blob: 215af2b155cb1946ff5f4b4d7d0fff19291573b8 [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;
Eric Dumazet73f156a2014-06-02 05:26:03 -0700151 ip_select_ident(skb, 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
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200214static int ip_finish_output_gso(struct sk_buff *skb)
215{
216 netdev_features_t features;
217 struct sk_buff *segs;
218 int ret = 0;
219
220 /* common case: locally created skb or seglen is <= mtu */
221 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
222 skb_gso_network_seglen(skb) <= ip_skb_dst_mtu(skb))
223 return ip_finish_output2(skb);
224
225 /* Slowpath - GSO segment length is exceeding the dst MTU.
226 *
227 * This can happen in two cases:
228 * 1) TCP GRO packet, DF bit not set
229 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
230 * from host network stack.
231 */
232 features = netif_skb_features(skb);
233 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
234 if (IS_ERR(segs)) {
235 kfree_skb(skb);
236 return -ENOMEM;
237 }
238
239 consume_skb(skb);
240
241 do {
242 struct sk_buff *nskb = segs->next;
243 int err;
244
245 segs->next = NULL;
246 err = ip_fragment(segs, ip_finish_output2);
247
248 if (err && ret == 0)
249 ret = err;
250 segs = nskb;
251 } while (segs);
252
253 return ret;
254}
255
Patrick McHardy861d0482007-10-15 01:48:39 -0700256static int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800258#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
259 /* Policy lookup after SNAT yielded a new policy */
Eric Dumazetadf30902009-06-02 05:19:30 +0000260 if (skb_dst(skb)->xfrm != NULL) {
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800261 IPCB(skb)->flags |= IPSKB_REROUTED;
262 return dst_output(skb);
263 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800264#endif
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200265 if (skb_is_gso(skb))
266 return ip_finish_output_gso(skb);
267
268 if (skb->len > ip_skb_dst_mtu(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800269 return ip_fragment(skb, ip_finish_output2);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200270
271 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272}
273
Eric Dumazetaad88722014-04-15 13:47:15 -0400274int ip_mc_output(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
Eric Dumazet511c3f92009-06-02 05:14:27 +0000276 struct rtable *rt = skb_rtable(skb);
Changli Gaod8d1f302010-06-10 23:31:35 -0700277 struct net_device *dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
279 /*
280 * If the indicated interface is up and running, send the packet.
281 */
Neil Hormanedf391f2009-04-27 02:45:02 -0700282 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 skb->dev = dev;
285 skb->protocol = htons(ETH_P_IP);
286
287 /*
288 * Multicasts are looped back for other local users
289 */
290
291 if (rt->rt_flags&RTCF_MULTICAST) {
Octavian Purdila7ad68482010-01-06 20:37:01 -0800292 if (sk_mc_loop(sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293#ifdef CONFIG_IP_MROUTE
294 /* Small optimization: do not loopback not local frames,
295 which returned after forwarding; they will be dropped
296 by ip_mr_input in any case.
297 Note, that local frames are looped back to be delivered
298 to local recipients.
299
300 This check is duplicated in ip_mr_input at the moment.
301 */
Joe Perches9d4fb272009-11-23 10:41:23 -0800302 &&
303 ((rt->rt_flags & RTCF_LOCAL) ||
304 !(IPCB(skb)->flags & IPSKB_FORWARDED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305#endif
Joe Perches9d4fb272009-11-23 10:41:23 -0800306 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
308 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100309 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
310 newskb, NULL, newskb->dev,
Michel Machado95603e22012-06-12 10:16:35 +0000311 dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 }
313
314 /* Multicasts with ttl 0 must not go beyond the host */
315
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700316 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 kfree_skb(skb);
318 return 0;
319 }
320 }
321
322 if (rt->rt_flags&RTCF_BROADCAST) {
323 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
324 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100325 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
Michel Machado95603e22012-06-12 10:16:35 +0000326 NULL, newskb->dev, dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 }
328
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100329 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
330 skb->dev, ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800331 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332}
333
Eric Dumazetaad88722014-04-15 13:47:15 -0400334int ip_output(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
Eric Dumazetadf30902009-06-02 05:19:30 +0000336 struct net_device *dev = skb_dst(skb)->dev;
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800337
Neil Hormanedf391f2009-04-27 02:45:02 -0700338 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800340 skb->dev = dev;
341 skb->protocol = htons(ETH_P_IP);
342
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100343 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900344 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800345 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
Eric Dumazet84f93072011-11-30 19:00:53 +0000348/*
349 * copy saddr and daddr, possibly using 64bit load/stores
350 * Equivalent to :
351 * iph->saddr = fl4->saddr;
352 * iph->daddr = fl4->daddr;
353 */
354static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
355{
356 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
357 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
358 memcpy(&iph->saddr, &fl4->saddr,
359 sizeof(fl4->saddr) + sizeof(fl4->daddr));
360}
361
Eric Dumazetb0270e92014-04-15 12:58:34 -0400362/* Note: skb->sk can be different from sk, in case of tunnels */
363int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000366 struct ip_options_rcu *inet_opt;
David S. Millerb57ae012011-05-06 16:24:06 -0700367 struct flowi4 *fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 struct rtable *rt;
369 struct iphdr *iph;
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000370 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 /* Skip all of this if the packet is already routed,
373 * f.e. by something like SCTP.
374 */
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000375 rcu_read_lock();
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000376 inet_opt = rcu_dereference(inet->inet_opt);
David S. Millerea4fc0d2011-05-06 22:30:20 -0700377 fl4 = &fl->u.ip4;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000378 rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (rt != NULL)
380 goto packet_routed;
381
382 /* Make sure we can route this packet. */
383 rt = (struct rtable *)__sk_dst_check(sk, 0);
384 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700385 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 /* Use correct destination address if we have options. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000388 daddr = inet->inet_daddr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000389 if (inet_opt && inet_opt->opt.srr)
390 daddr = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
David S. Miller78fbfd82011-03-12 00:00:52 -0500392 /* If this fails, retransmit mechanism of transport layer will
393 * keep trying until route appears or the connection times
394 * itself out.
