blob: 1cbeba5edff90fa1ac891d4dd23cfb65464878a4 [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
Eric Dumazetaad88722014-04-15 13:47:15 -0400104int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
Herbert Xuc439cb22008-01-11 19:14:00 -0800105{
106 int err;
107
108 err = __ip_local_out(skb);
109 if (likely(err == 1))
Eric Dumazetaad88722014-04-15 13:47:15 -0400110 err = dst_output_sk(sk, skb);
Herbert Xuc439cb22008-01-11 19:14:00 -0800111
112 return err;
113}
Eric Dumazetaad88722014-04-15 13:47:15 -0400114EXPORT_SYMBOL_GPL(ip_local_out_sk);
Herbert Xuc439cb22008-01-11 19:14:00 -0800115
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;
Ansis Atteka703133d2013-09-18 15:29:53 -0700151 ip_select_ident(skb, &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
Patrick McHardy861d0482007-10-15 01:48:39 -0700214static int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800216#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
217 /* Policy lookup after SNAT yielded a new policy */
Eric Dumazetadf30902009-06-02 05:19:30 +0000218 if (skb_dst(skb)->xfrm != NULL) {
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800219 IPCB(skb)->flags |= IPSKB_REROUTED;
220 return dst_output(skb);
221 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800222#endif
John Heffner628a5c52007-04-20 15:53:27 -0700223 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800224 return ip_fragment(skb, ip_finish_output2);
225 else
226 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
228
Eric Dumazetaad88722014-04-15 13:47:15 -0400229int ip_mc_output(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Eric Dumazet511c3f92009-06-02 05:14:27 +0000231 struct rtable *rt = skb_rtable(skb);
Changli Gaod8d1f302010-06-10 23:31:35 -0700232 struct net_device *dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
234 /*
235 * If the indicated interface is up and running, send the packet.
236 */
Neil Hormanedf391f2009-04-27 02:45:02 -0700237 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239 skb->dev = dev;
240 skb->protocol = htons(ETH_P_IP);
241
242 /*
243 * Multicasts are looped back for other local users
244 */
245
246 if (rt->rt_flags&RTCF_MULTICAST) {
Octavian Purdila7ad68482010-01-06 20:37:01 -0800247 if (sk_mc_loop(sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248#ifdef CONFIG_IP_MROUTE
249 /* Small optimization: do not loopback not local frames,
250 which returned after forwarding; they will be dropped
251 by ip_mr_input in any case.
252 Note, that local frames are looped back to be delivered
253 to local recipients.
254
255 This check is duplicated in ip_mr_input at the moment.
256 */
Joe Perches9d4fb272009-11-23 10:41:23 -0800257 &&
258 ((rt->rt_flags & RTCF_LOCAL) ||
259 !(IPCB(skb)->flags & IPSKB_FORWARDED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260#endif
Joe Perches9d4fb272009-11-23 10:41:23 -0800261 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
263 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100264 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
265 newskb, NULL, newskb->dev,
Michel Machado95603e22012-06-12 10:16:35 +0000266 dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 }
268
269 /* Multicasts with ttl 0 must not go beyond the host */
270
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700271 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 kfree_skb(skb);
273 return 0;
274 }
275 }
276
277 if (rt->rt_flags&RTCF_BROADCAST) {
278 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
279 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100280 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
Michel Machado95603e22012-06-12 10:16:35 +0000281 NULL, newskb->dev, dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
283
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100284 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
285 skb->dev, ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800286 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288
Eric Dumazetaad88722014-04-15 13:47:15 -0400289int ip_output(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Eric Dumazetadf30902009-06-02 05:19:30 +0000291 struct net_device *dev = skb_dst(skb)->dev;
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800292
Neil Hormanedf391f2009-04-27 02:45:02 -0700293 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800295 skb->dev = dev;
296 skb->protocol = htons(ETH_P_IP);
297
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100298 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900299 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800300 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
Eric Dumazet84f93072011-11-30 19:00:53 +0000303/*
304 * copy saddr and daddr, possibly using 64bit load/stores
305 * Equivalent to :
306 * iph->saddr = fl4->saddr;
307 * iph->daddr = fl4->daddr;
308 */
309static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
310{
311 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
312 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
313 memcpy(&iph->saddr, &fl4->saddr,
314 sizeof(fl4->saddr) + sizeof(fl4->daddr));
315}
316
Eric Dumazetb0270e92014-04-15 12:58:34 -0400317/* Note: skb->sk can be different from sk, in case of tunnels */
318int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000321 struct ip_options_rcu *inet_opt;
David S. Millerb57ae012011-05-06 16:24:06 -0700322 struct flowi4 *fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 struct rtable *rt;
324 struct iphdr *iph;
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000325 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
327 /* Skip all of this if the packet is already routed,
328 * f.e. by something like SCTP.
