blob: ac880beda8a76e971b1b64d743319aca7dddd972 [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
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080045#include <linux/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>
Roopa Prabhu14972cb2016-08-24 20:10:43 -070076#include <net/lwtunnel.h>
Daniel Mack33b48672016-11-23 16:52:29 +010077#include <linux/bpf-cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#include <linux/igmp.h>
79#include <linux/netfilter_ipv4.h>
80#include <linux/netfilter_bridge.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <linux/netlink.h>
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -070082#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Eric W. Biederman694869b2015-06-12 21:55:31 -050084static int
85ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
86 unsigned int mtu,
87 int (*output)(struct net *, struct sock *, struct sk_buff *));
Andy Zhou49d16b22015-05-15 14:15:37 -070088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/* Generate a checksum for an outgoing IP datagram. */
Denis Efremov2fbd9672013-05-08 23:19:42 +000090void ip_send_check(struct iphdr *iph)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 iph->check = 0;
93 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
94}
Eric Dumazet4bc2f182010-07-09 21:22:10 +000095EXPORT_SYMBOL(ip_send_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Eric W. Biedermancf91a992015-10-07 16:48:45 -050097int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
Herbert Xuc439cb22008-01-11 19:14:00 -080098{
99 struct iphdr *iph = ip_hdr(skb);
100
101 iph->tot_len = htons(skb->len);
102 ip_send_check(iph);
David Aherna8e3e1a2016-09-10 12:09:53 -0700103
104 /* if egress device is enslaved to an L3 master device pass the
105 * skb to its handler for processing
106 */
107 skb = l3mdev_ip_out(sk, skb);
108 if (unlikely(!skb))
109 return 0;
110
Eli Cooperf4180432016-12-01 10:05:10 +0800111 skb->protocol = htons(ETH_P_IP);
112
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500113 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
114 net, sk, skb, NULL, skb_dst(skb)->dev,
Eric W. Biederman13206b62015-10-07 16:48:35 -0500115 dst_output);
David Miller7026b1d2015-04-05 22:19:04 -0400116}
117
Eric W. Biederman33224b12015-10-07 16:48:46 -0500118int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
Herbert Xuc439cb22008-01-11 19:14:00 -0800119{
120 int err;
121
Eric W. Biedermancf91a992015-10-07 16:48:45 -0500122 err = __ip_local_out(net, sk, skb);
Herbert Xuc439cb22008-01-11 19:14:00 -0800123 if (likely(err == 1))
Eric W. Biederman13206b62015-10-07 16:48:35 -0500124 err = dst_output(net, sk, skb);
Herbert Xuc439cb22008-01-11 19:14:00 -0800125
126 return err;
127}
Eric W. Biedermane2cb77d2015-10-07 16:48:38 -0500128EXPORT_SYMBOL_GPL(ip_local_out);
Herbert Xuc439cb22008-01-11 19:14:00 -0800129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
131{
132 int ttl = inet->uc_ttl;
133
134 if (ttl < 0)
David S. Miller323e1262010-12-12 21:55:08 -0800135 ttl = ip4_dst_hoplimit(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 return ttl;
137}
138
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900139/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 * Add an ip header to a skbuff and send it out.
141 *
142 */
Eric Dumazetcfe673b2015-09-25 07:39:16 -0700143int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000144 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
146 struct inet_sock *inet = inet_sk(sk);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000147 struct rtable *rt = skb_rtable(skb);
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500148 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 struct iphdr *iph;
150
151 /* Build the IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000152 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300153 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700154 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 iph->version = 4;
156 iph->ihl = 5;
157 iph->tos = inet->tos;
Changli Gaod8d1f302010-06-10 23:31:35 -0700158 iph->ttl = ip_select_ttl(inet, &rt->dst);
David S. Millerdd927a22011-05-04 12:03:30 -0700159 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
160 iph->saddr = saddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 iph->protocol = sk->sk_protocol;
Eric Dumazetcfe673b2015-09-25 07:39:16 -0700162 if (ip_dont_fragment(sk, &rt->dst)) {
163 iph->frag_off = htons(IP_DF);
164 iph->id = 0;
165 } else {
166 iph->frag_off = 0;
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500167 __ip_select_ident(net, iph, 1);
Eric Dumazetcfe673b2015-09-25 07:39:16 -0700168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000170 if (opt && opt->opt.optlen) {
171 iph->ihl += opt->opt.optlen>>2;
172 ip_options_build(skb, &opt->opt, daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175 skb->priority = sk->sk_priority;
Jamal Hadi Salime05a90e2017-07-03 09:51:50 -0400176 if (!skb->mark)
177 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179 /* Send it out. */
Eric W. Biederman33224b12015-10-07 16:48:46 -0500180 return ip_local_out(net, skb->sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -0700182EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
183
Eric W. Biederman694869b2015-06-12 21:55:31 -0500184static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
Eric Dumazetadf30902009-06-02 05:19:30 +0000186 struct dst_entry *dst = skb_dst(skb);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700187 struct rtable *rt = (struct rtable *)dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 struct net_device *dev = dst->dev;
Chuck Leverc2636b42007-10-23 21:07:32 -0700189 unsigned int hh_len = LL_RESERVED_SPACE(dev);
David S. Millerf6b72b622011-07-14 07:53:20 -0700190 struct neighbour *neigh;
David Ahern5c9f7c12019-04-05 16:30:34 -0700191 bool is_v6gw = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Neil Hormanedf391f2009-04-27 02:45:02 -0700193 if (rt->rt_type == RTN_MULTICAST) {
Eric W. Biederman4ba1bf42015-09-15 20:04:01 -0500194 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
Neil Hormanedf391f2009-04-27 02:45:02 -0700195 } else if (rt->rt_type == RTN_BROADCAST)
Eric W. Biederman4ba1bf42015-09-15 20:04:01 -0500196 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
Mitsuru Chinen80787eb2007-04-30 00:48:20 -0700197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 /* Be paranoid, rather than too clever. */
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -0700199 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 struct sk_buff *skb2;
201
202 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
Ian Morris51456b22015-04-03 09:17:26 +0100203 if (!skb2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 kfree_skb(skb);
205 return -ENOMEM;
206 }
207 if (skb->sk)
208 skb_set_owner_w(skb2, skb->sk);
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000209 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 skb = skb2;
211 }
212
Roopa Prabhu14972cb2016-08-24 20:10:43 -0700213 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
214 int res = lwtunnel_xmit(skb);
215
216 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
217 return res;
218 }
219
David S. Millera263b302012-07-02 02:02:15 -0700220 rcu_read_lock_bh();
David Ahern5c9f7c12019-04-05 16:30:34 -0700221 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
Vasiliy Kulikov9871f1a2012-08-06 03:55:29 +0000222 if (!IS_ERR(neigh)) {
Julian Anastasov4ff06202017-02-06 23:14:12 +0200223 int res;
224
225 sock_confirm_neigh(skb, neigh);
David Ahern5c9f7c12019-04-05 16:30:34 -0700226 /* if crossing protocols, can not use the cached header */
227 res = neigh_output(neigh, skb, is_v6gw);
David S. Millera263b302012-07-02 02:02:15 -0700228 rcu_read_unlock_bh();
Eric Dumazetf2c31e32011-07-29 19:00:53 +0000229 return res;
230 }
David S. Millera263b302012-07-02 02:02:15 -0700231 rcu_read_unlock_bh();
David S. Miller05e3aa02011-07-16 17:26:00 -0700232
Joe Perchese87cc472012-05-13 21:56:26 +0000233 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
234 __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 kfree_skb(skb);
236 return -EINVAL;
237}
238
Eric W. Biederman694869b2015-06-12 21:55:31 -0500239static int ip_finish_output_gso(struct net *net, struct sock *sk,
240 struct sk_buff *skb, unsigned int mtu)
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200241{
242 netdev_features_t features;
243 struct sk_buff *segs;
244 int ret = 0;
245
Lance Richardson9ee6c5d2016-11-02 16:36:17 -0400246 /* common case: seglen is <= mtu
Shmulik Ladkani359ebda2016-07-18 14:49:33 +0300247 */
Daniel Axtens779b7932018-03-01 17:13:37 +1100248 if (skb_gso_validate_network_len(skb, mtu))
Eric W. Biederman694869b2015-06-12 21:55:31 -0500249 return ip_finish_output2(net, sk, skb);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200250
Lance Richardson0ace81e2016-11-09 15:04:39 -0500251 /* Slowpath - GSO segment length exceeds the egress MTU.
