blob: df184616493f707ea8a0ea353a33d3a5aa981fc7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The Internet Protocol (IP) output module.
7 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Richard Underwood
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
17 *
18 * See ip_input.c for original log
19 *
20 * Fixes:
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090023 * Bradford Johnson: Fix faulty handling of some frames when
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * no route is found.
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090034 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
41 * datagrams.
42 * Hirokazu Takahashi: sendfile() on UDP works now.
43 */
44
45#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/module.h>
47#include <linux/types.h>
48#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mm.h>
50#include <linux/string.h>
51#include <linux/errno.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040052#include <linux/highmem.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090053#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#include <linux/socket.h>
56#include <linux/sockios.h>
57#include <linux/in.h>
58#include <linux/inet.h>
59#include <linux/netdevice.h>
60#include <linux/etherdevice.h>
61#include <linux/proc_fs.h>
62#include <linux/stat.h>
63#include <linux/init.h>
64
65#include <net/snmp.h>
66#include <net/ip.h>
67#include <net/protocol.h>
68#include <net/route.h>
Patrick McHardycfacb052006-01-08 22:36:54 -080069#include <net/xfrm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/skbuff.h>
71#include <net/sock.h>
72#include <net/arp.h>
73#include <net/icmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <net/checksum.h>
75#include <net/inetpeer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/igmp.h>
77#include <linux/netfilter_ipv4.h>
78#include <linux/netfilter_bridge.h>
79#include <linux/mroute.h>
80#include <linux/netlink.h>
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -070081#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Brian Haleyab32ea52006-09-22 14:15:41 -070083int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
David S. Miller323e1262010-12-12 21:55:08 -080084EXPORT_SYMBOL(sysctl_ip_default_ttl);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86/* Generate a checksum for an outgoing IP datagram. */
Denis Efremov2fbd9672013-05-08 23:19:42 +000087void ip_send_check(struct iphdr *iph)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 iph->check = 0;
90 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
91}
Eric Dumazet4bc2f182010-07-09 21:22:10 +000092EXPORT_SYMBOL(ip_send_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
Herbert Xuc439cb22008-01-11 19:14:00 -080094int __ip_local_out(struct sk_buff *skb)
95{
96 struct iphdr *iph = ip_hdr(skb);
97
98 iph->tot_len = htons(skb->len);
99 ip_send_check(iph);
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100100 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
101 skb_dst(skb)->dev, dst_output);
Herbert Xuc439cb22008-01-11 19:14:00 -0800102}
103
104int ip_local_out(struct sk_buff *skb)
105{
106 int err;
107
108 err = __ip_local_out(skb);
109 if (likely(err == 1))
110 err = dst_output(skb);
111
112 return err;
113}
114EXPORT_SYMBOL_GPL(ip_local_out);
115
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
117{
118 int ttl = inet->uc_ttl;
119
120 if (ttl < 0)
David S. Miller323e1262010-12-12 21:55:08 -0800121 ttl = ip4_dst_hoplimit(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return ttl;
123}
124
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900125/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 * Add an ip header to a skbuff and send it out.
127 *
128 */
129int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000130 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
132 struct inet_sock *inet = inet_sk(sk);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000133 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 struct iphdr *iph;
135
136 /* Build the IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000137 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300138 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700139 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 iph->version = 4;
141 iph->ihl = 5;
142 iph->tos = inet->tos;
Changli Gaod8d1f302010-06-10 23:31:35 -0700143 if (ip_dont_fragment(sk, &rt->dst))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 iph->frag_off = htons(IP_DF);
145 else
146 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700147 iph->ttl = ip_select_ttl(inet, &rt->dst);
David S. Millerdd927a22011-05-04 12:03:30 -0700148 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
149 iph->saddr = saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 iph->protocol = sk->sk_protocol;
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
229int ip_mc_output(struct sk_buff *skb)
230{
231 struct sock *sk = skb->sk;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000232 struct rtable *rt = skb_rtable(skb);
Changli Gaod8d1f302010-06-10 23:31:35 -0700233 struct net_device *dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 /*
236 * If the indicated interface is up and running, send the packet.
237 */
Neil Hormanedf391f2009-04-27 02:45:02 -0700238 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 skb->dev = dev;
241 skb->protocol = htons(ETH_P_IP);
242
243 /*
244 * Multicasts are looped back for other local users
245 */
246
247 if (rt->rt_flags&RTCF_MULTICAST) {
Octavian Purdila7ad68482010-01-06 20:37:01 -0800248 if (sk_mc_loop(sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249#ifdef CONFIG_IP_MROUTE
250 /* Small optimization: do not loopback not local frames,
251 which returned after forwarding; they will be dropped
252 by ip_mr_input in any case.
253 Note, that local frames are looped back to be delivered
254 to local recipients.