329 */
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000330 rcu_read_lock();
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000331 inet_opt = rcu_dereference(inet->inet_opt);
David S. Millerea4fc0d2011-05-06 22:30:20 -0700332 fl4 = &fl->u.ip4;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000333 rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 if (rt != NULL)
335 goto packet_routed;
336
337 /* Make sure we can route this packet. */
338 rt = (struct rtable *)__sk_dst_check(sk, 0);
339 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700340 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342 /* Use correct destination address if we have options. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000343 daddr = inet->inet_daddr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000344 if (inet_opt && inet_opt->opt.srr)
345 daddr = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
David S. Miller78fbfd82011-03-12 00:00:52 -0500347 /* If this fails, retransmit mechanism of transport layer will
348 * keep trying until route appears or the connection times
349 * itself out.
350 */
David S. Millerb57ae012011-05-06 16:24:06 -0700351 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500352 daddr, inet->inet_saddr,
353 inet->inet_dport,
354 inet->inet_sport,
355 sk->sk_protocol,
356 RT_CONN_FLAGS(sk),
357 sk->sk_bound_dev_if);
358 if (IS_ERR(rt))
359 goto no_route;
Changli Gaod8d1f302010-06-10 23:31:35 -0700360 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 }
Changli Gaod8d1f302010-06-10 23:31:35 -0700362 skb_dst_set_noref(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364packet_routed:
Julian Anastasov155e8332012-10-08 11:41:18 +0000365 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 goto no_route;
367
368 /* OK, we know where to send it, allocate and build IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000369 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300370 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700371 iph = ip_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800372 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
Changli Gaod8d1f302010-06-10 23:31:35 -0700373 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 iph->frag_off = htons(IP_DF);
375 else
376 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700377 iph->ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +0000379 ip_copy_addrs(iph, fl4);
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /* Transport layer set skb->h.foo itself. */
382
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000383 if (inet_opt && inet_opt->opt.optlen) {
384 iph->ihl += inet_opt->opt.optlen >> 2;
385 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 }
387
Ansis Atteka703133d2013-09-18 15:29:53 -0700388 ip_select_ident_more(skb, &rt->dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700389 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
Eric Dumazetb0270e92014-04-15 12:58:34 -0400391 /* TODO : should we use skb->sk here instead of sk ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800393 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000395 res = ip_local_out(skb);
396 rcu_read_unlock();
397 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399no_route:
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000400 rcu_read_unlock();
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700401 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 kfree_skb(skb);
403 return -EHOSTUNREACH;
404}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000405EXPORT_SYMBOL(ip_queue_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407
408static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
409{
410 to->pkt_type = from->pkt_type;
411 to->priority = from->priority;
412 to->protocol = from->protocol;
Eric Dumazetadf30902009-06-02 05:19:30 +0000413 skb_dst_drop(to);
Eric Dumazetfe76cda2010-07-01 23:48:22 +0000414 skb_dst_copy(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800416 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418 /* Copy the flags to each fragment. */
419 IPCB(to)->flags = IPCB(from)->flags;
420
421#ifdef CONFIG_NET_SCHED
422 to->tc_index = from->tc_index;
423#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700424 nf_copy(to, from);
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300425#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
426 to->ipvs_property = from->ipvs_property;
427#endif
James Morris984bc162006-06-09 00:29:17 -0700428 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
430
431/*
432 * This IP datagram is too large to be sent in one piece. Break it up into
433 * smaller pieces (each of size equal to IP header plus
434 * a block of the data of the original IP data part) that will yet fit in a
435 * single device frame, and queue such a frame for sending.
436 */
437
Jianjun Kongd93191002008-11-03 00:23:42 -0800438int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
440 struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 int ptr;
442 struct net_device *dev;
443 struct sk_buff *skb2;
Changli Gaoc893b802010-07-31 13:25:08 +0000444 unsigned int mtu, hlen, left, len, ll_rs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800446 __be16 not_last_frag;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000447 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 int err = 0;
449
Changli Gaod8d1f302010-06-10 23:31:35 -0700450 dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 /*
453 * Point into the IP datagram header.
454 */
455
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700456 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Hannes Frederic Sowa69647ce2014-02-26 01:20:41 +0100458 mtu = ip_skb_dst_mtu(skb);
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200459 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
460 (IPCB(skb)->frag_max_size &&
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100461 IPCB(skb)->frag_max_size > mtu))) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700462 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100464 htonl(mtu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 kfree_skb(skb);
466 return -EMSGSIZE;
467 }
468
469 /*
470 * Setup starting values.
471 */
472
473 hlen = iph->ihl * 4;
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100474 mtu = mtu - hlen; /* Size of data space */
Bart De Schuymer6c79bf02010-04-20 16:22:01 +0200475#ifdef CONFIG_BRIDGE_NETFILTER
476 if (skb->nf_bridge)
477 mtu -= nf_bridge_mtu_reduction(skb);
478#endif
Herbert Xu89cee8b2005-12-13 23:14:27 -0800479 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 /* When frag_list is given, use it. First, check its validity:
482 * some transformers could create wrong frag_list or break existing
483 * one, it is not prohibited. In this case fall back to copying.
