;
412}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000413EXPORT_SYMBOL(ip_queue_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415
416static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
417{
418 to->pkt_type = from->pkt_type;
419 to->priority = from->priority;
420 to->protocol = from->protocol;
Eric Dumazetadf30902009-06-02 05:19:30 +0000421 skb_dst_drop(to);
Eric Dumazetfe76cda2010-07-01 23:48:22 +0000422 skb_dst_copy(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800424 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 /* Copy the flags to each fragment. */
427 IPCB(to)->flags = IPCB(from)->flags;
428
429#ifdef CONFIG_NET_SCHED
430 to->tc_index = from->tc_index;
431#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700432 nf_copy(to, from);
Gao fengf0165882013-03-21 19:48:42 +0000433#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700434 to->nf_trace = from->nf_trace;
435#endif
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300436#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
437 to->ipvs_property = from->ipvs_property;
438#endif
James Morris984bc162006-06-09 00:29:17 -0700439 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441
442/*
443 * This IP datagram is too large to be sent in one piece. Break it up into
444 * smaller pieces (each of size equal to IP header plus
445 * a block of the data of the original IP data part) that will yet fit in a
446 * single device frame, and queue such a frame for sending.
447 */
448
Jianjun Kongd93191002008-11-03 00:23:42 -0800449int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
451 struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 int ptr;
453 struct net_device *dev;
454 struct sk_buff *skb2;
Changli Gaoc893b802010-07-31 13:25:08 +0000455 unsigned int mtu, hlen, left, len, ll_rs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800457 __be16 not_last_frag;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000458 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 int err = 0;
460
Changli Gaod8d1f302010-06-10 23:31:35 -0700461 dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 /*
464 * Point into the IP datagram header.
465 */
466
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700467 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200469 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
470 (IPCB(skb)->frag_max_size &&
471 IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700472 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
John Heffner628a5c52007-04-20 15:53:27 -0700474 htonl(ip_skb_dst_mtu(skb)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 kfree_skb(skb);
476 return -EMSGSIZE;
477 }
478
479 /*
480 * Setup starting values.
481 */
482
483 hlen = iph->ihl * 4;
Changli Gaod8d1f302010-06-10 23:31:35 -0700484 mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */
Bart De Schuymer6c79bf02010-04-20 16:22:01 +0200485#ifdef CONFIG_BRIDGE_NETFILTER
486 if (skb->nf_bridge)
487 mtu -= nf_bridge_mtu_reduction(skb);
488#endif
Herbert Xu89cee8b2005-12-13 23:14:27 -0800489 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 /* When frag_list is given, use it. First, check its validity:
492 * some transformers could create wrong frag_list or break existing
493 * one, it is not prohibited. In this case fall back to copying.
494 *
495 * LATER: this step can be merged to real generation of fragments,
496 * we can switch to copy when see the first bad fragment.
497 */
David S. Miller21dc3302010-08-23 00:13:46 -0700498 if (skb_has_frag_list(skb)) {
Eric Dumazet3d130082010-09-21 08:47:45 +0000499 struct sk_buff *frag, *frag2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 int first_len = skb_pagelen(skb);
501
502 if (first_len - hlen > mtu ||
503 ((first_len - hlen) & 7) ||
Paul Gortmaker56f8a752011-06-21 20:33:34 -0700504 ip_is_fragment(iph) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 skb_cloned(skb))
506 goto slow_path;
507
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700508 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 /* Correct geometry. */
510 if (frag->len > mtu ||
511 ((frag->len & 7) && frag->next) ||
512 skb_headroom(frag) < hlen)
Eric Dumazet3d130082010-09-21 08:47:45 +0000513 goto slow_path_clean;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514
515 /* Partially cloned skb? */
516 if (skb_shared(frag))
Eric Dumazet3d130082010-09-21 08:47:45 +0000517 goto slow_path_clean;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700518
519 BUG_ON(frag->sk);
520 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700521 frag->sk = skb->sk;
522 frag->destructor = sock_wfree;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700523 }
Eric Dumazet3d130082010-09-21 08:47:45 +0000524 skb->truesize -= frag->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 }
526
527 /* Everything is OK. Generate! */
528
529 err = 0;
530 offset = 0;
531 frag = skb_shinfo(skb)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700532 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 skb->data_len = first_len - skb_headlen(skb);