255
256 This check is duplicated in ip_mr_input at the moment.
257 */
Joe Perches9d4fb272009-11-23 10:41:23 -0800258 &&
259 ((rt->rt_flags & RTCF_LOCAL) ||
260 !(IPCB(skb)->flags & IPSKB_FORWARDED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261#endif
Joe Perches9d4fb272009-11-23 10:41:23 -0800262 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
264 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100265 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
266 newskb, NULL, newskb->dev,
Michel Machado95603e22012-06-12 10:16:35 +0000267 dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 }
269
270 /* Multicasts with ttl 0 must not go beyond the host */
271
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700272 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 kfree_skb(skb);
274 return 0;
275 }
276 }
277
278 if (rt->rt_flags&RTCF_BROADCAST) {
279 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
280 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100281 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
Michel Machado95603e22012-06-12 10:16:35 +0000282 NULL, newskb->dev, dev_loopback_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 }
284
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100285 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
286 skb->dev, ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800287 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
290int ip_output(struct sk_buff *skb)
291{
Eric Dumazetadf30902009-06-02 05:19:30 +0000292 struct net_device *dev = skb_dst(skb)->dev;
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800293
Neil Hormanedf391f2009-04-27 02:45:02 -0700294 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800296 skb->dev = dev;
297 skb->protocol = htons(ETH_P_IP);
298
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100299 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900300 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800301 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302}
303
Eric Dumazet84f93072011-11-30 19:00:53 +0000304/*
305 * copy saddr and daddr, possibly using 64bit load/stores
306 * Equivalent to :
307 * iph->saddr = fl4->saddr;
308 * iph->daddr = fl4->daddr;
309 */
310static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
311{
312 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
313 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
314 memcpy(&iph->saddr, &fl4->saddr,
315 sizeof(fl4->saddr) + sizeof(fl4->daddr));
316}
317
David S. Millerd9d8da82011-05-06 22:23:20 -0700318int ip_queue_xmit(struct sk_buff *skb, struct flowi *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
David S. Millere89862f2007-01-26 01:04:55 -0800320 struct sock *sk = skb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 struct inet_sock *inet = inet_sk(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000322 struct ip_options_rcu *inet_opt;
David S. Millerb57ae012011-05-06 16:24:06 -0700323 struct flowi4 *fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 struct rtable *rt;
325 struct iphdr *iph;
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000326 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 /* Skip all of this if the packet is already routed,
329 * f.e. by something like SCTP.
330 */
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000331 rcu_read_lock();
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000332 inet_opt = rcu_dereference(inet->inet_opt);
David S. Millerea4fc0d2011-05-06 22:30:20 -0700333 fl4 = &fl->u.ip4;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000334 rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 if (rt != NULL)
336 goto packet_routed;
337
338 /* Make sure we can route this packet. */
339 rt = (struct rtable *)__sk_dst_check(sk, 0);
340 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700341 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 /* Use correct destination address if we have options. */
Eric Dumazetc720c7e2009-10-15 06:30:45 +0000344 daddr = inet->inet_daddr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000345 if (inet_opt && inet_opt->opt.srr)
346 daddr = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
David S. Miller78fbfd82011-03-12 00:00:52 -0500348 /* If this fails, retransmit mechanism of transport layer will
349 * keep trying until route appears or the connection times
350 * itself out.
351 */
David S. Millerb57ae012011-05-06 16:24:06 -0700352 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500353 daddr, inet->inet_saddr,
354 inet->inet_dport,
355 inet->inet_sport,
356 sk->sk_protocol,
357 RT_CONN_FLAGS(sk),
358 sk->sk_bound_dev_if);
359 if (IS_ERR(rt))
360 goto no_route;
Changli Gaod8d1f302010-06-10 23:31:35 -0700361 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
Changli Gaod8d1f302010-06-10 23:31:35 -0700363 skb_dst_set_noref(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365packet_routed:
Julian Anastasov155e8332012-10-08 11:41:18 +0000366 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 goto no_route;
368
369 /* OK, we know where to send it, allocate and build IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000370 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300371 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700372 iph = ip_hdr(skb);
Al Viro714e85b2006-11-14 20:51:49 -0800373 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
Changli Gaod8d1f302010-06-10 23:31:35 -0700374 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 iph->frag_off = htons(IP_DF);
376 else
377 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700378 iph->ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +0000380 ip_copy_addrs(iph, fl4);
381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 /* Transport layer set skb->h.foo itself. */
383
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000384 if (inet_opt && inet_opt->opt.optlen) {
385 iph->ihl += inet_opt->opt.optlen >> 2;
386 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
388
Ansis Atteka703133d2013-09-18 15:29:53 -0700389 ip_select_ident_more(skb, &rt->dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700390 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
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);
Gao fengf0165882013-03-21 19:48:42 +0000425#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700426 to->nf_trace = from->nf_trace;
427#endif
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300428#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
429 to->ipvs_property = from->ipvs_property;
430#endif
James Morris984bc162006-06-09 00:29:17 -0700431 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
434/*
435 * This IP datagram is too large to be sent in one piece. Break it up into
436 * smaller pieces (each of size equal to IP header plus
437 * a block of the data of the original IP data part) that will yet fit in a
438 * single device frame, and queue such a frame for sending.
439 */
440
Jianjun Kongd93191002008-11-03 00:23:42 -0800441int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
443 struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 int ptr;
445 struct net_device *dev;
446 struct sk_buff *skb2;
Changli Gaoc893b802010-07-31 13:25:08 +0000447 unsigned int mtu, hlen, left, len, ll_rs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800449 __be16 not_last_frag;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000450 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 int err = 0;
452
Changli Gaod8d1f302010-06-10 23:31:35 -0700453 dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 /*
456 * Point into the IP datagram header.
457 */
458
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700459 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Patrick McHardy5f2d04f2012-08-26 19:13:55 +0200461 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
462 (IPCB(skb)->frag_max_size &&
463 IPCB(skb)->frag_max_size > dst_mtu(&rt->dst)))) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700464 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
John Heffner628a5c52007-04-20 15:53:27 -0700466 htonl(ip_skb_dst_mtu(skb)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 kfree_skb(skb);
468 return -EMSGSIZE;
469 }
470
471 /*
472 * Setup starting values.
473 */
474
475 hlen = iph->ihl * 4;
Changli Gaod8d1f302010-06-10 23:31:35 -0700476 mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */
Bart De Schuymer6c79bf02010-04-20 16:22:01 +0200477#ifdef CONFIG_BRIDGE_NETFILTER
478 if (skb->nf_bridge)
479 mtu -= nf_bridge_mtu_reduction(skb);
480#endif
Herbert Xu89cee8b2005-12-13 23:14:27 -0800481 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 /* When frag_list is given, use it. First, check its validity:
484 * some transformers could create wrong frag_list or break existing
485 * one, it is not prohibited. In this case fall back to copying.
486 *
487 * LATER: this step can be merged to real generation of fragments,
488 * we can switch to copy when see the first bad fragment.