484 *
485 * LATER: this step can be merged to real generation of fragments,
486 * we can switch to copy when see the first bad fragment.
487 */
David S. Miller21dc3302010-08-23 00:13:46 -0700488 if (skb_has_frag_list(skb)) {
Eric Dumazet3d130082010-09-21 08:47:45 +0000489 struct sk_buff *frag, *frag2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 int first_len = skb_pagelen(skb);
491
492 if (first_len - hlen > mtu ||
493 ((first_len - hlen) & 7) ||
Paul Gortmaker56f8a752011-06-21 20:33:34 -0700494 ip_is_fragment(iph) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 skb_cloned(skb))
496 goto slow_path;
497
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700498 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 /* Correct geometry. */
500 if (frag->len > mtu ||
501 ((frag->len & 7) && frag->next) ||
502 skb_headroom(frag) < hlen)
Eric Dumazet3d130082010-09-21 08:47:45 +0000503 goto slow_path_clean;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 /* Partially cloned skb? */
506 if (skb_shared(frag))
Eric Dumazet3d130082010-09-21 08:47:45 +0000507 goto slow_path_clean;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700508
509 BUG_ON(frag->sk);
510 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700511 frag->sk = skb->sk;
512 frag->destructor = sock_wfree;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700513 }
Eric Dumazet3d130082010-09-21 08:47:45 +0000514 skb->truesize -= frag->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516
517 /* Everything is OK. Generate! */
518
519 err = 0;
520 offset = 0;
521 frag = skb_shinfo(skb)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700522 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 skb->data_len = first_len - skb_headlen(skb);
524 skb->len = first_len;
525 iph->tot_len = htons(first_len);
526 iph->frag_off = htons(IP_MF);
527 ip_send_check(iph);
528
529 for (;;) {
530 /* Prepare header of the next frame,
531 * before previous one went down. */
532 if (frag) {
533 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300534 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700535 __skb_push(frag, hlen);
536 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700537 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700538 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 iph->tot_len = htons(frag->len);
540 ip_copy_metadata(frag, skb);
541 if (offset == 0)
542 ip_options_fragment(frag);
543 offset += skb->len - hlen;
544 iph->frag_off = htons(offset>>3);
545 if (frag->next != NULL)
546 iph->frag_off |= htons(IP_MF);
547 /* Ready, complete checksum */
548 ip_send_check(iph);
549 }
550
551 err = output(skb);
552
Wei Dongdafee492006-08-02 13:41:21 -0700553 if (!err)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700554 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 if (err || !frag)
556 break;
557
558 skb = frag;
559 frag = skb->next;
560 skb->next = NULL;
561 }
562
563 if (err == 0) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700564 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 return 0;
566 }
567
568 while (frag) {
569 skb = frag->next;
570 kfree_skb(frag);
571 frag = skb;
572 }
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700573 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 return err;
Eric Dumazet3d130082010-09-21 08:47:45 +0000575
576slow_path_clean:
577 skb_walk_frags(skb, frag2) {
578 if (frag2 == frag)
579 break;
580 frag2->sk = NULL;
581 frag2->destructor = NULL;
582 skb->truesize += frag2->truesize;
583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 }
585
586slow_path:
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000587 /* for offloaded checksums cleanup checksum before fragmentation */
588 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
589 goto fail;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000590 iph = ip_hdr(skb);
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 left = skb->len - hlen; /* Space per frame */
George Kadianakis49085bd2010-07-06 11:44:12 +0000593 ptr = hlen; /* Where to start from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700596 * we need to make room for the encapsulating header
597 */
Changli Gaoc893b802010-07-31 13:25:08 +0000598 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 /*
601 * Fragment the datagram.
602 */
603
604 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
605 not_last_frag = iph->frag_off & htons(IP_MF);
606
607 /*
608 * Keep copying data until we run out.
609 */
610
Stephen Hemminger132adf52007-03-08 20:44:43 -0800611 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 len = left;
613 /* IF: it doesn't fit, use 'mtu' - the data space left */
614 if (len > mtu)
615 len = mtu;
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300616 /* IF: we are not sending up to and including the packet end
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 then align the next start on an eight byte boundary */
618 if (len < left) {
619 len &= ~7;
620 }
621 /*
622 * Allocate buffer.
623 */
624
625 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700626 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 err = -ENOMEM;
628 goto fail;
629 }
630
631 /*
632 * Set up data on packet
633 */
634
635 ip_copy_metadata(skb2, skb);
636 skb_reserve(skb2, ll_rs);
637 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700638 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700639 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
641 /*
642 * Charge the memory for the fragment to any owner
643 * it might possess
644 */
645
646 if (skb->sk)
647 skb_set_owner_w(skb2, skb->sk);
648
649 /*
650 * Copy the packet header into the new buffer.
651 */
652
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300653 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655 /*
656 * Copy a block of the IP datagram.