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200252 *
Lance Richardson0ace81e2016-11-09 15:04:39 -0500253 * This can happen in several cases:
254 * - Forwarding of a TCP GRO skb, when DF flag is not set.
255 * - Forwarding of an skb that arrived on a virtualization interface
256 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
257 * stack.
258 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
259 * interface with a smaller MTU.
260 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
261 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
262 * insufficent MTU.
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200263 */
264 features = netif_skb_features(skb);
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +0300265 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200266 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
Florian Westphal330966e2014-10-20 13:49:17 +0200267 if (IS_ERR_OR_NULL(segs)) {
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200268 kfree_skb(skb);
269 return -ENOMEM;
270 }
271
272 consume_skb(skb);
273
274 do {
275 struct sk_buff *nskb = segs->next;
276 int err;
277
David S. Millera8305bf2018-07-29 20:42:53 -0700278 skb_mark_not_on_list(segs);
Eric W. Biederman694869b2015-06-12 21:55:31 -0500279 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200280
281 if (err && ret == 0)
282 ret = err;
283 segs = nskb;
284 } while (segs);
285
286 return ret;
287}
288
Eric W. Biederman0c4b51f2015-09-15 20:04:18 -0500289static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
Florian Westphalc5501eb2015-05-22 16:32:50 +0200291 unsigned int mtu;
Daniel Mack33b48672016-11-23 16:52:29 +0100292 int ret;
293
294 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
295 if (ret) {
296 kfree_skb(skb);
297 return ret;
298 }
Florian Westphalc5501eb2015-05-22 16:32:50 +0200299
Patrick McHardy5c901da2006-01-06 23:05:36 -0800300#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
301 /* Policy lookup after SNAT yielded a new policy */
Ian Morris00db4122015-04-03 09:17:27 +0100302 if (skb_dst(skb)->xfrm) {
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800303 IPCB(skb)->flags |= IPSKB_REROUTED;
Eric W. Biederman13206b62015-10-07 16:48:35 -0500304 return dst_output(net, sk, skb);
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800305 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800306#endif
Shmulik Ladkanifedbb6b42016-06-29 21:47:03 +0300307 mtu = ip_skb_dst_mtu(sk, skb);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200308 if (skb_is_gso(skb))
Eric W. Biederman694869b2015-06-12 21:55:31 -0500309 return ip_finish_output_gso(net, sk, skb, mtu);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200310
Florian Westphald6b915e2015-05-22 16:32:51 +0200311 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
Eric W. Biederman694869b2015-06-12 21:55:31 -0500312 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
Florian Westphalc7ba65d2014-05-05 15:00:43 +0200313
Eric W. Biederman694869b2015-06-12 21:55:31 -0500314 return ip_finish_output2(net, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
Daniel Mack33b48672016-11-23 16:52:29 +0100317static int ip_mc_finish_output(struct net *net, struct sock *sk,
318 struct sk_buff *skb)
319{
320 int ret;
321
322 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
323 if (ret) {
324 kfree_skb(skb);
325 return ret;
326 }
327
328 return dev_loopback_xmit(net, sk, skb);
329}
330
Eric W. Biedermanede20592015-10-07 16:48:47 -0500331int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Eric Dumazet511c3f92009-06-02 05:14:27 +0000333 struct rtable *rt = skb_rtable(skb);
Changli Gaod8d1f302010-06-10 23:31:35 -0700334 struct net_device *dev = rt->dst.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /*
337 * If the indicated interface is up and running, send the packet.
338 */
Eric W. Biederman88f5cc22015-09-15 20:03:57 -0500339 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 skb->dev = dev;
342 skb->protocol = htons(ETH_P_IP);
343
344 /*
345 * Multicasts are looped back for other local users
346 */
347
348 if (rt->rt_flags&RTCF_MULTICAST) {
Octavian Purdila7ad68482010-01-06 20:37:01 -0800349 if (sk_mc_loop(sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350#ifdef CONFIG_IP_MROUTE
351 /* Small optimization: do not loopback not local frames,
352 which returned after forwarding; they will be dropped
353 by ip_mr_input in any case.
354 Note, that local frames are looped back to be delivered
355 to local recipients.
356
357 This check is duplicated in ip_mr_input at the moment.
358 */
Joe Perches9d4fb272009-11-23 10:41:23 -0800359 &&
360 ((rt->rt_flags & RTCF_LOCAL) ||
361 !(IPCB(skb)->flags & IPSKB_FORWARDED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362#endif
Joe Perches9d4fb272009-11-23 10:41:23 -0800363 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
365 if (newskb)
Jan Engelhardt9bbc7682010-03-23 04:07:29 +0100366 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500367 net, sk, newskb, NULL, newskb->dev,
Daniel Mack33b48672016-11-23 16:52:29 +0100368 ip_mc_finish_output);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 }
370
371 /* Multicasts with ttl 0 must not go beyond the host */
372
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700373 if (ip_hdr(skb)->ttl == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 kfree_skb(skb);
375 return 0;
376 }
377 }
378
379 if (rt->rt_flags&RTCF_BROADCAST) {
380 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
381 if (newskb)
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500382 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
383 net, sk, newskb, NULL, newskb->dev,
Daniel Mack33b48672016-11-23 16:52:29 +0100384 ip_mc_finish_output);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
386
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500387 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
388 net, sk, skb, NULL, skb->dev,
389 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800390 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391}
392
Eric W. Biedermanede20592015-10-07 16:48:47 -0500393int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394{
Eric Dumazetadf30902009-06-02 05:19:30 +0000395 struct net_device *dev = skb_dst(skb)->dev;
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800396
Eric W. Biederman88f5cc22015-09-15 20:03:57 -0500397 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800399 skb->dev = dev;
400 skb->protocol = htons(ETH_P_IP);
401
Eric W. Biederman29a26a52015-09-15 20:04:16 -0500402 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
403 net, sk, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900404 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800405 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
Eric Dumazet84f93072011-11-30 19:00:53 +0000408/*
409 * copy saddr and daddr, possibly using 64bit load/stores
410 * Equivalent to :
411 * iph->saddr = fl4->saddr;
412 * iph->daddr = fl4->daddr;
413 */
414static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
415{
416 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
417 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
418 memcpy(&iph->saddr, &fl4->saddr,
419 sizeof(fl4->saddr) + sizeof(fl4->daddr));
420}
421
Eric Dumazetb0270e92014-04-15 12:58:34 -0400422/* Note: skb->sk can be different from sk, in case of tunnels */
Xin Long69b9e1e2018-07-02 18:21:11 +0800423int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
424 __u8 tos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 struct inet_sock *inet = inet_sk(sk);
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500427 struct net *net = sock_net(sk);
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000428 struct ip_options_rcu *inet_opt;
David S. Millerb57ae012011-05-06 16:24:06 -0700429 struct flowi4 *fl4;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 struct rtable *rt;
431 struct iphdr *iph;
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000432 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
434 /* Skip all of this if the packet is already routed,
435 * f.e. by something like SCTP.
436 */
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000437 rcu_read_lock();
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000438 inet_opt = rcu_dereference(inet->inet_opt);
David S. Millerea4fc0d2011-05-06 22:30:20 -0700439 fl4 = &fl->u.ip4;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000440 rt = skb_rtable(skb);
Ian Morris00db4122015-04-03 09:17:27 +0100441 if (rt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 goto packet_routed;
443
444 /* Make sure we can route this packet. */
445 rt = (struct rtable *)__sk_dst_check(sk, 0);
Ian Morris51456b22015-04-03 09:17:26 +0100446 if (!rt) {
Al Viro3ca3c682006-09-27 18:28:07 -0700447 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
449 /* Use correct destination address if we have options. */
Eric Dumazetc720c7e82009-10-15 06:30:45 +0000450 daddr = inet->inet_daddr;
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000451 if (inet_opt && inet_opt->opt.srr)
452 daddr = inet_opt->opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
David S. Miller78fbfd82011-03-12 00:00:52 -0500454 /* If this fails, retransmit mechanism of transport layer will
455 * keep trying until route appears or the connection times
456 * itself out.