534 skb->len = first_len;
535 iph->tot_len = htons(first_len);
536 iph->frag_off = htons(IP_MF);
537 ip_send_check(iph);
538
539 for (;;) {
540 /* Prepare header of the next frame,
541 * before previous one went down. */
542 if (frag) {
543 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300544 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700545 __skb_push(frag, hlen);
546 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700547 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700548 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 iph->tot_len = htons(frag->len);
550 ip_copy_metadata(frag, skb);
551 if (offset == 0)
552 ip_options_fragment(frag);
553 offset += skb->len - hlen;
554 iph->frag_off = htons(offset>>3);
555 if (frag->next != NULL)
556 iph->frag_off |= htons(IP_MF);
557 /* Ready, complete checksum */
558 ip_send_check(iph);
559 }
560
561 err = output(skb);
562
Wei Dongdafee492006-08-02 13:41:21 -0700563 if (!err)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700564 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (err || !frag)
566 break;
567
568 skb = frag;
569 frag = skb->next;
570 skb->next = NULL;
571 }
572
573 if (err == 0) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700574 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 return 0;
576 }
577
578 while (frag) {
579 skb = frag->next;
580 kfree_skb(frag);
581 frag = skb;
582 }
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700583 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return err;
Eric Dumazet3d130082010-09-21 08:47:45 +0000585
586slow_path_clean:
587 skb_walk_frags(skb, frag2) {
588 if (frag2 == frag)
589 break;
590 frag2->sk = NULL;
591 frag2->destructor = NULL;
592 skb->truesize += frag2->truesize;
593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
596slow_path:
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000597 /* for offloaded checksums cleanup checksum before fragmentation */
598 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
599 goto fail;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000600 iph = ip_hdr(skb);
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 left = skb->len - hlen; /* Space per frame */
George Kadianakis49085bd2010-07-06 11:44:12 +0000603 ptr = hlen; /* Where to start from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700606 * we need to make room for the encapsulating header
607 */
Changli Gaoc893b802010-07-31 13:25:08 +0000608 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 /*
611 * Fragment the datagram.
612 */
613
614 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
615 not_last_frag = iph->frag_off & htons(IP_MF);
616
617 /*
618 * Keep copying data until we run out.
619 */
620
Stephen Hemminger132adf52007-03-08 20:44:43 -0800621 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 len = left;
623 /* IF: it doesn't fit, use 'mtu' - the data space left */
624 if (len > mtu)
625 len = mtu;
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300626 /* IF: we are not sending up to and including the packet end
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 then align the next start on an eight byte boundary */
628 if (len < left) {
629 len &= ~7;
630 }
631 /*
632 * Allocate buffer.
633 */
634
635 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700636 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 err = -ENOMEM;
638 goto fail;
639 }
640
641 /*
642 * Set up data on packet
643 */
644
645 ip_copy_metadata(skb2, skb);
646 skb_reserve(skb2, ll_rs);
647 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700648 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700649 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651 /*
652 * Charge the memory for the fragment to any owner
653 * it might possess
654 */
655
656 if (skb->sk)
657 skb_set_owner_w(skb2, skb->sk);
658
659 /*
660 * Copy the packet header into the new buffer.
661 */
662
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300663 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665 /*
666 * Copy a block of the IP datagram.
667 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300668 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 BUG();
670 left -= len;
671
672 /*
673 * Fill in the new header fields.
674 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700675 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 iph->frag_off = htons((offset >> 3));
677
678 /* ANK: dirty, but effective trick. Upgrade options only if
679 * the segment to be fragmented was THE FIRST (otherwise,
680 * options are already fixed) and make it ONCE
681 * on the initial skb, so that all the following fragments
682 * will inherit fixed options.
683 */
684 if (offset == 0)
685 ip_options_fragment(skb);
686
687 /*
688 * Added AC : If we are fragmenting a fragment that's not the
689 * last fragment then keep MF on each bit
690 */
691 if (left > 0 || not_last_frag)
692 iph->frag_off |= htons(IP_MF);
693 ptr += len;
694 offset += len;
695
696 /*
697 * Put this fragment into the sending queue.