657 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300658 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 BUG();
660 left -= len;
661
662 /*
663 * Fill in the new header fields.
664 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700665 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 iph->frag_off = htons((offset >> 3));
667
668 /* ANK: dirty, but effective trick. Upgrade options only if
669 * the segment to be fragmented was THE FIRST (otherwise,
670 * options are already fixed) and make it ONCE
671 * on the initial skb, so that all the following fragments
672 * will inherit fixed options.
673 */
674 if (offset == 0)
675 ip_options_fragment(skb);
676
677 /*
678 * Added AC : If we are fragmenting a fragment that's not the
679 * last fragment then keep MF on each bit
680 */
681 if (left > 0 || not_last_frag)
682 iph->frag_off |= htons(IP_MF);
683 ptr += len;
684 offset += len;
685
686 /*
687 * Put this fragment into the sending queue.
688 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 iph->tot_len = htons(len + hlen);
690
691 ip_send_check(iph);
692
693 err = output(skb2);
694 if (err)
695 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700696
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700697 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000699 consume_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700700 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 return err;
702
703fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900704 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700705 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return err;
707}
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700708EXPORT_SYMBOL(ip_fragment);
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710int
711ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
712{
713 struct iovec *iov = from;
714
Patrick McHardy84fa7932006-08-29 16:44:56 -0700715 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
717 return -EFAULT;
718 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800719 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
721 return -EFAULT;
722 skb->csum = csum_block_add(skb->csum, csum, odd);
723 }
724 return 0;
725}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000726EXPORT_SYMBOL(ip_generic_getfrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Al Viro44bb9362006-11-14 21:36:14 -0800728static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729csum_page(struct page *page, int offset, int copy)
730{
731 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800732 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 kaddr = kmap(page);
734 csum = csum_partial(kaddr + offset, copy, 0);
735 kunmap(page);
736 return csum;
737}
738
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800739static inline int ip_ufo_append_data(struct sock *sk,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000740 struct sk_buff_head *queue,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700741 int getfrag(void *from, char *to, int offset, int len,
742 int odd, struct sk_buff *skb),
743 void *from, int length, int hh_len, int fragheaderlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000744 int transhdrlen, int maxfraglen, unsigned int flags)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700745{
746 struct sk_buff *skb;
747 int err;
748
749 /* There is support for UDP fragmentation offload by network
750 * device, so create one single skb packet containing complete
751 * udp datagram
752 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000753 if ((skb = skb_peek_tail(queue)) == NULL) {
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700754 skb = sock_alloc_send_skb(sk,
755 hh_len + fragheaderlen + transhdrlen + 20,
756 (flags & MSG_DONTWAIT), &err);
757
758 if (skb == NULL)
759 return err;
760
761 /* reserve space for Hardware header */
762 skb_reserve(skb, hh_len);
763
764 /* create space for UDP/IP header */
Jianjun Kongd93191002008-11-03 00:23:42 -0800765 skb_put(skb, fragheaderlen + transhdrlen);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700766
767 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700768 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700769
770 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700771 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700772
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700773 skb->csum = 0;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700774
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200775
Herbert Xu1470ddf2011-03-01 02:36:47 +0000776 __skb_queue_tail(queue, skb);
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200777 } else if (skb_is_gso(skb)) {
778 goto append;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700779 }
Kostya Bbe9164e2008-04-29 22:36:30 -0700780
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200781 skb->ip_summed = CHECKSUM_PARTIAL;
782 /* specify the length of each IP datagram fragment */
783 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
784 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
785
786append:
Kostya Bbe9164e2008-04-29 22:36:30 -0700787 return skb_append_datato_frags(sk, skb, getfrag, from,
788 (length - transhdrlen));
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700789}
790
David S. Millerf5fca602011-05-08 17:24:10 -0700791static int __ip_append_data(struct sock *sk,
792 struct flowi4 *fl4,
793 struct sk_buff_head *queue,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000794 struct inet_cork *cork,
Eric Dumazet5640f762012-09-23 23:04:42 +0000795 struct page_frag *pfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000796 int getfrag(void *from, char *to, int offset,
797 int len, int odd, struct sk_buff *skb),
798 void *from, int length, int transhdrlen,
799 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
801 struct inet_sock *inet = inet_sk(sk);
802 struct sk_buff *skb;
803
Herbert Xu07df5292011-03-01 23:00:58 -0800804 struct ip_options *opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 int hh_len;
806 int exthdrlen;
807 int mtu;
808 int copy;
809 int err;
810 int offset = 0;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100811 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 int csummode = CHECKSUM_NONE;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000813 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Steffen Klassert96d73032011-06-05 20:48:47 +0000815 skb = skb_peek_tail(queue);
816
817 exthdrlen = !skb ? rt->dst.header_len : 0;
Herbert Xu07df5292011-03-01 23:00:58 -0800818 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Changli Gaod8d1f302010-06-10 23:31:35 -0700820 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
822 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
823 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +0100824 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100826 if (cork->length + length > maxnonfragsize - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -0700827 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +0100828 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return -EMSGSIZE;
830 }
831
832 /*
833 * transhdrlen > 0 means that this is the first fragment and we wish
834 * it won't be fragmented in the future.