457 */
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500458 rt = ip_route_output_ports(net, fl4, sk,
David S. Miller78fbfd82011-03-12 00:00:52 -0500459 daddr, inet->inet_saddr,
460 inet->inet_dport,
461 inet->inet_sport,
462 sk->sk_protocol,
Xin Long69b9e1e2018-07-02 18:21:11 +0800463 RT_CONN_FLAGS_TOS(sk, tos),
David S. Miller78fbfd82011-03-12 00:00:52 -0500464 sk->sk_bound_dev_if);
465 if (IS_ERR(rt))
466 goto no_route;
Changli Gaod8d1f302010-06-10 23:31:35 -0700467 sk_setup_caps(sk, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 }
Changli Gaod8d1f302010-06-10 23:31:35 -0700469 skb_dst_set_noref(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471packet_routed:
David Ahern1550c172019-04-05 16:30:27 -0700472 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_gw_family)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 goto no_route;
474
475 /* OK, we know where to send it, allocate and build IP header. */
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000476 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
Arnaldo Carvalho de Melo8856dfa2007-03-10 19:40:39 -0300477 skb_reset_network_header(skb);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700478 iph = ip_hdr(skb);
Xin Long69b9e1e2018-07-02 18:21:11 +0800479 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
WANG Cong60ff7462014-05-04 16:39:18 -0700480 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 iph->frag_off = htons(IP_DF);
482 else
483 iph->frag_off = 0;
Changli Gaod8d1f302010-06-10 23:31:35 -0700484 iph->ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +0000486 ip_copy_addrs(iph, fl4);
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 /* Transport layer set skb->h.foo itself. */
489
Eric Dumazetf6d8bd02011-04-21 09:45:37 +0000490 if (inet_opt && inet_opt->opt.optlen) {
491 iph->ihl += inet_opt->opt.optlen >> 2;
492 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 }
494
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500495 ip_select_ident_segs(net, skb, sk,
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +0100496 skb_shinfo(skb)->gso_segs ?: 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Eric Dumazetb0270e92014-04-15 12:58:34 -0400498 /* TODO : should we use skb->sk here instead of sk ? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 skb->priority = sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -0800500 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Eric W. Biederman33224b12015-10-07 16:48:46 -0500502 res = ip_local_out(net, sk, skb);
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000503 rcu_read_unlock();
504 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506no_route:
Eric Dumazetab6e3fe2010-05-10 11:31:49 +0000507 rcu_read_unlock();
Eric W. Biederman77589ce2015-10-07 16:48:42 -0500508 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 kfree_skb(skb);
510 return -EHOSTUNREACH;
511}
Xin Long69b9e1e2018-07-02 18:21:11 +0800512EXPORT_SYMBOL(__ip_queue_xmit);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
515{
516 to->pkt_type = from->pkt_type;
517 to->priority = from->priority;
518 to->protocol = from->protocol;
Shmulik Ladkanid2f0c962019-04-29 16:39:30 +0300519 to->skb_iif = from->skb_iif;
Eric Dumazetadf30902009-06-02 05:19:30 +0000520 skb_dst_drop(to);
Eric Dumazetfe76cda2010-07-01 23:48:22 +0000521 skb_dst_copy(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800523 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Paolo Abeni3dd1c9a2018-07-23 16:50:48 +0200525 skb_copy_hash(to, from);
526
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 /* Copy the flags to each fragment. */
528 IPCB(to)->flags = IPCB(from)->flags;
529
530#ifdef CONFIG_NET_SCHED
531 to->tc_index = from->tc_index;
532#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -0700533 nf_copy(to, from);
Florian Westphaldf5042f2018-12-18 17:15:16 +0100534 skb_ext_copy(to, from);
Javier Martinez Canillas6ca40d42016-09-09 08:43:16 -0400535#if IS_ENABLED(CONFIG_IP_VS)
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300536 to->ipvs_property = from->ipvs_property;
537#endif
James Morris984bc162006-06-09 00:29:17 -0700538 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
Eric W. Biederman694869b2015-06-12 21:55:31 -0500541static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
Florian Westphalc5501eb2015-05-22 16:32:50 +0200542 unsigned int mtu,
Eric W. Biederman694869b2015-06-12 21:55:31 -0500543 int (*output)(struct net *, struct sock *, struct sk_buff *))
Andy Zhou49d16b22015-05-15 14:15:37 -0700544{
545 struct iphdr *iph = ip_hdr(skb);
Andy Zhou49d16b22015-05-15 14:15:37 -0700546
Florian Westphald6b915e2015-05-22 16:32:51 +0200547 if ((iph->frag_off & htons(IP_DF)) == 0)
Eric W. Biederman694869b2015-06-12 21:55:31 -0500548 return ip_do_fragment(net, sk, skb, output);
Florian Westphald6b915e2015-05-22 16:32:51 +0200549
550 if (unlikely(!skb->ignore_df ||
Andy Zhou49d16b22015-05-15 14:15:37 -0700551 (IPCB(skb)->frag_max_size &&
552 IPCB(skb)->frag_max_size > mtu))) {
Eric W. Biederman9479b0a2015-09-15 20:04:00 -0500553 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
Andy Zhou49d16b22015-05-15 14:15:37 -0700554 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
555 htonl(mtu));
556 kfree_skb(skb);
557 return -EMSGSIZE;
558 }
559
Eric W. Biederman694869b2015-06-12 21:55:31 -0500560 return ip_do_fragment(net, sk, skb, output);
Andy Zhou49d16b22015-05-15 14:15:37 -0700561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * This IP datagram is too large to be sent in one piece. Break it up into
565 * smaller pieces (each of size equal to IP header plus
566 * a block of the data of the original IP data part) that will yet fit in a
567 * single device frame, and queue such a frame for sending.
568 */
569
Eric W. Biederman694869b2015-06-12 21:55:31 -0500570int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
571 int (*output)(struct net *, struct sock *, struct sk_buff *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
573 struct iphdr *iph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 int ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 struct sk_buff *skb2;
Changli Gaoc893b802010-07-31 13:25:08 +0000576 unsigned int mtu, hlen, left, len, ll_rs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800578 __be16 not_last_frag;
Eric Dumazet511c3f92009-06-02 05:14:27 +0000579 struct rtable *rt = skb_rtable(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 int err = 0;
581
Hannes Frederic Sowadbd33932015-10-27 22:40:40 +0100582 /* for offloaded checksums cleanup checksum before fragmentation */
583 if (skb->ip_summed == CHECKSUM_PARTIAL &&
584 (err = skb_checksum_help(skb)))
585 goto fail;
586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 /*
588 * Point into the IP datagram header.
589 */
590
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700591 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Shmulik Ladkanifedbb6b42016-06-29 21:47:03 +0300593 mtu = ip_skb_dst_mtu(sk, skb);
Florian Westphald6b915e2015-05-22 16:32:51 +0200594 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
595 mtu = IPCB(skb)->frag_max_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597 /*
598 * Setup starting values.
599 */
600
601 hlen = iph->ihl * 4;
Hannes Frederic Sowaf87c10a2014-01-09 10:01:15 +0100602 mtu = mtu - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800603 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Vasily Averin254d9002017-07-14 12:04:16 +0300604 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 /* When frag_list is given, use it. First, check its validity:
607 * some transformers could create wrong frag_list or break existing
608 * one, it is not prohibited. In this case fall back to copying.
609 *
610 * LATER: this step can be merged to real generation of fragments,
611 * we can switch to copy when see the first bad fragment.