698 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 iph->tot_len = htons(len + hlen);
700
701 ip_send_check(iph);
702
703 err = output(skb2);
704 if (err)
705 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700706
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700707 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000709 consume_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700710 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return err;
712
713fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900714 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700715 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 return err;
717}
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700718EXPORT_SYMBOL(ip_fragment);
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720int
721ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
722{
723 struct iovec *iov = from;
724
Patrick McHardy84fa7932006-08-29 16:44:56 -0700725 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
727 return -EFAULT;
728 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800729 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
731 return -EFAULT;
732 skb->csum = csum_block_add(skb->csum, csum, odd);
733 }
734 return 0;
735}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000736EXPORT_SYMBOL(ip_generic_getfrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
Al Viro44bb9362006-11-14 21:36:14 -0800738static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739csum_page(struct page *page, int offset, int copy)
740{
741 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800742 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 kaddr = kmap(page);
744 csum = csum_partial(kaddr + offset, copy, 0);
745 kunmap(page);
746 return csum;
747}
748
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800749static inline int ip_ufo_append_data(struct sock *sk,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000750 struct sk_buff_head *queue,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700751 int getfrag(void *from, char *to, int offset, int len,
752 int odd, struct sk_buff *skb),
753 void *from, int length, int hh_len, int fragheaderlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000754 int transhdrlen, int maxfraglen, unsigned int flags)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700755{
756 struct sk_buff *skb;
757 int err;
758
759 /* There is support for UDP fragmentation offload by network
760 * device, so create one single skb packet containing complete
761 * udp datagram
762 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000763 if ((skb = skb_peek_tail(queue)) == NULL) {
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700764 skb = sock_alloc_send_skb(sk,
765 hh_len + fragheaderlen + transhdrlen + 20,
766 (flags & MSG_DONTWAIT), &err);
767
768 if (skb == NULL)
769 return err;
770
771 /* reserve space for Hardware header */
772 skb_reserve(skb, hh_len);
773
774 /* create space for UDP/IP header */
Jianjun Kongd93191002008-11-03 00:23:42 -0800775 skb_put(skb, fragheaderlen + transhdrlen);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700776
777 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700778 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700779
780 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700781 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700782
Patrick McHardy84fa7932006-08-29 16:44:56 -0700783 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700784 skb->csum = 0;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700785
Kostya Bbe9164e2008-04-29 22:36:30 -0700786 /* specify the length of each IP datagram fragment */
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000787 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700788 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000789 __skb_queue_tail(queue, skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700790 }
Kostya Bbe9164e2008-04-29 22:36:30 -0700791
792 return skb_append_datato_frags(sk, skb, getfrag, from,
793 (length - transhdrlen));
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700794}
795
David S. Millerf5fca602011-05-08 17:24:10 -0700796static int __ip_append_data(struct sock *sk,
797 struct flowi4 *fl4,
798 struct sk_buff_head *queue,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000799 struct inet_cork *cork,
Eric Dumazet5640f762012-09-23 23:04:42 +0000800 struct page_frag *pfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000801 int getfrag(void *from, char *to, int offset,
802 int len, int odd, struct sk_buff *skb),
803 void *from, int length, int transhdrlen,
804 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
806 struct inet_sock *inet = inet_sk(sk);
807 struct sk_buff *skb;
808
Herbert Xu07df5292011-03-01 23:00:58 -0800809 struct ip_options *opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 int hh_len;
811 int exthdrlen;
812 int mtu;
813 int copy;
814 int err;
815 int offset = 0;
816 unsigned int maxfraglen, fragheaderlen;
817 int csummode = CHECKSUM_NONE;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000818 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Steffen Klassert96d73032011-06-05 20:48:47 +0000820 skb = skb_peek_tail(queue);
821
822 exthdrlen = !skb ? rt->dst.header_len : 0;
Herbert Xu07df5292011-03-01 23:00:58 -0800823 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Changli Gaod8d1f302010-06-10 23:31:35 -0700825 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
828 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
829
Herbert Xu1470ddf2011-03-01 02:36:47 +0000830 if (cork->length + length > 0xFFFF - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -0700831 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000832 mtu-exthdrlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return -EMSGSIZE;
834 }
835
836 /*
837 * transhdrlen > 0 means that this is the first fragment and we wish
838 * it won't be fragmented in the future.
839 */
840 if (transhdrlen &&
841 length + fragheaderlen <= mtu &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700842 rt->dst.dev->features & NETIF_F_V4_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700844 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845
Herbert Xu1470ddf2011-03-01 02:36:47 +0000846 cork->length += length;
Herbert Xu26cde9f2010-06-15 01:52:25 +0000847 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
Kostya Bbe9164e2008-04-29 22:36:30 -0700848 (sk->sk_protocol == IPPROTO_UDP) &&
Steffen Klassertc1460662011-06-29 23:19:32 +0000849 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
Herbert Xu1470ddf2011-03-01 02:36:47 +0000850 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
851 hh_len, fragheaderlen, transhdrlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000852 maxfraglen, flags);
Patrick McHardybaa829d2006-03-12 20:35:12 -0800853 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700854 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700855 return 0;
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 /* So, what's going on in the loop below?
859 *
860 * We use calculated fragment length to generate chained skb,
861 * each of segments is IP fragment ready for sending to network after
862 * adding appropriate IP header.