835 */
836 if (transhdrlen &&
837 length + fragheaderlen <= mtu &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700838 rt->dst.dev->features & NETIF_F_V4_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700840 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Herbert Xu1470ddf2011-03-01 02:36:47 +0000842 cork->length += length;
Herbert Xu26cde9f2010-06-15 01:52:25 +0000843 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
Kostya Bbe9164e2008-04-29 22:36:30 -0700844 (sk->sk_protocol == IPPROTO_UDP) &&
Steffen Klassertc1460662011-06-29 23:19:32 +0000845 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
Herbert Xu1470ddf2011-03-01 02:36:47 +0000846 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
847 hh_len, fragheaderlen, transhdrlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000848 maxfraglen, flags);
Patrick McHardybaa829d2006-03-12 20:35:12 -0800849 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700850 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700851 return 0;
852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 /* So, what's going on in the loop below?
855 *
856 * We use calculated fragment length to generate chained skb,
857 * each of segments is IP fragment ready for sending to network after
858 * adding appropriate IP header.
859 */
860
Herbert Xu26cde9f2010-06-15 01:52:25 +0000861 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 goto alloc_new_skb;
863
864 while (length > 0) {
865 /* Check if the remaining data fits into current packet. */
866 copy = mtu - skb->len;
867 if (copy < length)
868 copy = maxfraglen - skb->len;
869 if (copy <= 0) {
870 char *data;
871 unsigned int datalen;
872 unsigned int fraglen;
873 unsigned int fraggap;
874 unsigned int alloclen;
875 struct sk_buff *skb_prev;
876alloc_new_skb:
877 skb_prev = skb;
878 if (skb_prev)
879 fraggap = skb_prev->len - maxfraglen;
880 else
881 fraggap = 0;
882
883 /*
884 * If remaining data exceeds the mtu,
885 * we know we need more fragment(s).
886 */
887 datalen = length + fraggap;
888 if (datalen > mtu - fragheaderlen)
889 datalen = maxfraglen - fragheaderlen;
890 fraglen = datalen + fragheaderlen;
891
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900892 if ((flags & MSG_MORE) &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700893 !(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 alloclen = mtu;
895 else
Eric Dumazet59104f02010-09-20 20:16:27 +0000896 alloclen = fraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Steffen Klassert353e5c92011-06-22 01:05:37 +0000898 alloclen += exthdrlen;
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 /* The last fragment gets additional space at tail.
901 * Note, with MSG_MORE we overallocate on fragments,
902 * because we have no idea what fragment will be
903 * the last.
904 */
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000905 if (datalen == length + fraggap)
Changli Gaod8d1f302010-06-10 23:31:35 -0700906 alloclen += rt->dst.trailer_len;
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900909 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 alloclen + hh_len + 15,
911 (flags & MSG_DONTWAIT), &err);
912 } else {
913 skb = NULL;
914 if (atomic_read(&sk->sk_wmem_alloc) <=
915 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900916 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 alloclen + hh_len + 15, 1,
918 sk->sk_allocation);
919 if (unlikely(skb == NULL))
920 err = -ENOBUFS;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000921 else
922 /* only the initial fragment is
923 time stamped */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000924 cork->tx_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 }
926 if (skb == NULL)
927 goto error;
928
929 /*
930 * Fill in the control structures
931 */
932 skb->ip_summed = csummode;
933 skb->csum = 0;
934 skb_reserve(skb, hh_len);
Herbert Xu1470ddf2011-03-01 02:36:47 +0000935 skb_shinfo(skb)->tx_flags = cork->tx_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 /*
938 * Find where to start putting bytes.
939 */
Steffen Klassert353e5c92011-06-22 01:05:37 +0000940 data = skb_put(skb, fraglen + exthdrlen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -0300941 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700942 skb->transport_header = (skb->network_header +
943 fragheaderlen);
Steffen Klassert353e5c92011-06-22 01:05:37 +0000944 data += fragheaderlen + exthdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 if (fraggap) {
947 skb->csum = skb_copy_and_csum_bits(
948 skb_prev, maxfraglen,
949 data + transhdrlen, fraggap, 0);
950 skb_prev->csum = csum_sub(skb_prev->csum,
951 skb->csum);
952 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700953 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 }
955
956 copy = datalen - transhdrlen - fraggap;
957 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
958 err = -EFAULT;
959 kfree_skb(skb);
960 goto error;
961 }
962
963 offset += copy;
964 length -= datalen - fraggap;
965 transhdrlen = 0;
966 exthdrlen = 0;
967 csummode = CHECKSUM_NONE;
968
969 /*
970 * Put the packet on the pending queue.