489 */
David S. Miller21dc3302010-08-23 00:13:46 -0700490 if (skb_has_frag_list(skb)) {
Eric Dumazet3d130082010-09-21 08:47:45 +0000491 struct sk_buff *frag, *frag2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int first_len = skb_pagelen(skb);
493
494 if (first_len - hlen > mtu ||
495 ((first_len - hlen) & 7) ||
Paul Gortmaker56f8a752011-06-21 20:33:34 -0700496 ip_is_fragment(iph) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 skb_cloned(skb))
498 goto slow_path;
499
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700500 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 /* Correct geometry. */
502 if (frag->len > mtu ||
503 ((frag->len & 7) && frag->next) ||
504 skb_headroom(frag) < hlen)
Eric Dumazet3d130082010-09-21 08:47:45 +0000505 goto slow_path_clean;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507 /* Partially cloned skb? */
508 if (skb_shared(frag))
Eric Dumazet3d130082010-09-21 08:47:45 +0000509 goto slow_path_clean;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700510
511 BUG_ON(frag->sk);
512 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700513 frag->sk = skb->sk;
514 frag->destructor = sock_wfree;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700515 }
Eric Dumazet3d130082010-09-21 08:47:45 +0000516 skb->truesize -= frag->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 }
518
519 /* Everything is OK. Generate! */
520
521 err = 0;
522 offset = 0;
523 frag = skb_shinfo(skb)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700524 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 skb->data_len = first_len - skb_headlen(skb);
526 skb->len = first_len;
527 iph->tot_len = htons(first_len);
528 iph->frag_off = htons(IP_MF);
529 ip_send_check(iph);
530
531 for (;;) {
532 /* Prepare header of the next frame,
533 * before previous one went down. */
534 if (frag) {
535 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300536 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700537 __skb_push(frag, hlen);
538 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700539 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700540 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 iph->tot_len = htons(frag->len);
542 ip_copy_metadata(frag, skb);
543 if (offset == 0)
544 ip_options_fragment(frag);
545 offset += skb->len - hlen;
546 iph->frag_off = htons(offset>>3);
547 if (frag->next != NULL)
548 iph->frag_off |= htons(IP_MF);
549 /* Ready, complete checksum */
550 ip_send_check(iph);
551 }
552
553 err = output(skb);
554
Wei Dongdafee492006-08-02 13:41:21 -0700555 if (!err)
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700556 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (err || !frag)
558 break;
559
560 skb = frag;
561 frag = skb->next;
562 skb->next = NULL;
563 }
564
565 if (err == 0) {
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700566 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return 0;
568 }
569
570 while (frag) {
571 skb = frag->next;
572 kfree_skb(frag);
573 frag = skb;
574 }
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700575 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 return err;
Eric Dumazet3d130082010-09-21 08:47:45 +0000577
578slow_path_clean:
579 skb_walk_frags(skb, frag2) {
580 if (frag2 == frag)
581 break;
582 frag2->sk = NULL;
583 frag2->destructor = NULL;
584 skb->truesize += frag2->truesize;
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
587
588slow_path:
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000589 /* for offloaded checksums cleanup checksum before fragmentation */
590 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
591 goto fail;
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000592 iph = ip_hdr(skb);
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 left = skb->len - hlen; /* Space per frame */
George Kadianakis49085bd2010-07-06 11:44:12 +0000595 ptr = hlen; /* Where to start from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700598 * we need to make room for the encapsulating header
599 */
Changli Gaoc893b802010-07-31 13:25:08 +0000600 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 /*
603 * Fragment the datagram.
604 */
605
606 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
607 not_last_frag = iph->frag_off & htons(IP_MF);
608
609 /*
610 * Keep copying data until we run out.
611 */
612
Stephen Hemminger132adf52007-03-08 20:44:43 -0800613 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 len = left;
615 /* IF: it doesn't fit, use 'mtu' - the data space left */
616 if (len > mtu)
617 len = mtu;
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300618 /* IF: we are not sending up to and including the packet end
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 then align the next start on an eight byte boundary */
620 if (len < left) {
621 len &= ~7;
622 }
623 /*
624 * Allocate buffer.
625 */
626
627 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700628 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 err = -ENOMEM;
630 goto fail;
631 }
632
633 /*
634 * Set up data on packet
635 */
636
637 ip_copy_metadata(skb2, skb);
638 skb_reserve(skb2, ll_rs);
639 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700640 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700641 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 /*
644 * Charge the memory for the fragment to any owner
645 * it might possess
646 */
647
648 if (skb->sk)
649 skb_set_owner_w(skb2, skb->sk);
650
651 /*
652 * Copy the packet header into the new buffer.
653 */
654
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300655 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
657 /*
658 * Copy a block of the IP datagram.
659 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300660 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 BUG();
662 left -= len;
663
664 /*
665 * Fill in the new header fields.
666 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700667 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 iph->frag_off = htons((offset >> 3));
669
670 /* ANK: dirty, but effective trick. Upgrade options only if
671 * the segment to be fragmented was THE FIRST (otherwise,
672 * options are already fixed) and make it ONCE
673 * on the initial skb, so that all the following fragments
674 * will inherit fixed options.
675 */
676 if (offset == 0)
677 ip_options_fragment(skb);
678
679 /*
680 * Added AC : If we are fragmenting a fragment that's not the
681 * last fragment then keep MF on each bit
682 */
683 if (left > 0 || not_last_frag)
684 iph->frag_off |= htons(IP_MF);
685 ptr += len;
686 offset += len;
687
688 /*
689 * Put this fragment into the sending queue.