863 */
864
Herbert Xu26cde9f2010-06-15 01:52:25 +0000865 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 goto alloc_new_skb;
867
868 while (length > 0) {
869 /* Check if the remaining data fits into current packet. */
870 copy = mtu - skb->len;
871 if (copy < length)
872 copy = maxfraglen - skb->len;
873 if (copy <= 0) {
874 char *data;
875 unsigned int datalen;
876 unsigned int fraglen;
877 unsigned int fraggap;
878 unsigned int alloclen;
879 struct sk_buff *skb_prev;
880alloc_new_skb:
881 skb_prev = skb;
882 if (skb_prev)
883 fraggap = skb_prev->len - maxfraglen;
884 else
885 fraggap = 0;
886
887 /*
888 * If remaining data exceeds the mtu,
889 * we know we need more fragment(s).
890 */
891 datalen = length + fraggap;
892 if (datalen > mtu - fragheaderlen)
893 datalen = maxfraglen - fragheaderlen;
894 fraglen = datalen + fragheaderlen;
895
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900896 if ((flags & MSG_MORE) &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700897 !(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 alloclen = mtu;
899 else
Eric Dumazet59104f02010-09-20 20:16:27 +0000900 alloclen = fraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Steffen Klassert353e5c92011-06-22 01:05:37 +0000902 alloclen += exthdrlen;
903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 /* The last fragment gets additional space at tail.
905 * Note, with MSG_MORE we overallocate on fragments,
906 * because we have no idea what fragment will be
907 * the last.
908 */
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000909 if (datalen == length + fraggap)
Changli Gaod8d1f302010-06-10 23:31:35 -0700910 alloclen += rt->dst.trailer_len;
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900913 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 alloclen + hh_len + 15,
915 (flags & MSG_DONTWAIT), &err);
916 } else {
917 skb = NULL;
918 if (atomic_read(&sk->sk_wmem_alloc) <=
919 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900920 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 alloclen + hh_len + 15, 1,
922 sk->sk_allocation);
923 if (unlikely(skb == NULL))
924 err = -ENOBUFS;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000925 else
926 /* only the initial fragment is
927 time stamped */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000928 cork->tx_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 }
930 if (skb == NULL)
931 goto error;
932
933 /*
934 * Fill in the control structures
935 */
936 skb->ip_summed = csummode;
937 skb->csum = 0;
938 skb_reserve(skb, hh_len);
Herbert Xu1470ddf2011-03-01 02:36:47 +0000939 skb_shinfo(skb)->tx_flags = cork->tx_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
941 /*
942 * Find where to start putting bytes.
943 */
Steffen Klassert353e5c92011-06-22 01:05:37 +0000944 data = skb_put(skb, fraglen + exthdrlen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -0300945 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700946 skb->transport_header = (skb->network_header +
947 fragheaderlen);
Steffen Klassert353e5c92011-06-22 01:05:37 +0000948 data += fragheaderlen + exthdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 if (fraggap) {
951 skb->csum = skb_copy_and_csum_bits(
952 skb_prev, maxfraglen,
953 data + transhdrlen, fraggap, 0);
954 skb_prev->csum = csum_sub(skb_prev->csum,
955 skb->csum);
956 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700957 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 }
959
960 copy = datalen - transhdrlen - fraggap;
961 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
962 err = -EFAULT;
963 kfree_skb(skb);
964 goto error;
965 }
966
967 offset += copy;
968 length -= datalen - fraggap;
969 transhdrlen = 0;
970 exthdrlen = 0;
971 csummode = CHECKSUM_NONE;
972
973 /*
974 * Put the packet on the pending queue.
975 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000976 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 continue;
978 }
979
980 if (copy > length)
981 copy = length;
982
Changli Gaod8d1f302010-06-10 23:31:35 -0700983 if (!(rt->dst.dev->features&NETIF_F_SG)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 unsigned int off;
985
986 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900987 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 offset, copy, off, skb) < 0) {
989 __skb_trim(skb, off);
990 err = -EFAULT;
991 goto error;
992 }
993 } else {
994 int i = skb_shinfo(skb)->nr_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Eric Dumazet5640f762012-09-23 23:04:42 +0000996 err = -ENOMEM;
997 if (!sk_page_frag_refill(sk, pfrag))
998 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Eric Dumazet5640f762012-09-23 23:04:42 +00001000 if (!skb_can_coalesce(skb, i, pfrag->page,
1001 pfrag->offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 err = -EMSGSIZE;
Eric Dumazet5640f762012-09-23 23:04:42 +00001003 if (i == MAX_SKB_FRAGS)
1004 goto error;
1005
1006 __skb_fill_page_desc(skb, i, pfrag->page,
1007 pfrag->offset, 0);
1008 skb_shinfo(skb)->nr_frags = ++i;
1009 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001011 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1012 if (getfrag(from,
1013 page_address(pfrag->page) + pfrag->offset,
1014 offset, copy, skb->len, skb) < 0)
1015 goto error_efault;
1016
1017 pfrag->offset += copy;
1018 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 skb->len += copy;
1020 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001021 skb->truesize += copy;
1022 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
1024 offset += copy;
1025 length -= copy;
1026 }
1027
1028 return 0;
1029
Eric Dumazet5640f762012-09-23 23:04:42 +00001030error_efault:
1031 err = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032error:
Herbert Xu1470ddf2011-03-01 02:36:47 +00001033 cork->length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001034 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001035 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
1037
Herbert Xu1470ddf2011-03-01 02:36:47 +00001038static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1039 struct ipcm_cookie *ipc, struct rtable **rtp)
1040{
1041 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001042 struct ip_options_rcu *opt;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001043 struct rtable *rt;
1044
1045 /*
1046 * setup for corking.