971 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000972 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 continue;
974 }
975
976 if (copy > length)
977 copy = length;
978
Changli Gaod8d1f302010-06-10 23:31:35 -0700979 if (!(rt->dst.dev->features&NETIF_F_SG)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 unsigned int off;
981
982 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900983 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 offset, copy, off, skb) < 0) {
985 __skb_trim(skb, off);
986 err = -EFAULT;
987 goto error;
988 }
989 } else {
990 int i = skb_shinfo(skb)->nr_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Eric Dumazet5640f762012-09-23 23:04:42 +0000992 err = -ENOMEM;
993 if (!sk_page_frag_refill(sk, pfrag))
994 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Eric Dumazet5640f762012-09-23 23:04:42 +0000996 if (!skb_can_coalesce(skb, i, pfrag->page,
997 pfrag->offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 err = -EMSGSIZE;
Eric Dumazet5640f762012-09-23 23:04:42 +0000999 if (i == MAX_SKB_FRAGS)
1000 goto error;
1001
1002 __skb_fill_page_desc(skb, i, pfrag->page,
1003 pfrag->offset, 0);
1004 skb_shinfo(skb)->nr_frags = ++i;
1005 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001007 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1008 if (getfrag(from,
1009 page_address(pfrag->page) + pfrag->offset,
1010 offset, copy, skb->len, skb) < 0)
1011 goto error_efault;
1012
1013 pfrag->offset += copy;
1014 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 skb->len += copy;
1016 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001017 skb->truesize += copy;
1018 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
1020 offset += copy;
1021 length -= copy;
1022 }
1023
1024 return 0;
1025
Eric Dumazet5640f762012-09-23 23:04:42 +00001026error_efault:
1027 err = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028error:
Herbert Xu1470ddf2011-03-01 02:36:47 +00001029 cork->length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001030 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001031 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
1033
Herbert Xu1470ddf2011-03-01 02:36:47 +00001034static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1035 struct ipcm_cookie *ipc, struct rtable **rtp)
1036{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001037 struct ip_options_rcu *opt;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001038 struct rtable *rt;
1039
1040 /*
1041 * setup for corking.
1042 */
1043 opt = ipc->opt;
1044 if (opt) {
1045 if (cork->opt == NULL) {
1046 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1047 sk->sk_allocation);
1048 if (unlikely(cork->opt == NULL))
1049 return -ENOBUFS;
1050 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001051 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001052 cork->flags |= IPCORK_OPT;
1053 cork->addr = ipc->addr;
1054 }
1055 rt = *rtp;
1056 if (unlikely(!rt))
1057 return -EFAULT;
1058 /*
1059 * We steal reference to this route, caller should not release it
1060 */
1061 *rtp = NULL;
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001062 cork->fragsize = ip_sk_use_pmtu(sk) ?
1063 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001064 cork->dst = &rt->dst;
1065 cork->length = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001066 cork->ttl = ipc->ttl;
1067 cork->tos = ipc->tos;
1068 cork->priority = ipc->priority;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001069 cork->tx_flags = ipc->tx_flags;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001070
1071 return 0;
1072}
1073
1074/*
1075 * ip_append_data() and ip_append_page() can make one large IP datagram
1076 * from many pieces of data. Each pieces will be holded on the socket
1077 * until ip_push_pending_frames() is called. Each piece can be a page
1078 * or non-page data.
1079 *
1080 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1081 * this interface potentially.
1082 *
1083 * LATER: length must be adjusted by pad at tail, when it is required.
1084 */
David S. Millerf5fca602011-05-08 17:24:10 -07001085int ip_append_data(struct sock *sk, struct flowi4 *fl4,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001086 int getfrag(void *from, char *to, int offset, int len,
1087 int odd, struct sk_buff *skb),
1088 void *from, int length, int transhdrlen,
1089 struct ipcm_cookie *ipc, struct rtable **rtp,
1090 unsigned int flags)
1091{
1092 struct inet_sock *inet = inet_sk(sk);
1093 int err;
1094
1095 if (flags&MSG_PROBE)
1096 return 0;
1097
1098 if (skb_queue_empty(&sk->sk_write_queue)) {
David S. Millerbdc712b2011-05-06 15:02:07 -07001099 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001100 if (err)
1101 return err;
1102 } else {
1103 transhdrlen = 0;
1104 }
1105
Eric Dumazet5640f762012-09-23 23:04:42 +00001106 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1107 sk_page_frag(sk), getfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001108 from, length, transhdrlen, flags);
1109}
1110
David S. Millerf5fca602011-05-08 17:24:10 -07001111ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 int offset, size_t size, int flags)
1113{
1114 struct inet_sock *inet = inet_sk(sk);
1115 struct sk_buff *skb;
1116 struct rtable *rt;
1117 struct ip_options *opt = NULL;
David S. Millerbdc712b2011-05-06 15:02:07 -07001118 struct inet_cork *cork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 int hh_len;
1120 int mtu;
1121 int len;
1122 int err;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001123 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 if (inet->hdrincl)
1126 return -EPERM;
1127
1128 if (flags&MSG_PROBE)
1129 return 0;
1130
1131 if (skb_queue_empty(&sk->sk_write_queue))
1132 return -EINVAL;
1133
David S. Millerbdc712b2011-05-06 15:02:07 -07001134 cork = &inet->cork.base;
1135 rt = (struct rtable *)cork->dst;
1136 if (cork->flags & IPCORK_OPT)
1137 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Changli Gaod8d1f302010-06-10 23:31:35 -07001139 if (!(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return -EOPNOTSUPP;
1141
Changli Gaod8d1f302010-06-10 23:31:35 -07001142 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
David S. Millerbdc712b2011-05-06 15:02:07 -07001143 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1146 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +01001147 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001149 if (cork->length + size > maxnonfragsize - fragheaderlen) {
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +01001150 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1151 mtu - (opt ? opt->optlen : 0));
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 */
Hannes Frederic Sowa1b346572014-02-26 01:20:42 +01001308 skb->local_df = ip_sk_local_df(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 /* DF bit is set when we want to see DF on outgoing frames.