612 */
David S. Miller21dc3302010-08-23 00:13:46 -0700613 if (skb_has_frag_list(skb)) {
Eric Dumazet3d130082010-09-21 08:47:45 +0000614 struct sk_buff *frag, *frag2;
Alexey Dobriyanc72d8cd2016-11-19 04:08:08 +0300615 unsigned int first_len = skb_pagelen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 if (first_len - hlen > mtu ||
618 ((first_len - hlen) & 7) ||
Paul Gortmaker56f8a752011-06-21 20:33:34 -0700619 ip_is_fragment(iph) ||
Vasily Averin254d9002017-07-14 12:04:16 +0300620 skb_cloned(skb) ||
621 skb_headroom(skb) < ll_rs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 goto slow_path;
623
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700624 skb_walk_frags(skb, frag) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 /* Correct geometry. */
626 if (frag->len > mtu ||
627 ((frag->len & 7) && frag->next) ||
Vasily Averin254d9002017-07-14 12:04:16 +0300628 skb_headroom(frag) < hlen + ll_rs)
Eric Dumazet3d130082010-09-21 08:47:45 +0000629 goto slow_path_clean;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
631 /* Partially cloned skb? */
632 if (skb_shared(frag))
Eric Dumazet3d130082010-09-21 08:47:45 +0000633 goto slow_path_clean;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700634
635 BUG_ON(frag->sk);
636 if (skb->sk) {
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700637 frag->sk = skb->sk;
638 frag->destructor = sock_wfree;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700639 }
Eric Dumazet3d130082010-09-21 08:47:45 +0000640 skb->truesize -= frag->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642
643 /* Everything is OK. Generate! */
644
645 err = 0;
646 offset = 0;
647 frag = skb_shinfo(skb)->frag_list;
David S. Millerd7fcf1a2009-06-09 00:19:37 -0700648 skb_frag_list_init(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 skb->data_len = first_len - skb_headlen(skb);
650 skb->len = first_len;
651 iph->tot_len = htons(first_len);
652 iph->frag_off = htons(IP_MF);
653 ip_send_check(iph);
654
655 for (;;) {
656 /* Prepare header of the next frame,
657 * before previous one went down. */
658 if (frag) {
659 frag->ip_summed = CHECKSUM_NONE;
Arnaldo Carvalho de Melobadff6d2007-03-13 13:06:52 -0300660 skb_reset_transport_header(frag);
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700661 __skb_push(frag, hlen);
662 skb_reset_network_header(frag);
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -0700663 memcpy(skb_network_header(frag), iph, hlen);
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700664 iph = ip_hdr(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 iph->tot_len = htons(frag->len);
666 ip_copy_metadata(frag, skb);
667 if (offset == 0)
668 ip_options_fragment(frag);
669 offset += skb->len - hlen;
670 iph->frag_off = htons(offset>>3);
Ian Morris00db4122015-04-03 09:17:27 +0100671 if (frag->next)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 iph->frag_off |= htons(IP_MF);
673 /* Ready, complete checksum */
674 ip_send_check(iph);
675 }
676
Eric W. Biederman694869b2015-06-12 21:55:31 -0500677 err = output(net, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Wei Dongdafee492006-08-02 13:41:21 -0700679 if (!err)
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500680 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 if (err || !frag)
682 break;
683
684 skb = frag;
685 frag = skb->next;
David S. Millera8305bf2018-07-29 20:42:53 -0700686 skb_mark_not_on_list(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
688
689 if (err == 0) {
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500690 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 return 0;
692 }
693
Pablo Neira Ayuso942f1462019-04-04 13:54:20 +0200694 kfree_skb_list(frag);
695
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500696 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return err;
Eric Dumazet3d130082010-09-21 08:47:45 +0000698
699slow_path_clean:
700 skb_walk_frags(skb, frag2) {
701 if (frag2 == frag)
702 break;
703 frag2->sk = NULL;
704 frag2->destructor = NULL;
705 skb->truesize += frag2->truesize;
706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 }
708
709slow_path:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000710 iph = ip_hdr(skb);
Alexander Duyckfc70fb62012-12-07 14:14:15 +0000711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 left = skb->len - hlen; /* Space per frame */
George Kadianakis49085bd2010-07-06 11:44:12 +0000713 ptr = hlen; /* Where to start from */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 /*
716 * Fragment the datagram.
717 */
718
719 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
720 not_last_frag = iph->frag_off & htons(IP_MF);
721
722 /*
723 * Keep copying data until we run out.
724 */
725
Stephen Hemminger132adf52007-03-08 20:44:43 -0800726 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 len = left;
728 /* IF: it doesn't fit, use 'mtu' - the data space left */
729 if (len > mtu)
730 len = mtu;
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300731 /* IF: we are not sending up to and including the packet end
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 then align the next start on an eight byte boundary */
733 if (len < left) {
734 len &= ~7;
735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Joe Perchescbffccc2014-11-05 14:39:21 -0800737 /* Allocate buffer */
738 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
739 if (!skb2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 err = -ENOMEM;
741 goto fail;
742 }
743
744 /*
745 * Set up data on packet
746 */
747
748 ip_copy_metadata(skb2, skb);
749 skb_reserve(skb2, ll_rs);
750 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700751 skb_reset_network_header(skb2);
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700752 skb2->transport_header = skb2->network_header + hlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
754 /*
755 * Charge the memory for the fragment to any owner
756 * it might possess
757 */
758
759 if (skb->sk)
760 skb_set_owner_w(skb2, skb->sk);
761
762 /*
763 * Copy the packet header into the new buffer.
764 */
765
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -0300766 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 /*
769 * Copy a block of the IP datagram.
770 */
Arnaldo Carvalho de Melobff9b612007-03-16 17:19:57 -0300771 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 BUG();
773 left -= len;
774
775 /*
776 * Fill in the new header fields.
777 */
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700778 iph = ip_hdr(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 iph->frag_off = htons((offset >> 3));
780
Florian Westphald6b915e2015-05-22 16:32:51 +0200781 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
782 iph->frag_off |= htons(IP_DF);
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 /* ANK: dirty, but effective trick. Upgrade options only if
785 * the segment to be fragmented was THE FIRST (otherwise,
786 * options are already fixed) and make it ONCE
787 * on the initial skb, so that all the following fragments
788 * will inherit fixed options.
789 */
790 if (offset == 0)
791 ip_options_fragment(skb);
792
793 /*
794 * Added AC : If we are fragmenting a fragment that's not the
795 * last fragment then keep MF on each bit
796 */
797 if (left > 0 || not_last_frag)
798 iph->frag_off |= htons(IP_MF);
799 ptr += len;
800 offset += len;
801
802 /*
803 * Put this fragment into the sending queue.
804 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 iph->tot_len = htons(len + hlen);
806
807 ip_send_check(iph);
808
Eric W. Biederman694869b2015-06-12 21:55:31 -0500809 err = output(net, sk, skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 if (err)
811 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700812
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500813 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 }
Eric Dumazet5d0ba552012-06-04 01:17:19 +0000815 consume_skb(skb);
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500816 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return err;
818
819fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900820 kfree_skb(skb);
Eric W. Biederman26a949db2015-09-15 20:03:59 -0500821 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 return err;
823}
Andy Zhou49d16b22015-05-15 14:15:37 -0700824EXPORT_SYMBOL(ip_do_fragment);
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826int
827ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
828{
Al Virof69e6d12014-11-24 13:23:40 -0500829 struct msghdr *msg = from;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Patrick McHardy84fa7932006-08-29 16:44:56 -0700831 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Al Viro0b62fca2016-11-03 18:17:31 -0400832 if (!copy_from_iter_full(to, len, &msg->msg_iter))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return -EFAULT;
834 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800835 __wsum csum = 0;
Al Viro0b62fca2016-11-03 18:17:31 -0400836 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 return -EFAULT;
838 skb->csum = csum_block_add(skb->csum, csum, odd);
839 }
840 return 0;
841}
Eric Dumazet4bc2f182010-07-09 21:22:10 +0000842EXPORT_SYMBOL(ip_generic_getfrag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Al Viro44bb9362006-11-14 21:36:14 -0800844static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845csum_page(struct page *page, int offset, int copy)
846{
847 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800848 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 kaddr = kmap(page);
850 csum = csum_partial(kaddr + offset, copy, 0);
851 kunmap(page);
852 return csum;
853}
854
David S. Millerf5fca602011-05-08 17:24:10 -0700855static int __ip_append_data(struct sock *sk,
856 struct flowi4 *fl4,
857 struct sk_buff_head *queue,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000858 struct inet_cork *cork,
Eric Dumazet5640f762012-09-23 23:04:42 +0000859 struct page_frag *pfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +0000860 int getfrag(void *from, char *to, int offset,
861 int len, int odd, struct sk_buff *skb),
862 void *from, int length, int transhdrlen,
863 unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
865 struct inet_sock *inet = inet_sk(sk);
Willem de Bruijnb5947e52018-11-30 15:32:39 -0500866 struct ubuf_info *uarg = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct sk_buff *skb;
868
Herbert Xu07df5292011-03-01 23:00:58 -0800869 struct ip_options *opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 int hh_len;
871 int exthdrlen;
872 int mtu;
873 int copy;
874 int err;
875 int offset = 0;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100876 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 int csummode = CHECKSUM_NONE;
Herbert Xu1470ddf2011-03-01 02:36:47 +0000878 struct rtable *rt = (struct rtable *)cork->dst;
Eric Dumazet694aba62018-03-31 13:16:25 -0700879 unsigned int wmem_alloc_delta = 0;
Willem de Bruijn52900d22018-11-30 15:32:40 -0500880 bool paged, extra_uref;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400881 u32 tskey = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Steffen Klassert96d73032011-06-05 20:48:47 +0000883 skb = skb_peek_tail(queue);
884
885 exthdrlen = !skb ? rt->dst.header_len : 0;
Willem de Bruijnbec1f6f2018-04-26 13:42:17 -0400886 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
Willem de Bruijn15e36f52018-04-26 13:42:19 -0400887 paged = !!cork->gso_size;
Willem de Bruijnbec1f6f2018-04-26 13:42:17 -0400888
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400889 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
890 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
891 tskey = sk->sk_tskey++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
Changli Gaod8d1f302010-06-10 23:31:35 -0700893 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894
895 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
896 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
WANG Cong60ff7462014-05-04 16:39:18 -0700897 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +0100899 if (cork->length + length > maxnonfragsize - fragheaderlen) {
David S. Millerf5fca602011-05-08 17:24:10 -0700900 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +0100901 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return -EMSGSIZE;
903 }
904
905 /*
906 * transhdrlen > 0 means that this is the first fragment and we wish
907 * it won't be fragmented in the future.