1047 */
1048 opt = ipc->opt;
1049 if (opt) {
1050 if (cork->opt == NULL) {
1051 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1052 sk->sk_allocation);
1053 if (unlikely(cork->opt == NULL))
1054 return -ENOBUFS;
1055 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001056 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001057 cork->flags |= IPCORK_OPT;
1058 cork->addr = ipc->addr;
1059 }
1060 rt = *rtp;
1061 if (unlikely(!rt))
1062 return -EFAULT;
1063 /*
1064 * We steal reference to this route, caller should not release it
1065 */
1066 *rtp = NULL;
1067 cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ?
Steffen Klassert353e5c92011-06-22 01:05:37 +00001068 rt->dst.dev->mtu : dst_mtu(&rt->dst);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001069 cork->dst = &rt->dst;
1070 cork->length = 0;
1071 cork->tx_flags = ipc->tx_flags;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001072
1073 return 0;
1074}
1075
1076/*
1077 * ip_append_data() and ip_append_page() can make one large IP datagram
1078 * from many pieces of data. Each pieces will be holded on the socket
1079 * until ip_push_pending_frames() is called. Each piece can be a page
1080 * or non-page data.
1081 *
1082 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1083 * this interface potentially.
1084 *
1085 * LATER: length must be adjusted by pad at tail, when it is required.
1086 */
David S. Millerf5fca602011-05-08 17:24:10 -07001087int ip_append_data(struct sock *sk, struct flowi4 *fl4,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001088 int getfrag(void *from, char *to, int offset, int len,
1089 int odd, struct sk_buff *skb),
1090 void *from, int length, int transhdrlen,
1091 struct ipcm_cookie *ipc, struct rtable **rtp,
1092 unsigned int flags)
1093{
1094 struct inet_sock *inet = inet_sk(sk);
1095 int err;
1096
1097 if (flags&MSG_PROBE)
1098 return 0;
1099
1100 if (skb_queue_empty(&sk->sk_write_queue)) {
David S. Millerbdc712b2011-05-06 15:02:07 -07001101 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001102 if (err)
1103 return err;
1104 } else {
1105 transhdrlen = 0;
1106 }
1107
Eric Dumazet5640f762012-09-23 23:04:42 +00001108 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1109 sk_page_frag(sk), getfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001110 from, length, transhdrlen, flags);
1111}
1112
David S. Millerf5fca602011-05-08 17:24:10 -07001113ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 int offset, size_t size, int flags)
1115{
1116 struct inet_sock *inet = inet_sk(sk);
1117 struct sk_buff *skb;
1118 struct rtable *rt;
1119 struct ip_options *opt = NULL;
David S. Millerbdc712b2011-05-06 15:02:07 -07001120 struct inet_cork *cork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 int hh_len;
1122 int mtu;
1123 int len;
1124 int err;
1125 unsigned int maxfraglen, fragheaderlen, fraggap;
1126
1127 if (inet->hdrincl)
1128 return -EPERM;
1129
1130 if (flags&MSG_PROBE)
1131 return 0;
1132
1133 if (skb_queue_empty(&sk->sk_write_queue))
1134 return -EINVAL;
1135
David S. Millerbdc712b2011-05-06 15:02:07 -07001136 cork = &inet->cork.base;
1137 rt = (struct rtable *)cork->dst;
1138 if (cork->flags & IPCORK_OPT)
1139 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Changli Gaod8d1f302010-06-10 23:31:35 -07001141 if (!(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 return -EOPNOTSUPP;
1143
Changli Gaod8d1f302010-06-10 23:31:35 -07001144 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
David S. Millerbdc712b2011-05-06 15:02:07 -07001145 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
1147 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1148 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1149
David S. Millerbdc712b2011-05-06 15:02:07 -07001150 if (cork->length + size > 0xFFFF - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -07001151 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, mtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 return -EMSGSIZE;
1153 }
1154
1155 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1156 return -EINVAL;
1157
David S. Millerbdc712b2011-05-06 15:02:07 -07001158 cork->length += size;
Herbert Xu26cde9f2010-06-15 01:52:25 +00001159 if ((size + skb->len > mtu) &&
1160 (sk->sk_protocol == IPPROTO_UDP) &&
Changli Gaod8d1f302010-06-10 23:31:35 -07001161 (rt->dst.dev->features & NETIF_F_UFO)) {
Herbert Xu79671682006-06-22 02:40:14 -07001162 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001163 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001164 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
1167 while (size > 0) {
1168 int i;
1169
Herbert Xu89114af2006-07-08 13:34:32 -07001170 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001171 len = size;
1172 else {
1173
1174 /* Check if the remaining data fits into current packet. */
1175 len = mtu - skb->len;
1176 if (len < size)
1177 len = maxfraglen - skb->len;
1178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (len <= 0) {