1311 * If local_df is set too, we still allow to fragment this frame
1312 * locally. */
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001313 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1314 inet->pmtudisc == IP_PMTUDISC_PROBE ||
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
Francesco Fuscoaa661582013-09-24 15:43:09 +02001322 if (cork->ttl != 0)
1323 ttl = cork->ttl;
1324 else if (rt->rt_type == RTN_MULTICAST)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 ttl = inet->mc_ttl;
1326 else
Changli Gaod8d1f302010-06-10 23:31:35 -07001327 ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Ansis Atteka749154a2013-09-18 15:29:52 -07001329 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 iph->version = 4;
1331 iph->ihl = 5;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001332 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 iph->frag_off = df;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 iph->ttl = ttl;
1335 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +00001336 ip_copy_addrs(iph, fl4);
Ansis Atteka703133d2013-09-18 15:29:53 -07001337 ip_select_ident(skb, &rt->dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
David S. Miller22f728f2011-05-13 17:21:27 -04001339 if (opt) {
1340 iph->ihl += opt->optlen>>2;
1341 ip_options_build(skb, opt, cork->addr, rt, 0);
1342 }
1343
Francesco Fuscoaa661582013-09-24 15:43:09 +02001344 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001345 skb->mark = sk->sk_mark;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001346 /*
1347 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1348 * on dst refcount
1349 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001350 cork->dst = NULL;
Changli Gaod8d1f302010-06-10 23:31:35 -07001351 skb_dst_set(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
David L Stevens96793b42007-09-17 09:57:33 -07001353 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001354 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001355 skb_transport_header(skb))->type);
1356
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001357 ip_cork_release(cork);
1358out:
1359 return skb;
1360}
1361
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001362int ip_send_skb(struct net *net, struct sk_buff *skb)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001363{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001364 int err;
1365
Herbert Xuc439cb22008-01-11 19:14:00 -08001366 err = ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (err) {
1368 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001369 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (err)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001371 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 }
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
David S. Miller77968b72011-05-08 17:12:19 -07001377int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1470ddf2011-03-01 02:36:47 +00001378{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001379 struct sk_buff *skb;
1380
David S. Miller77968b72011-05-08 17:12:19 -07001381 skb = ip_finish_skb(sk, fl4);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001382 if (!skb)
1383 return 0;
1384
1385 /* Netfilter gets whole the not fragmented skb. */
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001386 return ip_send_skb(sock_net(sk), skb);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001387}
1388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389/*
1390 * Throw away all pending data on the socket.
1391 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001392static void __ip_flush_pending_frames(struct sock *sk,
1393 struct sk_buff_head *queue,
1394 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 struct sk_buff *skb;
1397
Herbert Xu1470ddf2011-03-01 02:36:47 +00001398 while ((skb = __skb_dequeue_tail(queue)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 kfree_skb(skb);
1400
Herbert Xu1470ddf2011-03-01 02:36:47 +00001401 ip_cork_release(cork);
1402}
1403
1404void ip_flush_pending_frames(struct sock *sk)
1405{
David S. Millerbdc712b2011-05-06 15:02:07 -07001406 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407}
1408
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001409struct sk_buff *ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001410 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001411 int getfrag(void *from, char *to, int offset,
1412 int len, int odd, struct sk_buff *skb),
1413 void *from, int length, int transhdrlen,
1414 struct ipcm_cookie *ipc, struct rtable **rtp,
1415 unsigned int flags)
1416{
David S. Millerb80d7222011-05-06 15:06:01 -07001417 struct inet_cork cork;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001418 struct sk_buff_head queue;
1419 int err;
1420
1421 if (flags & MSG_PROBE)
1422 return NULL;
1423
1424 __skb_queue_head_init(&queue);
1425
David S. Millerb80d7222011-05-06 15:06:01 -07001426 cork.flags = 0;
1427 cork.addr = 0;
David S. Miller70652722011-05-06 16:01:15 -07001428 cork.opt = NULL;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001429 err = ip_setup_cork(sk, &cork, ipc, rtp);
1430 if (err)
1431 return ERR_PTR(err);
1432
Eric Dumazet5640f762012-09-23 23:04:42 +00001433 err = __ip_append_data(sk, fl4, &queue, &cork,
1434 &current->task_frag, getfrag,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001435 from, length, transhdrlen, flags);
1436 if (err) {
1437 __ip_flush_pending_frames(sk, &queue, &cork);
1438 return ERR_PTR(err);
1439 }
1440
David S. Miller77968b72011-05-08 17:12:19 -07001441 return __ip_make_skb(sk, fl4, &queue, &cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001442}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444/*
1445 * Fetch data from kernel space and fill in checksum if needed.