395 */
David S. Millerb57ae012011-05-06 16:24:06 -0700396 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500397 daddr, inet->inet_saddr,
398 inet->inet_dport,
399 inet->inet_sport,
400 sk->sk_protocol,
401 RT_CONN_FLAGS(sk),
402 sk->sk_bound_dev_if);
403 if (IS_ERR(rt))
404 goto no_route;
Changli Gaod8d1f302010-06-10 23:31:35 -0700405 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 }
Changli Gaod8d1f302010-06-10 23:31:35 -0700407 skb_dst_set_noref(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409packet_routed:
Julian Anastasov155e8332012-10-08 11:41:18 +0000410 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 goto no_route;
412
413 /* OK, we know where to send it, allocate and build IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000414 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300415 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700416 iph = ip_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800417 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
WANG Cong60ff7462014-05-04 16:39:18 -0700418 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 iph->frag_off = htons(IP_DF);
420 else
421 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700422 iph->ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +0000424 ip_copy_addrs(iph, fl4);
425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 /* Transport layer set skb->h.foo itself. */
427
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000428 if (inet_opt && inet_opt->opt.optlen) {
429 iph->ihl += inet_opt->opt.optlen >> 2;
430 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 }
432
Eric Dumazet73f156a2014-06-02 05:26:03 -0700433 ip_select_ident_segs(skb, sk, skb_shinfo(skb)->gso_segs ?: 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Eric Dumazetb0270e92014-04-15 12:58:34 -0400435 /* TODO : should we use skb->sk here instead of sk ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800437 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000439 res = ip_local_out(skb);
440 rcu_read_unlock();
441 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443no_route:
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000444 rcu_read_unlock();
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700445 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 kfree_skb(skb);
447 return -EHOSTUNREACH;
448}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000449EXPORT_SYMBOL(ip_queue_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451
452static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
453{
454 to->pkt_type = from->pkt_type;
455 to->priority = from->priority;
456 to->protocol = from->protocol;
Eric Dumazetadf30902009-06-02 05:19:30 +0000457 skb_dst_drop(to);
Eric Dumazetfe76cda2010-07-01 23:48:22 +0000458 skb_dst_copy(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800460 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 /* Copy the flags to each fragment. */
463 IPCB(to)->flags = IPCB(from)->flags;
464
465#ifdef CONFIG_NET_SCHED
466 to->tc_index = from->tc_index;
467#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700468 nf_copy(to, from);
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300469#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
470 to->ipvs_property = from->ipvs_property;
471#endif
James Morris984bc162006-06-09 00:29:17 -0700472 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475/*
476 * This IP datagram is too large to be sent in one piece. Break it up into
477 * smaller pieces (each of size equal to IP header plus
478 * a block of the data of the original IP data part) that will yet fit in a
479 * single device frame, and queue such a frame for sending.
480 */
481
Jianjun Kongd93191002008-11-03 00:23:42 -0800482int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
484 struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 int ptr;
486 struct net_device *dev;
487 struct sk_buff *skb2;
Changli Gaoc893b802010-07-31 13:25:08 +0000488 unsigned int mtu, hlen, left, len, ll_rs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800490 __be16 not_last_frag;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000491 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int err = 0;
493
Changli Gaod8d1f302010-06-10 23:31:35 -0700494 dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
496 /*
497 * Point into the IP datagram header.
498 */
499
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700500 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Hannes Frederic Sowa69647ce2014-02-26 01:20:41 +0100502 mtu = ip_skb_dst_mtu(skb);
WANG Cong60ff7462014-05-04 16:39:18 -0700503 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200504 (IPCB(skb)->frag_max_size &&
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100505 IPCB(skb)->frag_max_size > mtu))) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700506 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100508 htonl(mtu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 kfree_skb(skb);
510 return -EMSGSIZE;
511 }
512
513 /*
514 * Setup starting values.
515 */
516
517 hlen = iph->ihl * 4;
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100518 mtu = mtu - hlen; /* Size of data space */
Bart De Schuymer6c79bf02010-04-20 16:22:01 +0200519#ifdef CONFIG_BRIDGE_NETFILTER
520 if (skb->nf_bridge)
521 mtu -= nf_bridge_mtu_reduction(skb);
522#endif
Herbert Xu89cee8b2005-12-13 23:14:27 -0800523 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 /* When frag_list is given, use it. First, check its validity:
526 * some transformers could create wrong frag_list or break existing
527 * one, it is not prohibited. In this case fall back to copying.
528 *
529 * LATER: this step can be merged to real generation of fragments,
530 * we can switch to copy when see the first bad fragment.