908 */
909 if (transhdrlen &&
910 length + fragheaderlen <= mtu &&
Tom Herbertc8cd0982015-12-14 11:19:44 -0800911 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
Willem de Bruijnbec1f6f2018-04-26 13:42:17 -0400912 (!(flags & MSG_MORE) || cork->gso_size) &&
Jacek Kalwascd027a52018-04-12 12:03:13 -0700913 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
Patrick McHardy84fa7932006-08-29 16:44:56 -0700914 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
Willem de Bruijnb5947e52018-11-30 15:32:39 -0500916 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
917 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
918 if (!uarg)
919 return -ENOBUFS;
Willem de Bruijn52900d22018-11-30 15:32:40 -0500920 extra_uref = true;
Willem de Bruijnb5947e52018-11-30 15:32:39 -0500921 if (rt->dst.dev->features & NETIF_F_SG &&
922 csummode == CHECKSUM_PARTIAL) {
923 paged = true;
924 } else {
925 uarg->zerocopy = 0;
Willem de Bruijn52900d22018-11-30 15:32:40 -0500926 skb_zcopy_set(skb, uarg, &extra_uref);
Willem de Bruijnb5947e52018-11-30 15:32:39 -0500927 }
928 }
929
Herbert Xu1470ddf2011-03-01 02:36:47 +0000930 cork->length += length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 /* So, what's going on in the loop below?
933 *
934 * We use calculated fragment length to generate chained skb,
935 * each of segments is IP fragment ready for sending to network after
936 * adding appropriate IP header.
937 */
938
Herbert Xu26cde9f2010-06-15 01:52:25 +0000939 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 goto alloc_new_skb;
941
942 while (length > 0) {
943 /* Check if the remaining data fits into current packet. */
944 copy = mtu - skb->len;
945 if (copy < length)
946 copy = maxfraglen - skb->len;
947 if (copy <= 0) {
948 char *data;
949 unsigned int datalen;
950 unsigned int fraglen;
951 unsigned int fraggap;
952 unsigned int alloclen;
Willem de Bruijnaba36932018-11-24 14:21:16 -0500953 unsigned int pagedlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 struct sk_buff *skb_prev;
955alloc_new_skb:
956 skb_prev = skb;
957 if (skb_prev)
958 fraggap = skb_prev->len - maxfraglen;
959 else
960 fraggap = 0;
961
962 /*
963 * If remaining data exceeds the mtu,
964 * we know we need more fragment(s).
965 */
966 datalen = length + fraggap;
967 if (datalen > mtu - fragheaderlen)
968 datalen = maxfraglen - fragheaderlen;
969 fraglen = datalen + fragheaderlen;
Willem de Bruijnaba36932018-11-24 14:21:16 -0500970 pagedlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900972 if ((flags & MSG_MORE) &&
Changli Gaod8d1f302010-06-10 23:31:35 -0700973 !(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 alloclen = mtu;
Willem de Bruijn15e36f52018-04-26 13:42:19 -0400975 else if (!paged)
Eric Dumazet59104f02010-09-20 20:16:27 +0000976 alloclen = fraglen;
Willem de Bruijn15e36f52018-04-26 13:42:19 -0400977 else {
978 alloclen = min_t(int, fraglen, MAX_HEADER);
979 pagedlen = fraglen - alloclen;
980 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
Steffen Klassert353e5c92011-06-22 01:05:37 +0000982 alloclen += exthdrlen;
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* The last fragment gets additional space at tail.
985 * Note, with MSG_MORE we overallocate on fragments,
986 * because we have no idea what fragment will be
987 * the last.
988 */
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000989 if (datalen == length + fraggap)
Changli Gaod8d1f302010-06-10 23:31:35 -0700990 alloclen += rt->dst.trailer_len;
Steffen Klassert33f99dc2011-06-22 01:04:37 +0000991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900993 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 alloclen + hh_len + 15,
995 (flags & MSG_DONTWAIT), &err);
996 } else {
997 skb = NULL;
Eric Dumazet694aba62018-03-31 13:16:25 -0700998 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 2 * sk->sk_sndbuf)
Eric Dumazet694aba62018-03-31 13:16:25 -07001000 skb = alloc_skb(alloclen + hh_len + 15,
1001 sk->sk_allocation);
Ian Morris51456b22015-04-03 09:17:26 +01001002 if (unlikely(!skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 err = -ENOBUFS;
1004 }
Ian Morris51456b22015-04-03 09:17:26 +01001005 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 goto error;
1007
1008 /*
1009 * Fill in the control structures
1010 */
1011 skb->ip_summed = csummode;
1012 skb->csum = 0;
1013 skb_reserve(skb, hh_len);
Willem de Bruijn11878b42014-07-14 17:55:06 -04001014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /*
1016 * Find where to start putting bytes.
1017 */
Willem de Bruijn15e36f52018-04-26 13:42:19 -04001018 data = skb_put(skb, fraglen + exthdrlen - pagedlen);
Arnaldo Carvalho de Meloc14d2452007-03-11 22:39:41 -03001019 skb_set_network_header(skb, exthdrlen);
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -07001020 skb->transport_header = (skb->network_header +
1021 fragheaderlen);
Steffen Klassert353e5c92011-06-22 01:05:37 +00001022 data += fragheaderlen + exthdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 if (fraggap) {
1025 skb->csum = skb_copy_and_csum_bits(
1026 skb_prev, maxfraglen,
1027 data + transhdrlen, fraggap, 0);
1028 skb_prev->csum = csum_sub(skb_prev->csum,
1029 skb->csum);
1030 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -07001031 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033
Willem de Bruijn15e36f52018-04-26 13:42:19 -04001034 copy = datalen - transhdrlen - fraggap - pagedlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1036 err = -EFAULT;
1037 kfree_skb(skb);
1038 goto error;
1039 }
1040
1041 offset += copy;
Willem de Bruijn15e36f52018-04-26 13:42:19 -04001042 length -= copy + transhdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 transhdrlen = 0;
1044 exthdrlen = 0;
1045 csummode = CHECKSUM_NONE;
1046
Willem de Bruijn52900d22018-11-30 15:32:40 -05001047 /* only the initial fragment is time stamped */
1048 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1049 cork->tx_flags = 0;
1050 skb_shinfo(skb)->tskey = tskey;
1051 tskey = 0;
1052 skb_zcopy_set(skb, uarg, &extra_uref);
1053
Julian Anastasov0dec8792017-02-06 23:14:16 +02001054 if ((flags & MSG_CONFIRM) && !skb_prev)
1055 skb_set_dst_pending_confirm(skb, 1);
1056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 /*
1058 * Put the packet on the pending queue.