690 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 iph->tot_len = htons(len + hlen);
692
693 ip_send_check(iph);
694
695 err = output(skb2);
696 if (err)
697 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700698
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700699 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000701 consume_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700702 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 return err;
704
705fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900706 kfree_skb(skb);
Pavel Emelyanov5e38e272008-07-16 20:19:49 -0700707 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return err;
709}
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700710EXPORT_SYMBOL(ip_fragment);
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712int
713ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
714{
715 struct iovec *iov = from;
716
Patrick McHardy84fa7932006-08-29 16:44:56 -0700717 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
719 return -EFAULT;
720 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800721 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
723 return -EFAULT;
724 skb->csum = csum_block_add(skb->csum, csum, odd);
725 }
726 return 0;
727}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000728EXPORT_SYMBOL(ip_generic_getfrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Al Viro44bb9362006-11-14 21:36:14 -0800730static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731csum_page(struct page *page, int offset, int copy)
732{
733 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800734 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 kaddr = kmap(page);
736 csum = csum_partial(kaddr + offset, copy, 0);
737 kunmap(page);
738 return csum;
739}
740
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800741static inline int ip_ufo_append_data(struct sock *sk,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000742 struct sk_buff_head *queue,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700743 int getfrag(void *from, char *to, int offset, int len,
744 int odd, struct sk_buff *skb),
745 void *from, int length, int hh_len, int fragheaderlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000746 int transhdrlen, int maxfraglen, unsigned int flags)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700747{
748 struct sk_buff *skb;
749 int err;
750
751 /* There is support for UDP fragmentation offload by network
752 * device, so create one single skb packet containing complete
753 * udp datagram
754 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000755 if ((skb = skb_peek_tail(queue)) == NULL) {
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700756 skb = sock_alloc_send_skb(sk,
757 hh_len + fragheaderlen + transhdrlen + 20,
758 (flags & MSG_DONTWAIT), &err);
759
760 if (skb == NULL)
761 return err;
762
763 /* reserve space for Hardware header */
764 skb_reserve(skb, hh_len);
765
766 /* create space for UDP/IP header */
Jianjun Kongd93191002008-11-03 00:23:42 -0800767 skb_put(skb, fragheaderlen + transhdrlen);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700768
769 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700770 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700771
772 /* initialize protocol header pointer */
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700773 skb->transport_header = skb->network_header + fragheaderlen;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700774
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700775 skb->csum = 0;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700776
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200777
Herbert Xu1470ddf2011-03-01 02:36:47 +0000778 __skb_queue_tail(queue, skb);
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200779 } else if (skb_is_gso(skb)) {
780 goto append;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700781 }
Kostya Bbe9164e2008-04-29 22:36:30 -0700782
Jiri Pirkoe93b7d72013-10-19 12:29:17 +0200783 skb->ip_summed = CHECKSUM_PARTIAL;
784 /* specify the length of each IP datagram fragment */
785 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
786 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
787
788append:
Kostya Bbe9164e2008-04-29 22:36:30 -0700789 return skb_append_datato_frags(sk, skb, getfrag, from,
790 (length - transhdrlen));
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700791}
792
David S. Millerf5fca602011-05-08 17:24:10 -0700793static int __ip_append_data(struct sock *sk,
794 struct flowi4 *fl4,
795 struct sk_buff_head *queue,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000796 struct inet_cork *cork,
Eric Dumazet5640f762012-09-23 23:04:42 +0000797 struct page_frag *pfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000798 int getfrag(void *from, char *to, int offset,
799 int len, int odd, struct sk_buff *skb),
800 void *from, int length, int transhdrlen,
801 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
803 struct inet_sock *inet = inet_sk(sk);
804 struct sk_buff *skb;
805
Herbert Xu07df5292011-03-01 23:00:58 -0800806 struct ip_options *opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 int hh_len;
808 int exthdrlen;
809 int mtu;
810 int copy;
811 int err;
812 int offset = 0;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100813 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 int csummode = CHECKSUM_NONE;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000815 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Steffen Klassert96d73032011-06-05 20:48:47 +0000817 skb = skb_peek_tail(queue);
818
819 exthdrlen = !skb ? rt->dst.header_len : 0;
Herbert Xu07df5292011-03-01 23:00:58 -0800820 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Changli Gaod8d1f302010-06-10 23:31:35 -0700822 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
824 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
825 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100826 maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ?
827 mtu : 0xFFFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100829 if (cork->length + length > maxnonfragsize - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -0700830 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +0100831 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 return -EMSGSIZE;
833 }
834
835 /*
836 * transhdrlen > 0 means that this is the first fragment and we wish
837 * it won't be fragmented in the future.
838 */
839 if (transhdrlen &&
840 length + fragheaderlen <= mtu &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700841 rt->dst.dev->features & NETIF_F_V4_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700843 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Herbert Xu1470ddf2011-03-01 02:36:47 +0000845 cork->length += length;
Herbert Xu26cde9f2010-06-15 01:52:25 +0000846 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
Kostya Bbe9164e2008-04-29 22:36:30 -0700847 (sk->sk_protocol == IPPROTO_UDP) &&
Steffen Klassertc1460662011-06-29 23:19:32 +0000848 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
Herbert Xu1470ddf2011-03-01 02:36:47 +0000849 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
850 hh_len, fragheaderlen, transhdrlen,
Bill Sommerfeldd9be4f72011-07-19 15:22:33 +0000851 maxfraglen, flags);
Patrick McHardybaa829d2006-03-12 20:35:12 -0800852 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700853 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700854 return 0;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 /* So, what's going on in the loop below?
858 *
859 * We use calculated fragment length to generate chained skb,
860 * each of segments is IP fragment ready for sending to network after
861 * adding appropriate IP header.
862 */
863
Herbert Xu26cde9f2010-06-15 01:52:25 +0000864 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 goto alloc_new_skb;
866
867 while (length > 0) {
868 /* Check if the remaining data fits into current packet. */
869 copy = mtu - skb->len;
870 if (copy < length)
871 copy = maxfraglen - skb->len;
872 if (copy <= 0) {
873 char *data;
874 unsigned int datalen;
875 unsigned int fraglen;
876 unsigned int fraggap;
877 unsigned int alloclen;
878 struct sk_buff *skb_prev;
879alloc_new_skb:
880 skb_prev = skb;
881 if (skb_prev)
882 fraggap = skb_prev->len - maxfraglen;
883 else
884 fraggap = 0;
885
886 /*
887 * If remaining data exceeds the mtu,
888 * we know we need more fragment(s).
