]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * 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. | |
e905a9ed | 23 | * Bradford Johnson: Fix faulty handling of some frames when |
1da177e4 LT |
24 | * 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. | |
e905a9ed YH |
34 | * Andi Kleen : Split fast and slow ip_build_xmit path |
35 | * for decreased register pressure on x86 | |
36 | * and more readibility. | |
1da177e4 LT |
37 | * 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> | |
1da177e4 LT |
46 | #include <linux/module.h> |
47 | #include <linux/types.h> | |
48 | #include <linux/kernel.h> | |
1da177e4 LT |
49 | #include <linux/mm.h> |
50 | #include <linux/string.h> | |
51 | #include <linux/errno.h> | |
a1f8e7f7 | 52 | #include <linux/highmem.h> |
5a0e3ad6 | 53 | #include <linux/slab.h> |
1da177e4 LT |
54 | |
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> | |
cfacb057 | 69 | #include <net/xfrm.h> |
1da177e4 LT |
70 | #include <linux/skbuff.h> |
71 | #include <net/sock.h> | |
72 | #include <net/arp.h> | |
73 | #include <net/icmp.h> | |
1da177e4 LT |
74 | #include <net/checksum.h> |
75 | #include <net/inetpeer.h> | |
1da177e4 LT |
76 | #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> | |
6cbb0df7 | 81 | #include <linux/tcp.h> |
1da177e4 | 82 | |
ab32ea5d | 83 | int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; |
323e126f | 84 | EXPORT_SYMBOL(sysctl_ip_default_ttl); |
1da177e4 | 85 | |
694869b3 EB |
86 | static int |
87 | ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
88 | unsigned int mtu, | |
89 | int (*output)(struct net *, struct sock *, struct sk_buff *)); | |
49d16b23 | 90 | |
1da177e4 | 91 | /* Generate a checksum for an outgoing IP datagram. */ |
2fbd9679 | 92 | void ip_send_check(struct iphdr *iph) |
1da177e4 LT |
93 | { |
94 | iph->check = 0; | |
95 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
96 | } | |
4bc2f18b | 97 | EXPORT_SYMBOL(ip_send_check); |
1da177e4 | 98 | |
7d771aaa | 99 | static int __ip_local_out_sk(struct sock *sk, struct sk_buff *skb) |
c439cb2e | 100 | { |
29a26a56 | 101 | struct net *net = dev_net(skb_dst(skb)->dev); |
c439cb2e HX |
102 | struct iphdr *iph = ip_hdr(skb); |
103 | ||
104 | iph->tot_len = htons(skb->len); | |
105 | ip_send_check(iph); | |
29a26a56 EB |
106 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, |
107 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
0c4b51f0 | 108 | dst_output_okfn); |
7026b1dd DM |
109 | } |
110 | ||
111 | int __ip_local_out(struct sk_buff *skb) | |
112 | { | |
113 | return __ip_local_out_sk(skb->sk, skb); | |
c439cb2e HX |
114 | } |
115 | ||
aad88724 | 116 | int ip_local_out_sk(struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
117 | { |
118 | int err; | |
119 | ||
fd2874b3 | 120 | err = __ip_local_out_sk(sk, skb); |
c439cb2e | 121 | if (likely(err == 1)) |
5a70649e | 122 | err = dst_output(sk, skb); |
c439cb2e HX |
123 | |
124 | return err; | |
125 | } | |
aad88724 | 126 | EXPORT_SYMBOL_GPL(ip_local_out_sk); |
c439cb2e | 127 | |
1da177e4 LT |
128 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) |
129 | { | |
130 | int ttl = inet->uc_ttl; | |
131 | ||
132 | if (ttl < 0) | |
323e126f | 133 | ttl = ip4_dst_hoplimit(dst); |
1da177e4 LT |
134 | return ttl; |
135 | } | |
136 | ||
e905a9ed | 137 | /* |
1da177e4 LT |
138 | * Add an ip header to a skbuff and send it out. |
139 | * | |
140 | */ | |
cfe673b0 | 141 | int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, |
f6d8bd05 | 142 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) |
1da177e4 LT |
143 | { |
144 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 145 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
146 | struct iphdr *iph; |
147 | ||
148 | /* Build the IP header. */ | |
f6d8bd05 | 149 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); |
8856dfa3 | 150 | skb_reset_network_header(skb); |
eddc9ec5 | 151 | iph = ip_hdr(skb); |
1da177e4 LT |
152 | iph->version = 4; |
153 | iph->ihl = 5; | |
154 | iph->tos = inet->tos; | |
d8d1f30b | 155 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
dd927a26 DM |
156 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); |
157 | iph->saddr = saddr; | |
1da177e4 | 158 | iph->protocol = sk->sk_protocol; |
cfe673b0 ED |
159 | if (ip_dont_fragment(sk, &rt->dst)) { |