531 */
David S. Miller21dc3302010-08-23 00:13:46 -0700532 if (skb_has_frag_list(skb)) {
Eric Dumazet3d130082010-09-21 08:47:45 +0000533 struct sk_buff *frag, *frag2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 int first_len = skb_pagelen(skb);
535
536 if (first_len - hlen > mtu ||
537 ((first_len - hlen) & 7) ||
Paul Gortmaker56f8a752011-06-21 20:33:34 -0700538 ip_is_fragment(iph) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 skb_cloned(skb))
540 goto slow_path;
541
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700542 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 /* Correct geometry. */
544 if (frag->len > mtu ||
545 ((frag->len & 7) && frag->next) ||
546 skb_headroom(frag) < hlen)
Eric Dumazet3d130082010-09-21 08:47:45 +0000547 goto slow_path_clean;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549 /* Partially cloned skb? */
550 if (skb_shared(frag))
Eric Dumazet3d130082010-09-21 08:47:45 +0000551 goto slow_path_clean;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700552
553 BUG_ON(frag->sk);
554 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700555 frag->sk = skb->sk;
556 frag->destructor = sock_wfree;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700557 }
Eric Dumazet3d130082010-09-21 08:47:45 +0000558 skb->truesize -= frag->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 }
560
561 /* Everything is OK. Generate! */
562
563 err = 0;
564 offset = 0;
565 frag = skb_shinfo(skb)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700566 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 skb->data_len = first_len - skb_headlen(skb);
568 skb->len = first_len;
569 iph->tot_len = htons(first_len);
570 iph->frag_off = htons(IP_MF);
571 ip_send_check(iph);
572
573 for (;;) {
574 /* Prepare header of the next frame,
575 * before previous one went down. */
576 if (frag) {
577 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300578 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700579 __skb_push(frag, hlen);
580 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700581 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700582 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 iph->tot_len = htons(frag->len);
584 ip_copy_metadata(frag, skb);
585 if (offset == 0)
586 ip_options_fragment(frag);
587 offset += skb->len - hlen;
588 iph->frag_off = htons(offset>>3);
589 if (frag->next != NULL)
590 iph->frag_off |= htons(IP_MF);
591 /* Ready, complete checksum */
592 ip_send_check(iph);
593 }
594
595 err = output(skb);
596
Wei Dongdafee492006-08-02 13:41:21 -0700597 if (!err)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700598 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 if (err || !frag)
600 break;
601
602 skb = frag;
603 frag = skb->next;
604 skb->next = NULL;
605 }
606
607 if (err == 0) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700608 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 return 0;
610 }
611
612 while (frag) {
613 skb = frag->next;
614 kfree_skb(frag);
615 frag = skb;
616 }
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700617 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 return err;
Eric Dumazet3d130082010-09-21 08:47:45 +0000619
620slow_path_clean:
621 skb_walk_frags(skb, frag2) {
622 if (frag2 == frag)
623 break;
624 frag2->sk = NULL;
625 frag2->destructor = NULL;
626 skb->truesize += frag2->truesize;
627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629
630slow_path:
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000631 /* for offloaded checksums cleanup checksum before fragmentation */
632 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
633 goto fail;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000634 iph = ip_hdr(skb);
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 left = skb->len - hlen; /* Space per frame */
George Kadianakis49085bd2010-07-06 11:44:12 +0000637 ptr = hlen; /* Where to start from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700640 * we need to make room for the encapsulating header
641 */
Changli Gaoc893b802010-07-31 13:25:08 +0000642 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 /*
645 * Fragment the datagram.
646 */
647
648 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
649 not_last_frag = iph->frag_off & htons(IP_MF);
650
651 /*
652 * Keep copying data until we run out.
653 */
654
Stephen Hemminger132adf52007-03-08 20:44:43 -0800655 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 len = left;
657 /* IF: it doesn't fit, use 'mtu' - the data space left */
658 if (len > mtu)
659 len = mtu;
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300660 /* IF: we are not sending up to and including the packet end
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 then align the next start on an eight byte boundary */
662 if (len < left) {
663 len &= ~7;
664 }
665 /*
666 * Allocate buffer.
667 */
668
669 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700670 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 err = -ENOMEM;
672 goto fail;
673 }
674
675 /*
676 * Set up data on packet
677 */
678
679 ip_copy_metadata(skb2, skb);
680 skb_reserve(skb2, ll_rs);
681 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700682 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700683 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
685 /*
686 * Charge the memory for the fragment to any owner
687 * it might possess
688 */
689
690 if (skb->sk)
691 skb_set_owner_w(skb2, skb->sk);
692
693 /*
694 * Copy the packet header into the new buffer.
695 */
696
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300697 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699 /*
700 * Copy a block of the IP datagram.
701 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300702 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 BUG();
704 left -= len;
705
706 /*
707 * Fill in the new header fields.
708 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700709 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 iph->frag_off = htons((offset >> 3));
711
712 /* ANK: dirty, but effective trick. Upgrade options only if
713 * the segment to be fragmented was THE FIRST (otherwise,
714 * options are already fixed) and make it ONCE
715 * on the initial skb, so that all the following fragments
716 * will inherit fixed options.
717 */
718 if (offset == 0)
719 ip_options_fragment(skb);
720
721 /*
722 * Added AC : If we are fragmenting a fragment that's not the
723 * last fragment then keep MF on each bit
724 */
725 if (left > 0 || not_last_frag)
726 iph->frag_off |= htons(IP_MF);
727 ptr += len;
728 offset += len;
729
730 /*
731 * Put this fragment into the sending queue.