1180 struct sk_buff *skb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 int alloclen;
1182
1183 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001184 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186 alloclen = fragheaderlen + hh_len + fraggap + 15;
1187 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1188 if (unlikely(!skb)) {
1189 err = -ENOBUFS;
1190 goto error;
1191 }
1192
1193 /*
1194 * Fill in the control structures
1195 */
1196 skb->ip_summed = CHECKSUM_NONE;
1197 skb->csum = 0;
1198 skb_reserve(skb, hh_len);
1199
1200 /*
1201 * Find where to start putting bytes.
1202 */
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001203 skb_put(skb, fragheaderlen + fraggap);
Arnaldo Carvalho de Melo2ca9e6f2007-03-10 19:15:25 -03001204 skb_reset_network_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001205 skb->transport_header = (skb->network_header +
1206 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 if (fraggap) {
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001208 skb->csum = skb_copy_and_csum_bits(skb_prev,
1209 maxfraglen,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001210 skb_transport_header(skb),
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001211 fraggap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 skb_prev->csum = csum_sub(skb_prev->csum,
1213 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001214 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 /*
1218 * Put the packet on the pending queue.
1219 */
1220 __skb_queue_tail(&sk->sk_write_queue, skb);
1221 continue;
1222 }
1223
1224 i = skb_shinfo(skb)->nr_frags;
1225 if (len > size)
1226 len = size;
1227 if (skb_can_coalesce(skb, i, page, offset)) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001228 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 } else if (i < MAX_SKB_FRAGS) {
1230 get_page(page);
1231 skb_fill_page_desc(skb, i, page, offset, len);
1232 } else {
1233 err = -EMSGSIZE;
1234 goto error;
1235 }
1236
1237 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001238 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 csum = csum_page(page, offset, len);
1240 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1241 }
1242
1243 skb->len += len;
1244 skb->data_len += len;
David S. Miller1e34a112008-01-22 23:44:31 -08001245 skb->truesize += len;
1246 atomic_add(len, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 offset += len;
1248 size -= len;
1249 }
1250 return 0;
1251
1252error:
David S. Millerbdc712b2011-05-06 15:02:07 -07001253 cork->length -= size;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001254 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return err;
1256}
1257
Herbert Xu1470ddf2011-03-01 02:36:47 +00001258static void ip_cork_release(struct inet_cork *cork)
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001259{
Herbert Xu1470ddf2011-03-01 02:36:47 +00001260 cork->flags &= ~IPCORK_OPT;
1261 kfree(cork->opt);
1262 cork->opt = NULL;
1263 dst_release(cork->dst);
1264 cork->dst = NULL;
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001265}
1266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267/*
1268 * Combined all pending IP fragments on the socket as one IP datagram
1269 * and push them out.
1270 */
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001271struct sk_buff *__ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001272 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001273 struct sk_buff_head *queue,
1274 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275{
1276 struct sk_buff *skb, *tmp_skb;
1277 struct sk_buff **tail_skb;
1278 struct inet_sock *inet = inet_sk(sk);
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001279 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 struct ip_options *opt = NULL;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001281 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001283 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 __u8 ttl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Herbert Xu1470ddf2011-03-01 02:36:47 +00001286 if ((skb = __skb_dequeue(queue)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 goto out;
1288 tail_skb = &(skb_shinfo(skb)->frag_list);
1289
1290 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001291 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001292 __skb_pull(skb, skb_network_offset(skb));
Herbert Xu1470ddf2011-03-01 02:36:47 +00001293 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001294 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 *tail_skb = tmp_skb;
1296 tail_skb = &(tmp_skb->next);
1297 skb->len += tmp_skb->len;
1298 skb->data_len += tmp_skb->len;
1299 skb->truesize += tmp_skb->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 tmp_skb->destructor = NULL;
1301 tmp_skb->sk = NULL;
1302 }
1303
1304 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1305 * to fragment the frame generated here. No matter, what transforms
1306 * how transforms change size of the packet, it will come out.