160 | iph->frag_off = htons(IP_DF); | |
161 | iph->id = 0; | |
162 | } else { | |
163 | iph->frag_off = 0; | |
164 | __ip_select_ident(sock_net(sk), iph, 1); | |
165 | } | |
1da177e4 | 166 | |
f6d8bd05 ED |
167 | if (opt && opt->opt.optlen) { |
168 | iph->ihl += opt->opt.optlen>>2; | |
169 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
1da177e4 | 170 | } |
1da177e4 LT |
171 | |
172 | skb->priority = sk->sk_priority; | |
4a19ec58 | 173 | skb->mark = sk->sk_mark; |
1da177e4 LT |
174 | |
175 | /* Send it out. */ | |
c439cb2e | 176 | return ip_local_out(skb); |
1da177e4 | 177 | } |
d8c97a94 ACM |
178 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
179 | ||
694869b3 | 180 | static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 181 | { |
adf30907 | 182 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 183 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 184 | struct net_device *dev = dst->dev; |
c2636b4d | 185 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
f6b72b62 | 186 | struct neighbour *neigh; |
a263b309 | 187 | u32 nexthop; |
1da177e4 | 188 | |
edf391ff | 189 | if (rt->rt_type == RTN_MULTICAST) { |
4ba1bf42 | 190 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); |
edf391ff | 191 | } else if (rt->rt_type == RTN_BROADCAST) |
4ba1bf42 | 192 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); |
80787ebc | 193 | |
1da177e4 | 194 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 195 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
196 | struct sk_buff *skb2; |
197 | ||
198 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
51456b29 | 199 | if (!skb2) { |
1da177e4 LT |
200 | kfree_skb(skb); |
201 | return -ENOMEM; | |
202 | } | |
203 | if (skb->sk) | |
204 | skb_set_owner_w(skb2, skb->sk); | |
5d0ba55b | 205 | consume_skb(skb); |
1da177e4 LT |
206 | skb = skb2; |
207 | } | |
208 | ||
a263b309 | 209 | rcu_read_lock_bh(); |
155e8336 | 210 | nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); |
a263b309 DM |
211 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); |
212 | if (unlikely(!neigh)) | |
213 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
9871f1ad | 214 | if (!IS_ERR(neigh)) { |
5110effe | 215 | int res = dst_neigh_output(dst, neigh, skb); |
f2c31e32 | 216 | |
a263b309 | 217 | rcu_read_unlock_bh(); |
f2c31e32 ED |
218 | return res; |
219 | } | |
a263b309 | 220 | rcu_read_unlock_bh(); |
05e3aa09 | 221 | |
e87cc472 JP |
222 | net_dbg_ratelimited("%s: No header cache and no neighbour!\n", |
223 | __func__); | |
1da177e4 LT |
224 | kfree_skb(skb); |
225 | return -EINVAL; | |
226 | } | |
227 | ||
694869b3 EB |
228 | static int ip_finish_output_gso(struct net *net, struct sock *sk, |
229 | struct sk_buff *skb, unsigned int mtu) | |
c7ba65d7 FW |
230 | { |
231 | netdev_features_t features; | |
232 | struct sk_buff *segs; | |
233 | int ret = 0; | |
234 | ||
235 | /* common case: locally created skb or seglen is <= mtu */ | |
236 | if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) || | |
c5501eb3 | 237 | skb_gso_network_seglen(skb) <= mtu) |
694869b3 | 238 | return ip_finish_output2(net, sk, skb); |
c7ba65d7 FW |
239 | |
240 | /* Slowpath - GSO segment length is exceeding the dst MTU. | |
241 | * | |
242 | * This can happen in two cases: | |
243 | * 1) TCP GRO packet, DF bit not set | |
244 | * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly | |
245 | * from host network stack. | |
246 | */ | |
247 | features = netif_skb_features(skb); | |
248 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); | |
330966e5 | 249 | if (IS_ERR_OR_NULL(segs)) { |
c7ba65d7 FW |
250 | kfree_skb(skb); |
251 | return -ENOMEM; | |
252 | } | |
253 | ||
254 | consume_skb(skb); | |
255 | ||
256 | do { | |
257 | struct sk_buff *nskb = segs->next; | |
258 | int err; | |
259 | ||
260 | segs->next = NULL; | |
694869b3 | 261 | err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); |
c7ba65d7 FW |
262 | |
263 | if (err && ret == 0) | |
264 | ret = err; | |
265 | segs = nskb; | |
266 | } while (segs); | |
267 | ||
268 | return ret; | |
269 | } | |
270 | ||
0c4b51f0 | 271 | static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 272 | { |
c5501eb3 FW |
273 | unsigned int mtu; |
274 | ||
5c901daa PM |
275 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
276 | /* Policy lookup after SNAT yielded a new policy */ | |