732 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 iph->tot_len = htons(len + hlen);
734
735 ip_send_check(iph);
736
737 err = output(skb2);
738 if (err)
739 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700740
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700741 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000743 consume_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700744 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return err;
746
747fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900748 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700749 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return err;
751}
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700752EXPORT_SYMBOL(ip_fragment);
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754int
755ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
756{
757 struct iovec *iov = from;
758
Patrick McHardy84fa7932006-08-29 16:44:56 -0700759 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
761 return -EFAULT;
762 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800763 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
765 return -EFAULT;
766 skb->csum = csum_block_add(skb->csum, csum, odd);
767 }
768 return 0;
769}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000770EXPORT_SYMBOL(ip_generic_getfrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Al Viro44bb9362006-11-14 21:36:14 -0800772static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773csum_page(struct page *page, int offset, int copy)
774{
775 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800776 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 kaddr = kmap(page);
778 csum = csum_partial(kaddr + offset, copy, 0);
779 kunmap(page);
780 return csum;
781}
782
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800783static inline int ip_ufo_append_data(struct sock *sk,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000784 struct sk_buff_head *queue,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700785 int getfrag(void *from, char *to, int offset, int len,
786 int odd, struct sk_buff *skb),
787 void *from, int length, int hh_len, int fragheaderlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000788 int transhdrlen, int maxfraglen, unsigned int flags)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700789{
790 struct sk_buff *skb;
791 int err;
792
793 /* There is support for UDP fragmentation offload by network
794 * device, so create one single skb packet containing complete
795 * udp datagram
796 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000797 if ((skb = skb_peek_tail(queue)) == NULL) {
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700798 skb = sock_alloc_send_skb(sk,
799 hh_len + fragheaderlen + transhdrlen + 20,
800 (flags & MSG_DONTWAIT), &err);
801
802 if (skb == NULL)
803 return err;
804
805 /* reserve space for Hardware header */
806 skb_reserve(skb, hh_len);
807
808 /* create space for UDP/IP header */
Jianjun Kongd93191002008-11-03 00:23:42 -0800809 skb_put(skb, fragheaderlen + transhdrlen);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700810
811 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700812 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700813
814 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700815 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700816
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700817 skb->csum = 0;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700818
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200819
Herbert Xu1470ddf2011-03-01 02:36:47 +0000820 __skb_queue_tail(queue, skb);
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200821 } else if (skb_is_gso(skb)) {
822 goto append;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700823 }
Kostya Bbe9164e2008-04-29 22:36:30 -0700824
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200825 skb->ip_summed = CHECKSUM_PARTIAL;
826 /* specify the length of each IP datagram fragment */
827 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
828 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
829
830append:
Kostya Bbe9164e2008-04-29 22:36:30 -0700831 return skb_append_datato_frags(sk, skb, getfrag, from,
832 (length - transhdrlen));
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700833}
834
David S. Millerf5fca602011-05-08 17:24:10 -0700835static int __ip_append_data(struct sock *sk,
836 struct flowi4 *fl4,
837 struct sk_buff_head *queue,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000838 struct inet_cork *cork,
Eric Dumazet5640f762012-09-23 23:04:42 +0000839 struct page_frag *pfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000840 int getfrag(void *from, char *to, int offset,
841 int len, int odd, struct sk_buff *skb),
842 void *from, int length, int transhdrlen,
843 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
845 struct inet_sock *inet = inet_sk(sk);
846 struct sk_buff *skb;
847
Herbert Xu07df5292011-03-01 23:00:58 -0800848 struct ip_options *opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 int hh_len;
850 int exthdrlen;
851 int mtu;
852 int copy;
853 int err;
854 int offset = 0;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100855 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 int csummode = CHECKSUM_NONE;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000857 struct rtable *rt = (struct rtable *)cork->dst;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400858 u32 tskey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Steffen Klassert96d73032011-06-05 20:48:47 +0000860 skb = skb_peek_tail(queue);
861
862 exthdrlen = !skb ? rt->dst.header_len : 0;
Herbert Xu07df5292011-03-01 23:00:58 -0800863 mtu = cork->fragsize;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400864 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
865 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
866 tskey = sk->sk_tskey++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867
Changli Gaod8d1f302010-06-10 23:31:35 -0700868 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
871 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
WANG Cong60ff7462014-05-04 16:39:18 -0700872 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100874 if (cork->length + length > maxnonfragsize - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -0700875 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +0100876 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 return -EMSGSIZE;
878 }
879
880 /*
881 * transhdrlen > 0 means that this is the first fragment and we wish
882 * it won't be fragmented in the future.
883 */
884 if (transhdrlen &&
885 length + fragheaderlen <= mtu &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700886 rt->dst.dev->features & NETIF_F_V4_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700888 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Herbert Xu1470ddf2011-03-01 02:36:47 +0000890 cork->length += length;
Herbert Xu26cde9f2010-06-15 01:52:25 +0000891 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
Kostya Bbe9164e2008-04-29 22:36:30 -0700892 (sk->sk_protocol == IPPROTO_UDP) &&
Steffen Klassertc1460662011-06-29 23:19:32 +0000893 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
Herbert Xu1470ddf2011-03-01 02:36:47 +0000894 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
895 hh_len, fragheaderlen, transhdrlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000896 maxfraglen, flags);
Patrick McHardybaa829d2006-03-12 20:35:12 -0800897 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700898 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700899 return 0;
900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 /* So, what's going on in the loop below?
903 *
904 * We use calculated fragment length to generate chained skb,
905 * each of segments is IP fragment ready for sending to network after
906 * adding appropriate IP header.
907 */
908
Herbert Xu26cde9f2010-06-15 01:52:25 +0000909 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 goto alloc_new_skb;
911
912 while (length > 0) {
913 /* Check if the remaining data fits into current packet. */
914 copy = mtu - skb->len;
915 if (copy < length)
916 copy = maxfraglen - skb->len;
917 if (copy <= 0) {
918 char *data;
919 unsigned int datalen;
920 unsigned int fraglen;
921 unsigned int fraggap;
922 unsigned int alloclen;
923 struct sk_buff *skb_prev;
924alloc_new_skb:
925 skb_prev = skb;
926 if (skb_prev)
927 fraggap = skb_prev->len - maxfraglen;
928 else
929 fraggap = 0;
930
931 /*
932 * If remaining data exceeds the mtu,
933 * we know we need more fragment(s).
934 */
935 datalen = length + fraggap;
936 if (datalen > mtu - fragheaderlen)
937 datalen = maxfraglen - fragheaderlen;
938 fraglen = datalen + fragheaderlen;
939
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900940 if ((flags & MSG_MORE) &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700941 !(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 alloclen = mtu;
943 else
Eric Dumazet59104f02010-09-20 20:16:27 +0000944 alloclen = fraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Steffen Klassert353e5c92011-06-22 01:05:37 +0000946 alloclen += exthdrlen;
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 /* The last fragment gets additional space at tail.