1307 */
John Heffner628a5c52007-04-20 15:53:27 -07001308 if (inet->pmtudisc < IP_PMTUDISC_DO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 skb->local_df = 1;
1310
1311 /* DF bit is set when we want to see DF on outgoing frames.
1312 * If local_df is set too, we still allow to fragment this frame
1313 * locally. */
John Heffner628a5c52007-04-20 15:53:27 -07001314 if (inet->pmtudisc >= IP_PMTUDISC_DO ||
Changli Gaod8d1f302010-06-10 23:31:35 -07001315 (skb->len <= dst_mtu(&rt->dst) &&
1316 ip_dont_fragment(sk, &rt->dst)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 df = htons(IP_DF);
1318
Herbert Xu1470ddf2011-03-01 02:36:47 +00001319 if (cork->flags & IPCORK_OPT)
1320 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
1322 if (rt->rt_type == RTN_MULTICAST)
1323 ttl = inet->mc_ttl;
1324 else
Changli Gaod8d1f302010-06-10 23:31:35 -07001325 ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
1327 iph = (struct iphdr *)skb->data;
1328 iph->version = 4;
1329 iph->ihl = 5;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 iph->tos = inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 iph->frag_off = df;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 iph->ttl = ttl;
1333 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +00001334 ip_copy_addrs(iph, fl4);
Eric Dumazeta9915a12012-08-20 07:26:45 +00001335 ip_select_ident(iph, &rt->dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
David S. Miller22f728f2011-05-13 17:21:27 -04001337 if (opt) {
1338 iph->ihl += opt->optlen>>2;
1339 ip_options_build(skb, opt, cork->addr, rt, 0);
1340 }
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001343 skb->mark = sk->sk_mark;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001344 /*
1345 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1346 * on dst refcount
1347 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001348 cork->dst = NULL;
Changli Gaod8d1f302010-06-10 23:31:35 -07001349 skb_dst_set(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
David L Stevens96793b42007-09-17 09:57:33 -07001351 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001352 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001353 skb_transport_header(skb))->type);
1354
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001355 ip_cork_release(cork);
1356out:
1357 return skb;
1358}
1359
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001360int ip_send_skb(struct net *net, struct sk_buff *skb)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001361{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001362 int err;
1363
Herbert Xuc439cb22008-01-11 19:14:00 -08001364 err = ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 if (err) {
1366 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001367 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 if (err)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001369 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373}
1374
David S. Miller77968b72011-05-08 17:12:19 -07001375int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1470ddf2011-03-01 02:36:47 +00001376{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001377 struct sk_buff *skb;
1378
David S. Miller77968b72011-05-08 17:12:19 -07001379 skb = ip_finish_skb(sk, fl4);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001380 if (!skb)
1381 return 0;
1382
1383 /* Netfilter gets whole the not fragmented skb. */
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001384 return ip_send_skb(sock_net(sk), skb);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001385}
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387/*
1388 * Throw away all pending data on the socket.
1389 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001390static void __ip_flush_pending_frames(struct sock *sk,
1391 struct sk_buff_head *queue,
1392 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 struct sk_buff *skb;
1395
Herbert Xu1470ddf2011-03-01 02:36:47 +00001396 while ((skb = __skb_dequeue_tail(queue)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 kfree_skb(skb);
1398
Herbert Xu1470ddf2011-03-01 02:36:47 +00001399 ip_cork_release(cork);
1400}
1401
1402void ip_flush_pending_frames(struct sock *sk)
1403{
David S. Millerbdc712b2011-05-06 15:02:07 -07001404 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001407struct sk_buff *ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001408 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001409 int getfrag(void *from, char *to, int offset,
1410 int len, int odd, struct sk_buff *skb),
1411 void *from, int length, int transhdrlen,
1412 struct ipcm_cookie *ipc, struct rtable **rtp,
1413 unsigned int flags)
1414{
David S. Millerb80d7222011-05-06 15:06:01 -07001415 struct inet_cork cork;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001416 struct sk_buff_head queue;
1417 int err;
1418
1419 if (flags & MSG_PROBE)
1420 return NULL;
1421
1422 __skb_queue_head_init(&queue);
1423
David S. Millerb80d7222011-05-06 15:06:01 -07001424 cork.flags = 0;
1425 cork.addr = 0;
David S. Miller70652722011-05-06 16:01:15 -07001426 cork.opt = NULL;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001427 err = ip_setup_cork(sk, &cork, ipc, rtp);
1428 if (err)
1429 return ERR_PTR(err);
1430
Eric Dumazet5640f762012-09-23 23:04:42 +00001431 err = __ip_append_data(sk, fl4, &queue, &cork,
1432 &current->task_frag, getfrag,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001433 from, length, transhdrlen, flags);
1434 if (err) {
1435 __ip_flush_pending_frames(sk, &queue, &cork);
1436 return ERR_PTR(err);
1437 }
1438
David S. Miller77968b72011-05-08 17:12:19 -07001439 return __ip_make_skb(sk, fl4, &queue, &cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001440}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442/*
1443 * Fetch data from kernel space and fill in checksum if needed.