1059 */
Eric Dumazet694aba62018-03-31 13:16:25 -07001060 if (!skb->destructor) {
1061 skb->destructor = sock_wfree;
1062 skb->sk = sk;
1063 wmem_alloc_delta += skb->truesize;
1064 }
Herbert Xu1470ddf2011-03-01 02:36:47 +00001065 __skb_queue_tail(queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 continue;
1067 }
1068
1069 if (copy > length)
1070 copy = length;
1071
Willem de Bruijn113f99c2018-05-17 13:13:29 -04001072 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1073 skb_tailroom(skb) >= copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 unsigned int off;
1075
1076 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001077 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 offset, copy, off, skb) < 0) {
1079 __skb_trim(skb, off);
1080 err = -EFAULT;
1081 goto error;
1082 }
Willem de Bruijnb5947e52018-11-30 15:32:39 -05001083 } else if (!uarg || !uarg->zerocopy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 int i = skb_shinfo(skb)->nr_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
Eric Dumazet5640f762012-09-23 23:04:42 +00001086 err = -ENOMEM;
1087 if (!sk_page_frag_refill(sk, pfrag))
1088 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Eric Dumazet5640f762012-09-23 23:04:42 +00001090 if (!skb_can_coalesce(skb, i, pfrag->page,
1091 pfrag->offset)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 err = -EMSGSIZE;
Eric Dumazet5640f762012-09-23 23:04:42 +00001093 if (i == MAX_SKB_FRAGS)
1094 goto error;
1095
1096 __skb_fill_page_desc(skb, i, pfrag->page,
1097 pfrag->offset, 0);
1098 skb_shinfo(skb)->nr_frags = ++i;
1099 get_page(pfrag->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
Eric Dumazet5640f762012-09-23 23:04:42 +00001101 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1102 if (getfrag(from,
1103 page_address(pfrag->page) + pfrag->offset,
1104 offset, copy, skb->len, skb) < 0)
1105 goto error_efault;
1106
1107 pfrag->offset += copy;
1108 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 skb->len += copy;
1110 skb->data_len += copy;
Herbert Xuf945fa72008-01-22 22:39:26 -08001111 skb->truesize += copy;
Eric Dumazet694aba62018-03-31 13:16:25 -07001112 wmem_alloc_delta += copy;
Willem de Bruijnb5947e52018-11-30 15:32:39 -05001113 } else {
1114 err = skb_zerocopy_iter_dgram(skb, from, copy);
1115 if (err < 0)
1116 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 }
1118 offset += copy;
1119 length -= copy;
1120 }
1121
Paolo Abeni9e8445a2018-04-04 14:30:01 +02001122 if (wmem_alloc_delta)
1123 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 return 0;
1125
Eric Dumazet5640f762012-09-23 23:04:42 +00001126error_efault:
1127 err = -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128error:
Willem de Bruijn97ef7b42018-12-08 06:22:46 -05001129 if (uarg)
1130 sock_zerocopy_put_abort(uarg, extra_uref);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001131 cork->length -= length;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001132 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Eric Dumazet694aba62018-03-31 13:16:25 -07001133 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001134 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
1136
Herbert Xu1470ddf2011-03-01 02:36:47 +00001137static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1138 struct ipcm_cookie *ipc, struct rtable **rtp)
1139{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001140 struct ip_options_rcu *opt;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001141 struct rtable *rt;
1142
Gao Feng9783ccd2018-04-16 10:16:45 +08001143 rt = *rtp;
1144 if (unlikely(!rt))
1145 return -EFAULT;
1146
Herbert Xu1470ddf2011-03-01 02:36:47 +00001147 /*
1148 * setup for corking.
1149 */
1150 opt = ipc->opt;
1151 if (opt) {
Ian Morris51456b22015-04-03 09:17:26 +01001152 if (!cork->opt) {
Herbert Xu1470ddf2011-03-01 02:36:47 +00001153 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1154 sk->sk_allocation);
Ian Morris51456b22015-04-03 09:17:26 +01001155 if (unlikely(!cork->opt))
Herbert Xu1470ddf2011-03-01 02:36:47 +00001156 return -ENOBUFS;
1157 }
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001158 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001159 cork->flags |= IPCORK_OPT;
1160 cork->addr = ipc->addr;
1161 }
Gao Feng9783ccd2018-04-16 10:16:45 +08001162
Herbert Xu1470ddf2011-03-01 02:36:47 +00001163 /*
1164 * We steal reference to this route, caller should not release it
1165 */
1166 *rtp = NULL;
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001167 cork->fragsize = ip_sk_use_pmtu(sk) ?
1168 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
Willem de Bruijnbec1f6f2018-04-26 13:42:17 -04001169
Willem de Bruijnfbf47812018-07-06 10:12:59 -04001170 cork->gso_size = ipc->gso_size;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001171 cork->dst = &rt->dst;
1172 cork->length = 0;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001173 cork->ttl = ipc->ttl;
1174 cork->tos = ipc->tos;
1175 cork->priority = ipc->priority;
Jesus Sanchez-Palenciabc969a92018-07-03 15:42:49 -07001176 cork->transmit_time = ipc->sockc.transmit_time;
Willem de Bruijn678ca422018-07-06 10:12:58 -04001177 cork->tx_flags = 0;
1178 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001179
1180 return 0;
1181}
1182
1183/*
1184 * ip_append_data() and ip_append_page() can make one large IP datagram
1185 * from many pieces of data. Each pieces will be holded on the socket
1186 * until ip_push_pending_frames() is called. Each piece can be a page
1187 * or non-page data.
1188 *
1189 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1190 * this interface potentially.
1191 *
1192 * LATER: length must be adjusted by pad at tail, when it is required.
1193 */
David S. Millerf5fca602011-05-08 17:24:10 -07001194int ip_append_data(struct sock *sk, struct flowi4 *fl4,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001195 int getfrag(void *from, char *to, int offset, int len,
1196 int odd, struct sk_buff *skb),
1197 void *from, int length, int transhdrlen,
1198 struct ipcm_cookie *ipc, struct rtable **rtp,
1199 unsigned int flags)
1200{
1201 struct inet_sock *inet = inet_sk(sk);
1202 int err;
1203
1204 if (flags&MSG_PROBE)
1205 return 0;
1206
1207 if (skb_queue_empty(&sk->sk_write_queue)) {
David S. Millerbdc712b2011-05-06 15:02:07 -07001208 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001209 if (err)
1210 return err;
1211 } else {
1212 transhdrlen = 0;
1213 }
1214
Eric Dumazet5640f762012-09-23 23:04:42 +00001215 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1216 sk_page_frag(sk), getfrag,
Herbert Xu1470ddf2011-03-01 02:36:47 +00001217 from, length, transhdrlen, flags);
1218}
1219
David S. Millerf5fca602011-05-08 17:24:10 -07001220ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 int offset, size_t size, int flags)
1222{
1223 struct inet_sock *inet = inet_sk(sk);
1224 struct sk_buff *skb;
1225 struct rtable *rt;
1226 struct ip_options *opt = NULL;
David S. Millerbdc712b2011-05-06 15:02:07 -07001227 struct inet_cork *cork;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 int hh_len;
1229 int mtu;
1230 int len;
1231 int err;
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001232 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 if (inet->hdrincl)
1235 return -EPERM;
1236
1237 if (flags&MSG_PROBE)
1238 return 0;
1239
1240 if (skb_queue_empty(&sk->sk_write_queue))
1241 return -EINVAL;
1242
David S. Millerbdc712b2011-05-06 15:02:07 -07001243 cork = &inet->cork.base;
1244 rt = (struct rtable *)cork->dst;
1245 if (cork->flags & IPCORK_OPT)
1246 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Changli Gaod8d1f302010-06-10 23:31:35 -07001248 if (!(rt->dst.dev->features&NETIF_F_SG))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 return -EOPNOTSUPP;
1250
Changli Gaod8d1f302010-06-10 23:31:35 -07001251 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
Willem de Bruijnbec1f6f2018-04-26 13:42:17 -04001252 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1255 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
WANG Cong60ff7462014-05-04 16:39:18 -07001256 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Hannes Frederic Sowadaba2872013-10-27 17:29:11 +01001258 if (cork->length + size > maxnonfragsize - fragheaderlen) {
Hannes Frederic Sowa61e7f092013-12-19 02:13:36 +01001259 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1260 mtu - (opt ? opt->optlen : 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 return -EMSGSIZE;
1262 }
1263
Ian Morris51456b22015-04-03 09:17:26 +01001264 skb = skb_peek_tail(&sk->sk_write_queue);
1265 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 return -EINVAL;
1267
Hannes Frederic Sowaa8c4a252016-02-22 18:43:25 +01001268 cork->length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 while (size > 0) {
Willem de Bruijnab2fb7e2017-08-22 11:39:57 -04001271 /* Check if the remaining data fits into current packet. */
1272 len = mtu - skb->len;
1273 if (len < size)
1274 len = maxfraglen - skb->len;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (len <= 0) {
1277 struct sk_buff *skb_prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 int alloclen;
1279
1280 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001281 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 alloclen = fragheaderlen + hh_len + fraggap + 15;
1284 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1285 if (unlikely(!skb)) {
1286 err = -ENOBUFS;
1287 goto error;
1288 }
1289
1290 /*
1291 * Fill in the control structures
1292 */
1293 skb->ip_summed = CHECKSUM_NONE;
1294 skb->csum = 0;
1295 skb_reserve(skb, hh_len);
1296
1297 /*
1298 * Find where to start putting bytes.