949 * Note, with MSG_MORE we overallocate on fragments,
950 * because we have no idea what fragment will be
951 * the last.
952 */
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000953 if (datalen == length + fraggap)
Changli Gaod8d1f302010-06-10 23:31:35 -0700954 alloclen += rt->dst.trailer_len;
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900957 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 alloclen + hh_len + 15,
959 (flags & MSG_DONTWAIT), &err);
960 } else {
961 skb = NULL;
962 if (atomic_read(&sk->sk_wmem_alloc) <=
963 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900964 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 alloclen + hh_len + 15, 1,
966 sk->sk_allocation);
967 if (unlikely(skb == NULL))
968 err = -ENOBUFS;
969 }
970 if (skb == NULL)
971 goto error;
972
973 /*
974 * Fill in the control structures
975 */
976 skb->ip_summed = csummode;
977 skb->csum = 0;
978 skb_reserve(skb, hh_len);
Willem de Bruijn11878b42014-07-14 17:55:06 -0400979
980 /* only the initial fragment is time stamped */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000981 skb_shinfo(skb)->tx_flags = cork->tx_flags;
Willem de Bruijn11878b42014-07-14 17:55:06 -0400982 cork->tx_flags = 0;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400983 skb_shinfo(skb)->tskey = tskey;
984 tskey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 /*
987 * Find where to start putting bytes.
988 */
Steffen Klassert353e5c92011-06-22 01:05:37 +0000989 data = skb_put(skb, fraglen + exthdrlen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -0300990 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700991 skb->transport_header = (skb->network_header +
992 fragheaderlen);
Steffen Klassert353e5c92011-06-22 01:05:37 +0000993 data += fragheaderlen + exthdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 if (fraggap) {
996 skb->csum = skb_copy_and_csum_bits(
997 skb_prev, maxfraglen,
998 data + transhdrlen, fraggap, 0);
999 skb_prev->csum = csum_sub(skb_prev->csum,
1000 skb->csum);
1001 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -07001002 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 }
1004
1005 copy = datalen - transhdrlen - fraggap;
1006 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1007 err = -EFAULT;
1008 kfree_skb(skb);
1009 goto error;
1010 }
1011
1012 offset += copy;
1013 length -= datalen - fraggap;
1014 transhdrlen = 0;
1015 exthdrlen = 0;
1016 csummode = CHECKSUM_NONE;
1017
1018 /*
1019 * Put the packet on the pending queue.
1020 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001021 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 continue;
1023 }
1024
1025 if (copy > length)
1026 copy = length;
1027
Changli Gaod8d1f302010-06-10 23:31:35 -07001028 if (!(rt->dst.dev->features&NETIF_F_SG)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 unsigned int off;
1030
1031 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001032 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 offset, copy, off, skb) < 0) {
1034 __skb_trim(skb, off);
1035 err = -EFAULT;
1036 goto error;
1037 }
1038 } else {
1039 int i = skb_shinfo(skb)->nr_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Eric Dumazet5640f762012-09-23 23:04:42 +00001041 err = -ENOMEM;
1042 if (!sk_page_frag_refill(sk, pfrag))
1043 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Eric Dumazet5640f762012-09-23 23:04:42 +00001045 if (!skb_can_coalesce(skb, i, pfrag->page,
1046 pfrag->offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 err = -EMSGSIZE;
Eric Dumazet5640f762012-09-23 23:04:42 +00001048 if (i == MAX_SKB_FRAGS)
1049 goto error;
1050
1051 __skb_fill_page_desc(skb, i, pfrag->page,
1052 pfrag->offset, 0);
1053 skb_shinfo(skb)->nr_frags = ++i;
1054 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001056 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1057 if (getfrag(from,
1058 page_address(pfrag->page) + pfrag->offset,
1059 offset, copy, skb->len, skb) < 0)
1060 goto error_efault;
1061
1062 pfrag->offset += copy;
1063 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 skb->len += copy;
1065 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001066 skb->truesize += copy;
1067 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
1069 offset += copy;
1070 length -= copy;
1071 }
1072
1073 return 0;
1074
Eric Dumazet5640f762012-09-23 23:04:42 +00001075error_efault:
1076 err = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077error:
Herbert Xu1470ddf2011-03-01 02:36:47 +00001078 cork->length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001079 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001080 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081}
1082
Herbert Xu1470ddf2011-03-01 02:36:47 +00001083static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1084 struct ipcm_cookie *ipc, struct rtable **rtp)
1085{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001086 struct ip_options_rcu *opt;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001087 struct rtable *rt;
1088
1089 /*
1090 * setup for corking.
1091 */
1092 opt = ipc->opt;
1093 if (opt) {
1094 if (cork->opt == NULL) {
1095 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1096 sk->sk_allocation);
1097 if (unlikely(cork->opt == NULL))
1098 return -ENOBUFS;
1099 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001100 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001101 cork->flags |= IPCORK_OPT;
1102 cork->addr = ipc->addr;
1103 }
1104 rt = *rtp;
1105 if (unlikely(!rt))
1106 return -EFAULT;
1107 /*
1108 * We steal reference to this route, caller should not release it
1109 */
1110 *rtp = NULL;
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001111 cork->fragsize = ip_sk_use_pmtu(sk) ?
1112 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001113 cork->dst = &rt->dst;
1114 cork->length = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001115 cork->ttl = ipc->ttl;
1116 cork->tos = ipc->tos;
1117 cork->priority = ipc->priority;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001118 cork->tx_flags = ipc->tx_flags;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001119
1120 return 0;
1121}
1122
1123/*
1124 * ip_append_data() and ip_append_page() can make one large IP datagram
1125 * from many pieces of data. Each pieces will be holded on the socket
1126 * until ip_push_pending_frames() is called. Each piece can be a page
1127 * or non-page data.