889 */
890 datalen = length + fraggap;
891 if (datalen > mtu - fragheaderlen)
892 datalen = maxfraglen - fragheaderlen;
893 fraglen = datalen + fragheaderlen;
894
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900895 if ((flags & MSG_MORE) &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700896 !(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 alloclen = mtu;
898 else
Eric Dumazet59104f02010-09-20 20:16:27 +0000899 alloclen = fraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Steffen Klassert353e5c92011-06-22 01:05:37 +0000901 alloclen += exthdrlen;
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 /* The last fragment gets additional space at tail.
904 * Note, with MSG_MORE we overallocate on fragments,
905 * because we have no idea what fragment will be
906 * the last.
907 */
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000908 if (datalen == length + fraggap)
Changli Gaod8d1f302010-06-10 23:31:35 -0700909 alloclen += rt->dst.trailer_len;
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900912 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 alloclen + hh_len + 15,
914 (flags & MSG_DONTWAIT), &err);
915 } else {
916 skb = NULL;
917 if (atomic_read(&sk->sk_wmem_alloc) <=
918 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900919 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 alloclen + hh_len + 15, 1,
921 sk->sk_allocation);
922 if (unlikely(skb == NULL))
923 err = -ENOBUFS;
Patrick Ohly51f31ca2009-02-12 05:03:39 +0000924 else
925 /* only the initial fragment is
926 time stamped */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000927 cork->tx_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 }
929 if (skb == NULL)
930 goto error;
931
932 /*
933 * Fill in the control structures
934 */
935 skb->ip_summed = csummode;
936 skb->csum = 0;
937 skb_reserve(skb, hh_len);
Herbert Xu1470ddf2011-03-01 02:36:47 +0000938 skb_shinfo(skb)->tx_flags = cork->tx_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
940 /*
941 * Find where to start putting bytes.
942 */
Steffen Klassert353e5c92011-06-22 01:05:37 +0000943 data = skb_put(skb, fraglen + exthdrlen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -0300944 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700945 skb->transport_header = (skb->network_header +
946 fragheaderlen);
Steffen Klassert353e5c92011-06-22 01:05:37 +0000947 data += fragheaderlen + exthdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949 if (fraggap) {
950 skb->csum = skb_copy_and_csum_bits(
951 skb_prev, maxfraglen,
952 data + transhdrlen, fraggap, 0);
953 skb_prev->csum = csum_sub(skb_prev->csum,
954 skb->csum);
955 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700956 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
958
959 copy = datalen - transhdrlen - fraggap;
960 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
961 err = -EFAULT;
962 kfree_skb(skb);
963 goto error;
964 }
965
966 offset += copy;
967 length -= datalen - fraggap;
968 transhdrlen = 0;
969 exthdrlen = 0;
970 csummode = CHECKSUM_NONE;
971
972 /*
973 * Put the packet on the pending queue.
974 */
Herbert Xu1470ddf2011-03-01 02:36:47 +0000975 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 continue;
977 }
978
979 if (copy > length)
980 copy = length;
981
Changli Gaod8d1f302010-06-10 23:31:35 -0700982 if (!(rt->dst.dev->features&NETIF_F_SG)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 unsigned int off;
984
985 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900986 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 offset, copy, off, skb) < 0) {
988 __skb_trim(skb, off);
989 err = -EFAULT;
990 goto error;
991 }
992 } else {
993 int i = skb_shinfo(skb)->nr_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Eric Dumazet5640f762012-09-23 23:04:42 +0000995 err = -ENOMEM;
996 if (!sk_page_frag_refill(sk, pfrag))
997 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Eric Dumazet5640f762012-09-23 23:04:42 +0000999 if (!skb_can_coalesce(skb, i, pfrag->page,
1000 pfrag->offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 err = -EMSGSIZE;
Eric Dumazet5640f762012-09-23 23:04:42 +00001002 if (i == MAX_SKB_FRAGS)
1003 goto error;
1004
1005 __skb_fill_page_desc(skb, i, pfrag->page,
1006 pfrag->offset, 0);
1007 skb_shinfo(skb)->nr_frags = ++i;
1008 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001010 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1011 if (getfrag(from,
1012 page_address(pfrag->page) + pfrag->offset,
1013 offset, copy, skb->len, skb) < 0)
1014 goto error_efault;
1015
1016 pfrag->offset += copy;
1017 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 skb->len += copy;
1019 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001020 skb->truesize += copy;
1021 atomic_add(copy, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
1023 offset += copy;
1024 length -= copy;
1025 }
1026
1027 return 0;
1028
Eric Dumazet5640f762012-09-23 23:04:42 +00001029error_efault:
1030 err = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031error:
Herbert Xu1470ddf2011-03-01 02:36:47 +00001032 cork->length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001033 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001034 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035}
1036
Herbert Xu1470ddf2011-03-01 02:36:47 +00001037static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1038 struct ipcm_cookie *ipc, struct rtable **rtp)
1039{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001040 struct ip_options_rcu *opt;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001041 struct rtable *rt;
1042
1043 /*
1044 * setup for corking.
1045 */
1046 opt = ipc->opt;
1047 if (opt) {
1048 if (cork->opt == NULL) {
1049 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1050 sk->sk_allocation);
1051 if (unlikely(cork->opt == NULL))
1052 return -ENOBUFS;
1053 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001054 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001055 cork->flags |= IPCORK_OPT;
1056 cork->addr = ipc->addr;
1057 }
1058 rt = *rtp;
1059 if (unlikely(!rt))
1060 return -EFAULT;
1061 /*
1062 * We steal reference to this route, caller should not release it
1063 */
1064 *rtp = NULL;
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001065 cork->fragsize = ip_sk_use_pmtu(sk) ?