1299 */
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001300 skb_put(skb, fragheaderlen + fraggap);
Arnaldo Carvalho de Melo2ca9e6f2007-03-10 19:15:25 -03001301 skb_reset_network_header(skb);
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -07001302 skb->transport_header = (skb->network_header +
1303 fragheaderlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (fraggap) {
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001305 skb->csum = skb_copy_and_csum_bits(skb_prev,
1306 maxfraglen,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001307 skb_transport_header(skb),
Arnaldo Carvalho de Melo967b05f2007-03-13 13:51:52 -03001308 fraggap, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 skb_prev->csum = csum_sub(skb_prev->csum,
1310 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001311 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 }
1313
1314 /*
1315 * Put the packet on the pending queue.
1316 */
1317 __skb_queue_tail(&sk->sk_write_queue, skb);
1318 continue;
1319 }
1320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (len > size)
1322 len = size;
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +02001323
1324 if (skb_append_pagefrags(skb, page, offset, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 err = -EMSGSIZE;
1326 goto error;
1327 }
1328
1329 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001330 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 csum = csum_page(page, offset, len);
1332 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1333 }
1334
1335 skb->len += len;
1336 skb->data_len += len;
David S. Miller1e34a112008-01-22 23:44:31 -08001337 skb->truesize += len;
Reshetova, Elena14afee42017-06-30 13:08:00 +03001338 refcount_add(len, &sk->sk_wmem_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 offset += len;
1340 size -= len;
1341 }
1342 return 0;
1343
1344error:
David S. Millerbdc712b2011-05-06 15:02:07 -07001345 cork->length -= size;
Pavel Emelyanov5e38e272008-07-16 20:19:49 -07001346 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 return err;
1348}
1349
Herbert Xu1470ddf2011-03-01 02:36:47 +00001350static void ip_cork_release(struct inet_cork *cork)
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001351{
Herbert Xu1470ddf2011-03-01 02:36:47 +00001352 cork->flags &= ~IPCORK_OPT;
1353 kfree(cork->opt);
1354 cork->opt = NULL;
1355 dst_release(cork->dst);
1356 cork->dst = NULL;
Pavel Emelyanov429f08e2007-11-05 21:03:24 -08001357}
1358
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359/*
1360 * Combined all pending IP fragments on the socket as one IP datagram
1361 * and push them out.
1362 */
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001363struct sk_buff *__ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001364 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001365 struct sk_buff_head *queue,
1366 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
1368 struct sk_buff *skb, *tmp_skb;
1369 struct sk_buff **tail_skb;
1370 struct inet_sock *inet = inet_sk(sk);
Pavel Emelyanov0388b0042008-07-14 23:00:43 -07001371 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 struct ip_options *opt = NULL;
Herbert Xu1470ddf2011-03-01 02:36:47 +00001373 struct rtable *rt = (struct rtable *)cork->dst;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001375 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 __u8 ttl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Ian Morris51456b22015-04-03 09:17:26 +01001378 skb = __skb_dequeue(queue);
1379 if (!skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 goto out;
1381 tail_skb = &(skb_shinfo(skb)->frag_list);
1382
1383 /* move skb->data to ip header from ext header */
Arnaldo Carvalho de Melod56f90a2007-04-10 20:50:43 -07001384 if (skb->data < skb_network_header(skb))
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001385 __skb_pull(skb, skb_network_offset(skb));
Herbert Xu1470ddf2011-03-01 02:36:47 +00001386 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
Arnaldo Carvalho de Melocfe1fc72007-03-16 17:26:39 -03001387 __skb_pull(tmp_skb, skb_network_header_len(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 *tail_skb = tmp_skb;
1389 tail_skb = &(tmp_skb->next);
1390 skb->len += tmp_skb->len;
1391 skb->data_len += tmp_skb->len;
1392 skb->truesize += tmp_skb->truesize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 tmp_skb->destructor = NULL;
1394 tmp_skb->sk = NULL;
1395 }
1396
1397 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1398 * to fragment the frame generated here. No matter, what transforms
1399 * how transforms change size of the packet, it will come out.
1400 */
WANG Cong60ff7462014-05-04 16:39:18 -07001401 skb->ignore_df = ip_sk_ignore_df(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403 /* DF bit is set when we want to see DF on outgoing frames.
WANG Cong60ff7462014-05-04 16:39:18 -07001404 * If ignore_df is set too, we still allow to fragment this frame
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 * locally. */
Hannes Frederic Sowa482fc602013-11-05 02:24:17 +01001406 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1407 inet->pmtudisc == IP_PMTUDISC_PROBE ||
Changli Gaod8d1f302010-06-10 23:31:35 -07001408 (skb->len <= dst_mtu(&rt->dst) &&
1409 ip_dont_fragment(sk, &rt->dst)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 df = htons(IP_DF);
1411
Herbert Xu1470ddf2011-03-01 02:36:47 +00001412 if (cork->flags & IPCORK_OPT)
1413 opt = cork->opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Francesco Fuscoaa661582013-09-24 15:43:09 +02001415 if (cork->ttl != 0)
1416 ttl = cork->ttl;
1417 else if (rt->rt_type == RTN_MULTICAST)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 ttl = inet->mc_ttl;
1419 else
Changli Gaod8d1f302010-06-10 23:31:35 -07001420 ttl = ip_select_ttl(inet, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Ansis Atteka749154a2013-09-18 15:29:52 -07001422 iph = ip_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 iph->version = 4;
1424 iph->ihl = 5;
Francesco Fuscoaa661582013-09-24 15:43:09 +02001425 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 iph->frag_off = df;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 iph->ttl = ttl;
1428 iph->protocol = sk->sk_protocol;
Eric Dumazet84f93072011-11-30 19:00:53 +00001429 ip_copy_addrs(iph, fl4);
Hannes Frederic Sowab6a77192015-03-25 17:07:44 +01001430 ip_select_ident(net, skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
David S. Miller22f728f2011-05-13 17:21:27 -04001432 if (opt) {
1433 iph->ihl += opt->optlen>>2;
1434 ip_options_build(skb, opt, cork->addr, rt, 0);
1435 }
1436
Francesco Fuscoaa661582013-09-24 15:43:09 +02001437 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
Laszlo Attila Toth4a19ec52008-01-30 19:08:16 -08001438 skb->mark = sk->sk_mark;
Jesus Sanchez-Palenciabc969a92018-07-03 15:42:49 -07001439 skb->tstamp = cork->transmit_time;
Eric Dumazeta21bba92008-11-24 16:07:50 -08001440 /*
1441 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1442 * on dst refcount
1443 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001444 cork->dst = NULL;
Changli Gaod8d1f302010-06-10 23:31:35 -07001445 skb_dst_set(skb, &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
David L Stevens96793b42007-09-17 09:57:33 -07001447 if (iph->protocol == IPPROTO_ICMP)
Pavel Emelyanov0388b0042008-07-14 23:00:43 -07001448 icmp_out_count(net, ((struct icmphdr *)
David L Stevens96793b42007-09-17 09:57:33 -07001449 skb_transport_header(skb))->type);
1450
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001451 ip_cork_release(cork);
1452out:
1453 return skb;
1454}
1455
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001456int ip_send_skb(struct net *net, struct sk_buff *skb)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001457{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001458 int err;
1459
Eric W. Biederman33224b12015-10-07 16:48:46 -05001460 err = ip_local_out(net, skb->sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 if (err) {
1462 if (err > 0)
Eric Dumazet6ce9e7b2009-09-02 18:05:33 -07001463 err = net_xmit_errno(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 if (err)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001465 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 }
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
1470
David S. Miller77968b72011-05-08 17:12:19 -07001471int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
Herbert Xu1470ddf2011-03-01 02:36:47 +00001472{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001473 struct sk_buff *skb;
1474
David S. Miller77968b72011-05-08 17:12:19 -07001475 skb = ip_finish_skb(sk, fl4);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001476 if (!skb)
1477 return 0;
1478
1479 /* Netfilter gets whole the not fragmented skb. */
Eric Dumazetb5ec8ee2012-08-10 02:22:47 +00001480 return ip_send_skb(sock_net(sk), skb);
Herbert Xu1470ddf2011-03-01 02:36:47 +00001481}
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483/*
1484 * Throw away all pending data on the socket.