1128 *
1129 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1130 * this interface potentially.
1131 *
1132 * LATER: length must be adjusted by pad at tail, when it is required.
1133 */
David S. Millerf5fca602011-05-08 17:24:10 -07001134int ip_append_data(struct sock *sk, struct flowi4 *fl4,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001135 int getfrag(void *from, char *to, int offset, int len,
1136 int odd, struct sk_buff *skb),
1137 void *from, int length, int transhdrlen,
1138 struct ipcm_cookie *ipc, struct rtable **rtp,
1139 unsigned int flags)
1140{
1141 struct inet_sock *inet = inet_sk(sk);
1142 int err;
1143
1144 if (flags&MSG_PROBE)
1145 return 0;
1146
1147 if (skb_queue_empty(&sk->sk_write_queue)) {
David S. Millerbdc712b2011-05-06 15:02:07 -07001148 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001149 if (err)
1150 return err;
1151 } else {
1152 transhdrlen = 0;
1153 }
1154
Eric Dumazet5640f762012-09-23 23:04:42 +00001155 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1156 sk_page_frag(sk), getfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001157 from, length, transhdrlen, flags);
1158}
1159
David S. Millerf5fca602011-05-08 17:24:10 -07001160ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 int offset, size_t size, int flags)
1162{
1163 struct inet_sock *inet = inet_sk(sk);
1164 struct sk_buff *skb;
1165 struct rtable *rt;
1166 struct ip_options *opt = NULL;
David S. Millerbdc712b2011-05-06 15:02:07 -07001167 struct inet_cork *cork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 int hh_len;
1169 int mtu;
1170 int len;
1171 int err;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001172 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173
1174 if (inet->hdrincl)
1175 return -EPERM;
1176
1177 if (flags&MSG_PROBE)
1178 return 0;
1179
1180 if (skb_queue_empty(&sk->sk_write_queue))
1181 return -EINVAL;
1182
David S. Millerbdc712b2011-05-06 15:02:07 -07001183 cork = &inet->cork.base;
1184 rt = (struct rtable *)cork->dst;
1185 if (cork->flags & IPCORK_OPT)
1186 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Changli Gaod8d1f302010-06-10 23:31:35 -07001188 if (!(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return -EOPNOTSUPP;
1190
Changli Gaod8d1f302010-06-10 23:31:35 -07001191 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
David S. Millerbdc712b2011-05-06 15:02:07 -07001192 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1195 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
WANG Cong60ff7462014-05-04 16:39:18 -07001196 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001198 if (cork->length + size > maxnonfragsize - fragheaderlen) {
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +01001199 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1200 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return -EMSGSIZE;
1202 }
1203
1204 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1205 return -EINVAL;
1206
David S. Millerbdc712b2011-05-06 15:02:07 -07001207 cork->length += size;
Herbert Xu26cde9f2010-06-15 01:52:25 +00001208 if ((size + skb->len > mtu) &&
1209 (sk->sk_protocol == IPPROTO_UDP) &&
Changli Gaod8d1f302010-06-10 23:31:35 -07001210 (rt->dst.dev->features & NETIF_F_UFO)) {
Herbert Xu79671682006-06-22 02:40:14 -07001211 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001212 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001213 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
1216 while (size > 0) {
1217 int i;
1218
Herbert Xu89114af2006-07-08 13:34:32 -07001219 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001220 len = size;
1221 else {
1222
1223 /* Check if the remaining data fits into current packet. */
1224 len = mtu - skb->len;
1225 if (len < size)
1226 len = maxfraglen - skb->len;
1227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (len <= 0) {
1229 struct sk_buff *skb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 int alloclen;
1231
1232 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001233 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234
1235 alloclen = fragheaderlen + hh_len + fraggap + 15;
1236 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1237 if (unlikely(!skb)) {
1238 err = -ENOBUFS;
1239 goto error;
1240 }
1241
1242 /*
1243 * Fill in the control structures
1244 */
1245 skb->ip_summed = CHECKSUM_NONE;
1246 skb->csum = 0;
1247 skb_reserve(skb, hh_len);
1248
1249 /*
1250 * Find where to start putting bytes.
1251 */
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001252 skb_put(skb, fragheaderlen + fraggap);
Arnaldo Carvalho de Melo2ca9e6f2007-03-10 19:15:25 -03001253 skb_reset_network_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001254 skb->transport_header = (skb->network_header +
1255 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 if (fraggap) {
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001257 skb->csum = skb_copy_and_csum_bits(skb_prev,
1258 maxfraglen,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001259 skb_transport_header(skb),
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001260 fraggap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 skb_prev->csum = csum_sub(skb_prev->csum,
1262 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001263 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 }
1265
1266 /*
1267 * Put the packet on the pending queue.