1066 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001067 cork->dst = &rt->dst;
1068 cork->length = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001069 cork->ttl = ipc->ttl;
1070 cork->tos = ipc->tos;
1071 cork->priority = ipc->priority;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001072 cork->tx_flags = ipc->tx_flags;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001073
1074 return 0;
1075}
1076
1077/*
1078 * ip_append_data() and ip_append_page() can make one large IP datagram
1079 * from many pieces of data. Each pieces will be holded on the socket
1080 * until ip_push_pending_frames() is called. Each piece can be a page
1081 * or non-page data.
1082 *
1083 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1084 * this interface potentially.
1085 *
1086 * LATER: length must be adjusted by pad at tail, when it is required.
1087 */
David S. Millerf5fca602011-05-08 17:24:10 -07001088int ip_append_data(struct sock *sk, struct flowi4 *fl4,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001089 int getfrag(void *from, char *to, int offset, int len,
1090 int odd, struct sk_buff *skb),
1091 void *from, int length, int transhdrlen,
1092 struct ipcm_cookie *ipc, struct rtable **rtp,
1093 unsigned int flags)
1094{
1095 struct inet_sock *inet = inet_sk(sk);
1096 int err;
1097
1098 if (flags&MSG_PROBE)
1099 return 0;
1100
1101 if (skb_queue_empty(&sk->sk_write_queue)) {
David S. Millerbdc712b2011-05-06 15:02:07 -07001102 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001103 if (err)
1104 return err;
1105 } else {
1106 transhdrlen = 0;
1107 }
1108
Eric Dumazet5640f762012-09-23 23:04:42 +00001109 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1110 sk_page_frag(sk), getfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001111 from, length, transhdrlen, flags);
1112}
1113
David S. Millerf5fca602011-05-08 17:24:10 -07001114ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 int offset, size_t size, int flags)
1116{
1117 struct inet_sock *inet = inet_sk(sk);
1118 struct sk_buff *skb;
1119 struct rtable *rt;
1120 struct ip_options *opt = NULL;
David S. Millerbdc712b2011-05-06 15:02:07 -07001121 struct inet_cork *cork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 int hh_len;
1123 int mtu;
1124 int len;
1125 int err;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001126 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128 if (inet->hdrincl)
1129 return -EPERM;
1130
1131 if (flags&MSG_PROBE)
1132 return 0;
1133
1134 if (skb_queue_empty(&sk->sk_write_queue))
1135 return -EINVAL;
1136
David S. Millerbdc712b2011-05-06 15:02:07 -07001137 cork = &inet->cork.base;
1138 rt = (struct rtable *)cork->dst;
1139 if (cork->flags & IPCORK_OPT)
1140 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Changli Gaod8d1f302010-06-10 23:31:35 -07001142 if (!(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 return -EOPNOTSUPP;
1144
Changli Gaod8d1f302010-06-10 23:31:35 -07001145 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
David S. Millerbdc712b2011-05-06 15:02:07 -07001146 mtu = cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147
1148 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1149 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001150 maxnonfragsize = (inet->pmtudisc >= IP_PMTUDISC_DO) ?
1151 mtu : 0xFFFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001153 if (cork->length + size > maxnonfragsize - fragheaderlen) {
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +01001154 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1155 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return -EMSGSIZE;
1157 }
1158
1159 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1160 return -EINVAL;
1161
David S. Millerbdc712b2011-05-06 15:02:07 -07001162 cork->length += size;
Herbert Xu26cde9f2010-06-15 01:52:25 +00001163 if ((size + skb->len > mtu) &&
1164 (sk->sk_protocol == IPPROTO_UDP) &&
Changli Gaod8d1f302010-06-10 23:31:35 -07001165 (rt->dst.dev->features & NETIF_F_UFO)) {
Herbert Xu79671682006-06-22 02:40:14 -07001166 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001167 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001168 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 while (size > 0) {
1172 int i;
1173
Herbert Xu89114af2006-07-08 13:34:32 -07001174 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001175 len = size;
1176 else {
1177
1178 /* Check if the remaining data fits into current packet. */
1179 len = mtu - skb->len;
1180 if (len < size)
1181 len = maxfraglen - skb->len;
1182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if (len <= 0) {
1184 struct sk_buff *skb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 int alloclen;
1186
1187 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001188 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 alloclen = fragheaderlen + hh_len + fraggap + 15;
1191 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1192 if (unlikely(!skb)) {
1193 err = -ENOBUFS;
1194 goto error;
1195 }
1196
1197 /*
1198 * Fill in the control structures
1199 */
1200 skb->ip_summed = CHECKSUM_NONE;
1201 skb->csum = 0;
1202 skb_reserve(skb, hh_len);
1203
1204 /*
1205 * Find where to start putting bytes.
1206 */
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001207 skb_put(skb, fragheaderlen + fraggap);
Arnaldo Carvalho de Melo2ca9e6f2007-03-10 19:15:25 -03001208 skb_reset_network_header(skb);
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -07001209 skb->transport_header = (skb->network_header +
1210 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 if (fraggap) {
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001212 skb->csum = skb_copy_and_csum_bits(skb_prev,
1213 maxfraglen,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001214 skb_transport_header(skb),
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001215 fraggap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 skb_prev->csum = csum_sub(skb_prev->csum,
1217 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001218 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
1220
1221 /*
1222 * Put the packet on the pending queue.