1485 */
Herbert Xu1470ddf2011-03-01 02:36:47 +00001486static void __ip_flush_pending_frames(struct sock *sk,
1487 struct sk_buff_head *queue,
1488 struct inet_cork *cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 struct sk_buff *skb;
1491
Herbert Xu1470ddf2011-03-01 02:36:47 +00001492 while ((skb = __skb_dequeue_tail(queue)) != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 kfree_skb(skb);
1494
Herbert Xu1470ddf2011-03-01 02:36:47 +00001495 ip_cork_release(cork);
1496}
1497
1498void ip_flush_pending_frames(struct sock *sk)
1499{
David S. Millerbdc712b2011-05-06 15:02:07 -07001500 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
1502
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001503struct sk_buff *ip_make_skb(struct sock *sk,
David S. Miller77968b72011-05-08 17:12:19 -07001504 struct flowi4 *fl4,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001505 int getfrag(void *from, char *to, int offset,
1506 int len, int odd, struct sk_buff *skb),
1507 void *from, int length, int transhdrlen,
1508 struct ipcm_cookie *ipc, struct rtable **rtp,
Willem de Bruijn1cd78842018-04-26 13:42:15 -04001509 struct inet_cork *cork, unsigned int flags)
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001510{
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001511 struct sk_buff_head queue;
1512 int err;
1513
1514 if (flags & MSG_PROBE)
1515 return NULL;
1516
1517 __skb_queue_head_init(&queue);
1518
Willem de Bruijn1cd78842018-04-26 13:42:15 -04001519 cork->flags = 0;
1520 cork->addr = 0;
1521 cork->opt = NULL;
1522 err = ip_setup_cork(sk, cork, ipc, rtp);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001523 if (err)
1524 return ERR_PTR(err);
1525
Willem de Bruijn1cd78842018-04-26 13:42:15 -04001526 err = __ip_append_data(sk, fl4, &queue, cork,
Eric Dumazet5640f762012-09-23 23:04:42 +00001527 &current->task_frag, getfrag,
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001528 from, length, transhdrlen, flags);
1529 if (err) {
Willem de Bruijn1cd78842018-04-26 13:42:15 -04001530 __ip_flush_pending_frames(sk, &queue, cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001531 return ERR_PTR(err);
1532 }
1533
Willem de Bruijn1cd78842018-04-26 13:42:15 -04001534 return __ip_make_skb(sk, fl4, &queue, cork);
Herbert Xu1c32c5a2011-03-01 02:36:47 +00001535}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
1537/*
1538 * Fetch data from kernel space and fill in checksum if needed.
1539 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001540static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 int len, int odd, struct sk_buff *skb)
1542{
Al Viro50842052006-11-14 21:36:34 -08001543 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1546 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001547 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548}
1549
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001550/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * Generic function to send a packet as reply to another packet.
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001552 * Used to send some TCP resets/acks so far.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Eric Dumazetbdbbb852015-01-29 21:35:05 -08001554void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
Eric Dumazet24a2d432014-09-27 09:50:55 -07001555 const struct ip_options *sopt,
1556 __be32 daddr, __be32 saddr,
1557 const struct ip_reply_arg *arg,
David S. Miller70e73412012-06-28 03:21:41 -07001558 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001560 struct ip_options_data replyopts;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 struct ipcm_cookie ipc;
David S. Miller77968b72011-05-08 17:12:19 -07001562 struct flowi4 fl4;
Eric Dumazet511c3f92009-06-02 05:14:27 +00001563 struct rtable *rt = skb_rtable(skb);
Eric Dumazetbdbbb852015-01-29 21:35:05 -08001564 struct net *net = sock_net(sk);
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001565 struct sk_buff *nskb;
Vasily Averin40620902014-10-15 16:24:02 +04001566 int err;
David Ahernf7ba8682015-08-13 14:59:08 -06001567 int oif;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
Paolo Abeni91ed1e62017-08-03 18:07:06 +02001569 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return;
1571
Willem de Bruijn35178202018-07-06 10:12:54 -04001572 ipcm_init(&ipc);
David S. Miller0a5ebb82011-05-09 13:22:43 -07001573 ipc.addr = daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001575 if (replyopts.opt.opt.optlen) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 ipc.opt = &replyopts.opt;
1577
Eric Dumazetf6d8bd02011-04-21 09:45:37 +00001578 if (replyopts.opt.opt.srr)
1579 daddr = replyopts.opt.opt.faddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 }
1581
David Ahernf7ba8682015-08-13 14:59:08 -06001582 oif = arg->bound_dev_if;
David Ahern9b6c14d2016-11-09 09:07:26 -08001583 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1584 oif = skb->skb_iif;
David Ahernf7ba8682015-08-13 14:59:08 -06001585
1586 flowi4_init_output(&fl4, oif,
Jon Maxwell00483692018-05-10 16:53:51 +10001587 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
Eric Dumazet66b13d92011-10-24 03:06:21 -04001588 RT_TOS(arg->tos),
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001589 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
David S. Miller77968b72011-05-08 17:12:19 -07001590 ip_reply_arg_flowi_flags(arg),
David S. Miller70e73412012-06-28 03:21:41 -07001591 daddr, saddr,
Lorenzo Colittie2d118a2016-11-04 02:23:43 +09001592 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1593 arg->uid);
David S. Miller77968b72011-05-08 17:12:19 -07001594 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001595 rt = ip_route_output_key(net, &fl4);
David S. Miller77968b72011-05-08 17:12:19 -07001596 if (IS_ERR(rt))
1597 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Eric Dumazetbdbbb852015-01-29 21:35:05 -08001599 inet_sk(sk)->tos = arg->tos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 sk->sk_priority = skb->priority;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -07001602 sk->sk_protocol = ip_hdr(skb)->protocol;
Patrick McHardyf0e48db2007-06-04 21:32:46 -07001603 sk->sk_bound_dev_if = arg->bound_dev_if;
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001604 sk->sk_sndbuf = sysctl_wmem_default;
Pau Espin Pedrolbf99b4d2017-01-06 20:33:28 +01001605 sk->sk_mark = fl4.flowi4_mark;
Vasily Averin40620902014-10-15 16:24:02 +04001606 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1607 len, 0, &ipc, &rt, MSG_DONTWAIT);
1608 if (unlikely(err)) {
1609 ip_flush_pending_frames(sk);
1610 goto out;
1611 }
1612
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001613 nskb = skb_peek(&sk->sk_write_queue);
1614 if (nskb) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 if (arg->csumoffset >= 0)
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001616 *((__sum16 *)skb_transport_header(nskb) +
1617 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
Arnaldo Carvalho de Melo9c702202007-04-25 18:04:18 -07001618 arg->csum));
Eric Dumazetbe9f4a42012-07-19 07:34:03 +00001619 nskb->ip_summed = CHECKSUM_NONE;
David S. Miller77968b72011-05-08 17:12:19 -07001620 ip_push_pending_frames(sk, &fl4);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
Vasily Averin40620902014-10-15 16:24:02 +04001622out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 ip_rt_put(rt);
1624}
1625
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626void __init ip_init(void)
1627{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 ip_rt_init();
1629 inet_initpeers();
1630
WANG Cong72c1d3b2014-01-10 16:09:45 -08001631#if defined(CONFIG_IP_MULTICAST)
1632 igmp_mc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633#endif
1634}