1268 */
1269 __skb_queue_tail(&sk->sk_write_queue, skb);
1270 continue;
1271 }
1272
1273 i = skb_shinfo(skb)->nr_frags;
1274 if (len > size)
1275 len = size;
1276 if (skb_can_coalesce(skb, i, page, offset)) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001277 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 } else if (i < MAX_SKB_FRAGS) {
1279 get_page(page);
1280 skb_fill_page_desc(skb, i, page, offset, len);
1281 } else {
1282 err = -EMSGSIZE;
1283 goto error;
1284 }
1285
1286 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001287 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 csum = csum_page(page, offset, len);
1289 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1290 }
1291
1292 skb->len += len;
1293 skb->data_len += len;
David S. Miller1e34a112008-01-22 23:44:31 -08001294 skb->truesize += len;
1295 atomic_add(len, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 offset += len;
1297 size -= len;
1298 }
1299 return 0;
1300
1301error:
David S. Millerbdc712b2011-05-06 15:02:07 -07001302 cork->length -= size;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001303 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 return err;
1305}
1306
Herbert Xu1470ddf2011-03-01 02:36:47 +00001307static void ip_cork_release(struct inet_cork *cork)
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001308{
Herbert Xu1470ddf2011-03-01 02:36:47 +00001309 cork->flags &= ~IPCORK_OPT;
1310 kfree(cork->opt);
1311 cork->opt = NULL;
1312 dst_release(cork->dst);
1313 cork->dst = NULL;
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001314}
1315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316/*
1317 * Combined all pending IP fragments on the socket as one IP datagram
1318 * and push them out.
1319 */
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001320struct sk_buff *__ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001321 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001322 struct sk_buff_head *queue,
1323 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324{
1325 struct sk_buff *skb, *tmp_skb;
1326 struct sk_buff **tail_skb;
1327 struct inet_sock *inet = inet_sk(sk);
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001328 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 struct ip_options *opt = NULL;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001330 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001332 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 __u8 ttl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Herbert Xu1470ddf2011-03-01 02:36:47 +00001335 if ((skb = __skb_dequeue(queue)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 goto out;
1337 tail_skb = &(skb_shinfo(skb)->frag_list);
1338
1339 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001340 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001341 __skb_pull(skb, skb_network_offset(skb));
Herbert Xu1470ddf2011-03-01 02:36:47 +00001342 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001343 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 *tail_skb = tmp_skb;
1345 tail_skb = &(tmp_skb->next);
1346 skb->len += tmp_skb->len;
1347 skb->data_len += tmp_skb->len;
1348 skb->truesize += tmp_skb->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 tmp_skb->destructor = NULL;
1350 tmp_skb->sk = NULL;
1351 }
1352
1353 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1354 * to fragment the frame generated here. No matter, what transforms
1355 * how transforms change size of the packet, it will come out.
1356 */
WANG Cong60ff7462014-05-04 16:39:18 -07001357 skb->ignore_df = ip_sk_ignore_df(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
1359 /* DF bit is set when we want to see DF on outgoing frames.
WANG Cong60ff7462014-05-04 16:39:18 -07001360 * If ignore_df is set too, we still allow to fragment this frame
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * locally. */
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001362 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1363 inet->pmtudisc == IP_PMTUDISC_PROBE ||
Changli Gaod8d1f302010-06-10 23:31:35 -07001364 (skb->len <= dst_mtu(&rt->dst) &&
1365 ip_dont_fragment(sk, &rt->dst)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 df = htons(IP_DF);
1367
Herbert Xu1470ddf2011-03-01 02:36:47 +00001368 if (cork->flags & IPCORK_OPT)
1369 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Francesco Fuscoaa661582013-09-24 15:43:09 +02001371 if (cork->ttl != 0)
1372 ttl = cork->ttl;
1373 else if (rt->rt_type == RTN_MULTICAST)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 ttl = inet->mc_ttl;
1375 else
Changli Gaod8d1f302010-06-10 23:31:35 -07001376 ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Ansis Atteka749154a2013-09-18 15:29:52 -07001378 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 iph->version = 4;
1380 iph->ihl = 5;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001381 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 iph->frag_off = df;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 iph->ttl = ttl;
1384 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +00001385 ip_copy_addrs(iph, fl4);
Eric Dumazet73f156a2014-06-02 05:26:03 -07001386 ip_select_ident(skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
David S. Miller22f728f2011-05-13 17:21:27 -04001388 if (opt) {
1389 iph->ihl += opt->optlen>>2;
1390 ip_options_build(skb, opt, cork->addr, rt, 0);
1391 }
1392
Francesco Fuscoaa661582013-09-24 15:43:09 +02001393 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001394 skb->mark = sk->sk_mark;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001395 /*
1396 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1397 * on dst refcount
1398 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001399 cork->dst = NULL;
Changli Gaod8d1f302010-06-10 23:31:35 -07001400 skb_dst_set(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
David L Stevens96793b42007-09-17 09:57:33 -07001402 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001403 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001404 skb_transport_header(skb))->type);
1405
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001406 ip_cork_release(cork);
1407out:
1408 return skb;
1409}
1410
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001411int ip_send_skb(struct net *net, struct sk_buff *skb)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001412{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001413 int err;
1414
Herbert Xuc439cb22008-01-11 19:14:00 -08001415 err = ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 if (err) {
1417 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001418 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (err)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001420 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 }
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
1425
David S. Miller77968b72011-05-08 17:12:19 -07001426int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1470ddf2011-03-01 02:36:47 +00001427{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001428 struct sk_buff *skb;
1429
David S. Miller77968b72011-05-08 17:12:19 -07001430 skb = ip_finish_skb(sk, fl4);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001431 if (!skb)
1432 return 0;
1433
1434 /* Netfilter gets whole the not fragmented skb. */
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001435 return ip_send_skb(sock_net(sk), skb);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001436}
1437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438/*
1439 * Throw away all pending data on the socket.