1223 */
1224 __skb_queue_tail(&sk->sk_write_queue, skb);
1225 continue;
1226 }
1227
1228 i = skb_shinfo(skb)->nr_frags;
1229 if (len > size)
1230 len = size;
1231 if (skb_can_coalesce(skb, i, page, offset)) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00001232 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 } else if (i < MAX_SKB_FRAGS) {
1234 get_page(page);
1235 skb_fill_page_desc(skb, i, page, offset, len);
1236 } else {
1237 err = -EMSGSIZE;
1238 goto error;
1239 }
1240
1241 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001242 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 csum = csum_page(page, offset, len);
1244 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1245 }
1246
1247 skb->len += len;
1248 skb->data_len += len;
David S. Miller1e34a112008-01-22 23:44:31 -08001249 skb->truesize += len;
1250 atomic_add(len, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 offset += len;
1252 size -= len;
1253 }
1254 return 0;
1255
1256error:
David S. Millerbdc712b2011-05-06 15:02:07 -07001257 cork->length -= size;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001258 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return err;
1260}
1261
Herbert Xu1470ddf2011-03-01 02:36:47 +00001262static void ip_cork_release(struct inet_cork *cork)
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001263{
Herbert Xu1470ddf2011-03-01 02:36:47 +00001264 cork->flags &= ~IPCORK_OPT;
1265 kfree(cork->opt);
1266 cork->opt = NULL;
1267 dst_release(cork->dst);
1268 cork->dst = NULL;
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001269}
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271/*
1272 * Combined all pending IP fragments on the socket as one IP datagram
1273 * and push them out.
1274 */
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001275struct sk_buff *__ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001276 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001277 struct sk_buff_head *queue,
1278 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279{
1280 struct sk_buff *skb, *tmp_skb;
1281 struct sk_buff **tail_skb;
1282 struct inet_sock *inet = inet_sk(sk);
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001283 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 struct ip_options *opt = NULL;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001285 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001287 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 __u8 ttl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Herbert Xu1470ddf2011-03-01 02:36:47 +00001290 if ((skb = __skb_dequeue(queue)) == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 goto out;
1292 tail_skb = &(skb_shinfo(skb)->frag_list);
1293
1294 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001295 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001296 __skb_pull(skb, skb_network_offset(skb));
Herbert Xu1470ddf2011-03-01 02:36:47 +00001297 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001298 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 *tail_skb = tmp_skb;
1300 tail_skb = &(tmp_skb->next);
1301 skb->len += tmp_skb->len;
1302 skb->data_len += tmp_skb->len;
1303 skb->truesize += tmp_skb->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 tmp_skb->destructor = NULL;
1305 tmp_skb->sk = NULL;
1306 }
1307
1308 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1309 * to fragment the frame generated here. No matter, what transforms
1310 * how transforms change size of the packet, it will come out.
1311 */
John Heffner628a5c52007-04-20 15:53:27 -07001312 if (inet->pmtudisc < IP_PMTUDISC_DO)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 skb->local_df = 1;
1314
1315 /* DF bit is set when we want to see DF on outgoing frames.
1316 * If local_df is set too, we still allow to fragment this frame
1317 * locally. */
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001318 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1319 inet->pmtudisc == IP_PMTUDISC_PROBE ||
Changli Gaod8d1f302010-06-10 23:31:35 -07001320 (skb->len <= dst_mtu(&rt->dst) &&
1321 ip_dont_fragment(sk, &rt->dst)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 df = htons(IP_DF);
1323
Herbert Xu1470ddf2011-03-01 02:36:47 +00001324 if (cork->flags & IPCORK_OPT)
1325 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Francesco Fuscoaa661582013-09-24 15:43:09 +02001327 if (cork->ttl != 0)
1328 ttl = cork->ttl;
1329 else if (rt->rt_type == RTN_MULTICAST)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 ttl = inet->mc_ttl;
1331 else
Changli Gaod8d1f302010-06-10 23:31:35 -07001332 ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Ansis Atteka749154a2013-09-18 15:29:52 -07001334 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 iph->version = 4;
1336 iph->ihl = 5;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001337 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 iph->frag_off = df;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 iph->ttl = ttl;
1340 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +00001341 ip_copy_addrs(iph, fl4);
Ansis Atteka703133d2013-09-18 15:29:53 -07001342 ip_select_ident(skb, &rt->dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
David S. Miller22f728f2011-05-13 17:21:27 -04001344 if (opt) {
1345 iph->ihl += opt->optlen>>2;
1346 ip_options_build(skb, opt, cork->addr, rt, 0);
1347 }
1348
Francesco Fuscoaa661582013-09-24 15:43:09 +02001349 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001350 skb->mark = sk->sk_mark;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001351 /*
1352 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1353 * on dst refcount
1354 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001355 cork->dst = NULL;
Changli Gaod8d1f302010-06-10 23:31:35 -07001356 skb_dst_set(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
David L Stevens96793b42007-09-17 09:57:33 -07001358 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b002008-07-14 23:00:43 -07001359 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001360 skb_transport_header(skb))->type);
1361
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001362 ip_cork_release(cork);
1363out:
1364 return skb;
1365}
1366
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001367int ip_send_skb(struct net *net, struct sk_buff *skb)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001368{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001369 int err;
1370
Herbert Xuc439cb22008-01-11 19:14:00 -08001371 err = ip_local_out(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (err) {
1373 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001374 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 if (err)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001376 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380}
1381
David S. Miller77968b72011-05-08 17:12:19 -07001382int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1470ddf2011-03-01 02:36:47 +00001383{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001384 struct sk_buff *skb;
1385
David S. Miller77968b72011-05-08 17:12:19 -07001386 skb = ip_finish_skb(sk, fl4);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001387 if (!skb)
1388 return 0;
1389
1390 /* Netfilter gets whole the not fragmented skb. */
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001391 return ip_send_skb(sock_net(sk), skb);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394/*
1395 * Throw away all pending data on the socket.