1446 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001447static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 int len, int odd, struct sk_buff *skb)
1449{
Al Viro50842052006-11-14 21:36:34 -08001450 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
1452 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1453 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001454 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455}
1456
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001457/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * Generic function to send a packet as reply to another packet.
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001459 * Used to send some TCP resets/acks so far.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 *
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001461 * Use a fake percpu inet socket to avoid false sharing and contention.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 */
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001463static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1464 .sk = {
1465 .__sk_common = {
1466 .skc_refcnt = ATOMIC_INIT(1),
1467 },
1468 .sk_wmem_alloc = ATOMIC_INIT(1),
1469 .sk_allocation = GFP_ATOMIC,
1470 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
1471 },
Eric Dumazet0980e562012-07-20 22:28:51 +00001472 .pmtudisc = IP_PMTUDISC_WANT,
1473 .uc_ttl = -1,
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001474};
1475
1476void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
David S. Miller70e73412012-06-28 03:21:41 -07001477 __be32 saddr, const struct ip_reply_arg *arg,
1478 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001480 struct ip_options_data replyopts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 struct ipcm_cookie ipc;
David S. Miller77968b72011-05-08 17:12:19 -07001482 struct flowi4 fl4;
Eric Dumazet511c3f92009-06-02 05:14:27 +00001483 struct rtable *rt = skb_rtable(skb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001484 struct sk_buff *nskb;
1485 struct sock *sk;
1486 struct inet_sock *inet;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001488 if (ip_options_echo(&replyopts.opt.opt, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return;
1490
David S. Miller0a5ebb82011-05-09 13:22:43 -07001491 ipc.addr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +00001493 ipc.tx_flags = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001494 ipc.ttl = 0;
1495 ipc.tos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001497 if (replyopts.opt.opt.optlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 ipc.opt = &replyopts.opt;
1499
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001500 if (replyopts.opt.opt.srr)
1501 daddr = replyopts.opt.opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 }
1503
David S. Miller77968b72011-05-08 17:12:19 -07001504 flowi4_init_output(&fl4, arg->bound_dev_if, 0,
Eric Dumazet66b13d92011-10-24 03:06:21 -04001505 RT_TOS(arg->tos),
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001506 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
David S. Miller77968b72011-05-08 17:12:19 -07001507 ip_reply_arg_flowi_flags(arg),
David S. Miller70e73412012-06-28 03:21:41 -07001508 daddr, saddr,
David S. Miller77968b72011-05-08 17:12:19 -07001509 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1510 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001511 rt = ip_route_output_key(net, &fl4);
David S. Miller77968b72011-05-08 17:12:19 -07001512 if (IS_ERR(rt))
1513 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001515 inet = &get_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Eric Dumazet66b13d92011-10-24 03:06:21 -04001517 inet->tos = arg->tos;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001518 sk = &inet->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001520 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001521 sk->sk_bound_dev_if = arg->bound_dev_if;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001522 sock_net_set(sk, net);
1523 __skb_queue_head_init(&sk->sk_write_queue);
1524 sk->sk_sndbuf = sysctl_wmem_default;
David S. Millerf5fca602011-05-08 17:24:10 -07001525 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
Eric Dumazet2e77d892008-11-24 15:52:46 -08001526 &ipc, &rt, MSG_DONTWAIT);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001527 nskb = skb_peek(&sk->sk_write_queue);
1528 if (nskb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 if (arg->csumoffset >= 0)
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001530 *((__sum16 *)skb_transport_header(nskb) +
1531 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001532 arg->csum));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001533 nskb->ip_summed = CHECKSUM_NONE;
Eric Dumazet3a7c3842012-08-09 13:56:06 +00001534 skb_orphan(nskb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001535 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
David S. Miller77968b72011-05-08 17:12:19 -07001536 ip_push_pending_frames(sk, &fl4);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 }
1538
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001539 put_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541 ip_rt_put(rt);
1542}
1543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544void __init ip_init(void)
1545{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 ip_rt_init();
1547 inet_initpeers();
1548
WANG Cong72c1d3b2014-01-10 16:09:45 -08001549#if defined(CONFIG_IP_MULTICAST)
1550 igmp_mc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551#endif
1552}