1440 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001441static void __ip_flush_pending_frames(struct sock *sk,
1442 struct sk_buff_head *queue,
1443 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 struct sk_buff *skb;
1446
Herbert Xu1470ddf2011-03-01 02:36:47 +00001447 while ((skb = __skb_dequeue_tail(queue)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 kfree_skb(skb);
1449
Herbert Xu1470ddf2011-03-01 02:36:47 +00001450 ip_cork_release(cork);
1451}
1452
1453void ip_flush_pending_frames(struct sock *sk)
1454{
David S. Millerbdc712b2011-05-06 15:02:07 -07001455 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001458struct sk_buff *ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001459 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001460 int getfrag(void *from, char *to, int offset,
1461 int len, int odd, struct sk_buff *skb),
1462 void *from, int length, int transhdrlen,
1463 struct ipcm_cookie *ipc, struct rtable **rtp,
1464 unsigned int flags)
1465{
David S. Millerb80d7222011-05-06 15:06:01 -07001466 struct inet_cork cork;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001467 struct sk_buff_head queue;
1468 int err;
1469
1470 if (flags & MSG_PROBE)
1471 return NULL;
1472
1473 __skb_queue_head_init(&queue);
1474
David S. Millerb80d7222011-05-06 15:06:01 -07001475 cork.flags = 0;
1476 cork.addr = 0;
David S. Miller70652722011-05-06 16:01:15 -07001477 cork.opt = NULL;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001478 err = ip_setup_cork(sk, &cork, ipc, rtp);
1479 if (err)
1480 return ERR_PTR(err);
1481
Eric Dumazet5640f762012-09-23 23:04:42 +00001482 err = __ip_append_data(sk, fl4, &queue, &cork,
1483 &current->task_frag, getfrag,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001484 from, length, transhdrlen, flags);
1485 if (err) {
1486 __ip_flush_pending_frames(sk, &queue, &cork);
1487 return ERR_PTR(err);
1488 }
1489
David S. Miller77968b72011-05-08 17:12:19 -07001490 return __ip_make_skb(sk, fl4, &queue, &cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001491}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
1493/*
1494 * Fetch data from kernel space and fill in checksum if needed.
1495 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001496static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 int len, int odd, struct sk_buff *skb)
1498{
Al Viro50842052006-11-14 21:36:34 -08001499 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1502 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001503 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
1505
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001506/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 * Generic function to send a packet as reply to another packet.
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001508 * Used to send some TCP resets/acks so far.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 *
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001510 * Use a fake percpu inet socket to avoid false sharing and contention.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 */
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001512static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1513 .sk = {
1514 .__sk_common = {
1515 .skc_refcnt = ATOMIC_INIT(1),
1516 },
1517 .sk_wmem_alloc = ATOMIC_INIT(1),
1518 .sk_allocation = GFP_ATOMIC,
1519 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
1520 },
Eric Dumazet0980e562012-07-20 22:28:51 +00001521 .pmtudisc = IP_PMTUDISC_WANT,
1522 .uc_ttl = -1,
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001523};
1524
1525void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
David S. Miller70e73412012-06-28 03:21:41 -07001526 __be32 saddr, const struct ip_reply_arg *arg,
1527 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001529 struct ip_options_data replyopts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 struct ipcm_cookie ipc;
David S. Miller77968b72011-05-08 17:12:19 -07001531 struct flowi4 fl4;
Eric Dumazet511c3f92009-06-02 05:14:27 +00001532 struct rtable *rt = skb_rtable(skb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001533 struct sk_buff *nskb;
1534 struct sock *sk;
1535 struct inet_sock *inet;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001537 if (ip_options_echo(&replyopts.opt.opt, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 return;
1539
David S. Miller0a5ebb82011-05-09 13:22:43 -07001540 ipc.addr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +00001542 ipc.tx_flags = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001543 ipc.ttl = 0;
1544 ipc.tos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001546 if (replyopts.opt.opt.optlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 ipc.opt = &replyopts.opt;
1548
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001549 if (replyopts.opt.opt.srr)
1550 daddr = replyopts.opt.opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 }
1552
Lorenzo Colittie1108612014-05-13 10:17:33 -07001553 flowi4_init_output(&fl4, arg->bound_dev_if,
1554 IP4_REPLY_MARK(net, skb->mark),
Eric Dumazet66b13d92011-10-24 03:06:21 -04001555 RT_TOS(arg->tos),
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001556 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
David S. Miller77968b72011-05-08 17:12:19 -07001557 ip_reply_arg_flowi_flags(arg),
David S. Miller70e73412012-06-28 03:21:41 -07001558 daddr, saddr,
David S. Miller77968b72011-05-08 17:12:19 -07001559 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1560 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001561 rt = ip_route_output_key(net, &fl4);
David S. Miller77968b72011-05-08 17:12:19 -07001562 if (IS_ERR(rt))
1563 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001565 inet = &get_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Eric Dumazet66b13d92011-10-24 03:06:21 -04001567 inet->tos = arg->tos;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001568 sk = &inet->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001570 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001571 sk->sk_bound_dev_if = arg->bound_dev_if;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001572 sock_net_set(sk, net);
1573 __skb_queue_head_init(&sk->sk_write_queue);
1574 sk->sk_sndbuf = sysctl_wmem_default;
David S. Millerf5fca602011-05-08 17:24:10 -07001575 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
Eric Dumazet2e77d892008-11-24 15:52:46 -08001576 &ipc, &rt, MSG_DONTWAIT);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001577 nskb = skb_peek(&sk->sk_write_queue);
1578 if (nskb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 if (arg->csumoffset >= 0)
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001580 *((__sum16 *)skb_transport_header(nskb) +
1581 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001582 arg->csum));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001583 nskb->ip_summed = CHECKSUM_NONE;
Eric Dumazet3a7c3842012-08-09 13:56:06 +00001584 skb_orphan(nskb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001585 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
David S. Miller77968b72011-05-08 17:12:19 -07001586 ip_push_pending_frames(sk, &fl4);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
1588
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001589 put_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 ip_rt_put(rt);
1592}
1593
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594void __init ip_init(void)
1595{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 ip_rt_init();
1597 inet_initpeers();
1598
WANG Cong72c1d3b2014-01-10 16:09:45 -08001599#if defined(CONFIG_IP_MULTICAST)
1600 igmp_mc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601#endif
1602}