1396 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001397static void __ip_flush_pending_frames(struct sock *sk,
1398 struct sk_buff_head *queue,
1399 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 struct sk_buff *skb;
1402
Herbert Xu1470ddf2011-03-01 02:36:47 +00001403 while ((skb = __skb_dequeue_tail(queue)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 kfree_skb(skb);
1405
Herbert Xu1470ddf2011-03-01 02:36:47 +00001406 ip_cork_release(cork);
1407}
1408
1409void ip_flush_pending_frames(struct sock *sk)
1410{
David S. Millerbdc712b2011-05-06 15:02:07 -07001411 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001414struct sk_buff *ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001415 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001416 int getfrag(void *from, char *to, int offset,
1417 int len, int odd, struct sk_buff *skb),
1418 void *from, int length, int transhdrlen,
1419 struct ipcm_cookie *ipc, struct rtable **rtp,
1420 unsigned int flags)
1421{
David S. Millerb80d7222011-05-06 15:06:01 -07001422 struct inet_cork cork;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001423 struct sk_buff_head queue;
1424 int err;
1425
1426 if (flags & MSG_PROBE)
1427 return NULL;
1428
1429 __skb_queue_head_init(&queue);
1430
David S. Millerb80d7222011-05-06 15:06:01 -07001431 cork.flags = 0;
1432 cork.addr = 0;
David S. Miller70652722011-05-06 16:01:15 -07001433 cork.opt = NULL;
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001434 err = ip_setup_cork(sk, &cork, ipc, rtp);
1435 if (err)
1436 return ERR_PTR(err);
1437
Eric Dumazet5640f762012-09-23 23:04:42 +00001438 err = __ip_append_data(sk, fl4, &queue, &cork,
1439 &current->task_frag, getfrag,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001440 from, length, transhdrlen, flags);
1441 if (err) {
1442 __ip_flush_pending_frames(sk, &queue, &cork);
1443 return ERR_PTR(err);
1444 }
1445
David S. Miller77968b72011-05-08 17:12:19 -07001446 return __ip_make_skb(sk, fl4, &queue, &cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001447}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449/*
1450 * Fetch data from kernel space and fill in checksum if needed.
1451 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001452static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 int len, int odd, struct sk_buff *skb)
1454{
Al Viro50842052006-11-14 21:36:34 -08001455 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1458 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001459 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
1461
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001462/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 * Generic function to send a packet as reply to another packet.
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001464 * Used to send some TCP resets/acks so far.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 *
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001466 * Use a fake percpu inet socket to avoid false sharing and contention.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 */
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001468static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1469 .sk = {
1470 .__sk_common = {
1471 .skc_refcnt = ATOMIC_INIT(1),
1472 },
1473 .sk_wmem_alloc = ATOMIC_INIT(1),
1474 .sk_allocation = GFP_ATOMIC,
1475 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
1476 },
Eric Dumazet0980e562012-07-20 22:28:51 +00001477 .pmtudisc = IP_PMTUDISC_WANT,
1478 .uc_ttl = -1,
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001479};
1480
1481void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
David S. Miller70e73412012-06-28 03:21:41 -07001482 __be32 saddr, const struct ip_reply_arg *arg,
1483 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001485 struct ip_options_data replyopts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 struct ipcm_cookie ipc;
David S. Miller77968b72011-05-08 17:12:19 -07001487 struct flowi4 fl4;
Eric Dumazet511c3f92009-06-02 05:14:27 +00001488 struct rtable *rt = skb_rtable(skb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001489 struct sk_buff *nskb;
1490 struct sock *sk;
1491 struct inet_sock *inet;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001493 if (ip_options_echo(&replyopts.opt.opt, skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 return;
1495
David S. Miller0a5ebb82011-05-09 13:22:43 -07001496 ipc.addr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 ipc.opt = NULL;
Oliver Hartkopp2244d072010-08-17 08:59:14 +00001498 ipc.tx_flags = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001499 ipc.ttl = 0;
1500 ipc.tos = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001502 if (replyopts.opt.opt.optlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 ipc.opt = &replyopts.opt;
1504
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001505 if (replyopts.opt.opt.srr)
1506 daddr = replyopts.opt.opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 }
1508
David S. Miller77968b72011-05-08 17:12:19 -07001509 flowi4_init_output(&fl4, arg->bound_dev_if, 0,
Eric Dumazet66b13d92011-10-24 03:06:21 -04001510 RT_TOS(arg->tos),
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001511 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
David S. Miller77968b72011-05-08 17:12:19 -07001512 ip_reply_arg_flowi_flags(arg),
David S. Miller70e73412012-06-28 03:21:41 -07001513 daddr, saddr,
David S. Miller77968b72011-05-08 17:12:19 -07001514 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1515 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001516 rt = ip_route_output_key(net, &fl4);
David S. Miller77968b72011-05-08 17:12:19 -07001517 if (IS_ERR(rt))
1518 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001520 inet = &get_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Eric Dumazet66b13d92011-10-24 03:06:21 -04001522 inet->tos = arg->tos;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001523 sk = &inet->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001525 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001526 sk->sk_bound_dev_if = arg->bound_dev_if;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001527 sock_net_set(sk, net);
1528 __skb_queue_head_init(&sk->sk_write_queue);
1529 sk->sk_sndbuf = sysctl_wmem_default;
David S. Millerf5fca602011-05-08 17:24:10 -07001530 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
Eric Dumazet2e77d892008-11-24 15:52:46 -08001531 &ipc, &rt, MSG_DONTWAIT);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001532 nskb = skb_peek(&sk->sk_write_queue);
1533 if (nskb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 if (arg->csumoffset >= 0)
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001535 *((__sum16 *)skb_transport_header(nskb) +
1536 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001537 arg->csum));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001538 nskb->ip_summed = CHECKSUM_NONE;
Eric Dumazet3a7c3842012-08-09 13:56:06 +00001539 skb_orphan(nskb);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001540 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
David S. Miller77968b72011-05-08 17:12:19 -07001541 ip_push_pending_frames(sk, &fl4);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 }
1543
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001544 put_cpu_var(unicast_sock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
1546 ip_rt_put(rt);
1547}
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549void __init ip_init(void)
1550{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 ip_rt_init();
1552 inet_initpeers();
1553
1554#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1555 igmp_mc_proc_init();
1556#endif
1557}