1444 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001445static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 int len, int odd, struct sk_buff *skb)
1447{
Al Viro50842052006-11-14 21:36:34 -08001448 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1451 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001452 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
1454
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001455/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 * Generic function to send a packet as reply to another packet.
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001457 * Used to send some TCP resets/acks so far.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 *
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001459 * Use a fake percpu inet socket to avoid false sharing and contention.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 */
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001461static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1462 .sk = {
1463 .__sk_common = {
1464 .skc_refcnt = ATOMIC_INIT(1),
1465 },
1466 .sk_wmem_alloc = ATOMIC_INIT(1),
1467 .sk_allocation = GFP_ATOMIC,
1468 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
1469 },
Eric Dumazet0980e562012-07-20 22:28:51 +00001470 .pmtudisc = IP_PMTUDISC_WANT,
1471 .uc_ttl = -1,
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001472};
1473
1474void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
David S. Miller70e73412012-06-28 03:21:41 -07001475 __be32 saddr, const struct ip_reply_arg *arg,
1476 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001478 struct ip_options_data replyopts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 struct ipcm_cookie ipc;
David S. Miller77968b72011-05-08 17:12:19 -07001480 struct flowi4 fl4;
Eric Dumazet511c3f92009-06-02 05:14:27 +00001481 struct rtable *rt = skb_rtable(skb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001482 struct sk_buff *nskb;
1483 struct sock *sk;
1484 struct inet_sock *inet;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001486 if (ip_options_echo(&replyopts.opt.opt, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return;
1488
David S. Miller0a5ebb82011-05-09 13:22:43 -07001489 ipc.addr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +00001491 ipc.tx_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001493 if (replyopts.opt.opt.optlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 ipc.opt = &replyopts.opt;
1495
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001496 if (replyopts.opt.opt.srr)
1497 daddr = replyopts.opt.opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 }
1499
David S. Miller77968b72011-05-08 17:12:19 -07001500 flowi4_init_output(&fl4, arg->bound_dev_if, 0,
Eric Dumazet66b13d92011-10-24 03:06:21 -04001501 RT_TOS(arg->tos),
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001502 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
David S. Miller77968b72011-05-08 17:12:19 -07001503 ip_reply_arg_flowi_flags(arg),
David S. Miller70e73412012-06-28 03:21:41 -07001504 daddr, saddr,
David S. Miller77968b72011-05-08 17:12:19 -07001505 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1506 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001507 rt = ip_route_output_key(net, &fl4);
David S. Miller77968b72011-05-08 17:12:19 -07001508 if (IS_ERR(rt))
1509 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001511 inet = &get_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Eric Dumazet66b13d92011-10-24 03:06:21 -04001513 inet->tos = arg->tos;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001514 sk = &inet->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001516 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001517 sk->sk_bound_dev_if = arg->bound_dev_if;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001518 sock_net_set(sk, net);
1519 __skb_queue_head_init(&sk->sk_write_queue);
1520 sk->sk_sndbuf = sysctl_wmem_default;
David S. Millerf5fca602011-05-08 17:24:10 -07001521 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
Eric Dumazet2e77d892008-11-24 15:52:46 -08001522 &ipc, &rt, MSG_DONTWAIT);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001523 nskb = skb_peek(&sk->sk_write_queue);
1524 if (nskb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 if (arg->csumoffset >= 0)
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001526 *((__sum16 *)skb_transport_header(nskb) +
1527 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001528 arg->csum));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001529 nskb->ip_summed = CHECKSUM_NONE;
Eric Dumazet3a7c3842012-08-09 13:56:06 +00001530 skb_orphan(nskb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001531 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
David S. Miller77968b72011-05-08 17:12:19 -07001532 ip_push_pending_frames(sk, &fl4);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 }
1534
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001535 put_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
1537 ip_rt_put(rt);
1538}
1539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540void __init ip_init(void)
1541{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 ip_rt_init();
1543 inet_initpeers();
1544
1545#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1546 igmp_mc_proc_init();
1547#endif
1548}