00db4124 | 277 | if (skb_dst(skb)->xfrm) { |
48d5cad8 | 278 | IPCB(skb)->flags |= IPSKB_REROUTED; |
5a70649e | 279 | return dst_output(sk, skb); |
48d5cad8 | 280 | } |
5c901daa | 281 | #endif |
c5501eb3 | 282 | mtu = ip_skb_dst_mtu(skb); |
c7ba65d7 | 283 | if (skb_is_gso(skb)) |
694869b3 | 284 | return ip_finish_output_gso(net, sk, skb, mtu); |
c7ba65d7 | 285 | |
d6b915e2 | 286 | if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) |
694869b3 | 287 | return ip_fragment(net, sk, skb, mtu, ip_finish_output2); |
c7ba65d7 | 288 | |
694869b3 | 289 | return ip_finish_output2(net, sk, skb); |
1da177e4 LT |
290 | } |
291 | ||
aad88724 | 292 | int ip_mc_output(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 293 | { |
511c3f92 | 294 | struct rtable *rt = skb_rtable(skb); |
d8d1f30b | 295 | struct net_device *dev = rt->dst.dev; |
88f5cc24 | 296 | struct net *net = dev_net(dev); |
1da177e4 LT |
297 | |
298 | /* | |
299 | * If the indicated interface is up and running, send the packet. | |
300 | */ | |
88f5cc24 | 301 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
302 | |
303 | skb->dev = dev; | |
304 | skb->protocol = htons(ETH_P_IP); | |
305 | ||
306 | /* | |
307 | * Multicasts are looped back for other local users | |
308 | */ | |
309 | ||
310 | if (rt->rt_flags&RTCF_MULTICAST) { | |
7ad6848c | 311 | if (sk_mc_loop(sk) |
1da177e4 LT |
312 | #ifdef CONFIG_IP_MROUTE |
313 | /* Small optimization: do not loopback not local frames, | |
314 | which returned after forwarding; they will be dropped | |
315 | by ip_mr_input in any case. | |
316 | Note, that local frames are looped back to be delivered | |
317 | to local recipients. | |
318 | ||
319 | This check is duplicated in ip_mr_input at the moment. | |
320 | */ | |
9d4fb27d JP |
321 | && |
322 | ((rt->rt_flags & RTCF_LOCAL) || | |
323 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 324 | #endif |
9d4fb27d | 325 | ) { |
1da177e4 LT |
326 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
327 | if (newskb) | |
9bbc768a | 328 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
29a26a56 | 329 | net, sk, newskb, NULL, newskb->dev, |
95603e22 | 330 | dev_loopback_xmit); |
1da177e4 LT |
331 | } |
332 | ||
333 | /* Multicasts with ttl 0 must not go beyond the host */ | |
334 | ||
eddc9ec5 | 335 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
336 | kfree_skb(skb); |
337 | return 0; | |
338 | } | |
339 | } | |
340 | ||
341 | if (rt->rt_flags&RTCF_BROADCAST) { | |
342 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
343 | if (newskb) | |
29a26a56 EB |
344 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
345 | net, sk, newskb, NULL, newskb->dev, | |
346 | dev_loopback_xmit); | |
1da177e4 LT |
347 | } |
348 | ||
29a26a56 EB |
349 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
350 | net, sk, skb, NULL, skb->dev, | |
351 | ip_finish_output, | |
48d5cad8 | 352 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
353 | } |
354 | ||
aad88724 | 355 | int ip_output(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 356 | { |
adf30907 | 357 | struct net_device *dev = skb_dst(skb)->dev; |
88f5cc24 | 358 | struct net *net = dev_net(dev); |
1bd9bef6 | 359 | |
88f5cc24 | 360 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 361 | |
1bd9bef6 PM |
362 | skb->dev = dev; |
363 | skb->protocol = htons(ETH_P_IP); | |
364 | ||
29a26a56 EB |
365 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
366 | net, sk, skb, NULL, dev, | |
e905a9ed | 367 | ip_finish_output, |
48d5cad8 | 368 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
369 | } |
370 | ||
84f9307c ED |
371 | /* |
372 | * copy saddr and daddr, possibly using 64bit load/stores | |
373 | * Equivalent to : | |
374 | * iph->saddr = fl4->saddr; | |
375 | * iph->daddr = fl4->daddr; | |
376 | */ | |
377 | static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) | |
378 | { | |
379 | BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != | |
380 | offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); | |
381 | memcpy(&iph->saddr, &fl4->saddr, | |
382 | sizeof(fl4->saddr) + sizeof(fl4->daddr)); | |
383 | } | |
384 | ||
b0270e91 ED |
385 | /* Note: skb->sk can be different from sk, in case of tunnels */ |
386 | int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) | |
1da177e4 | 387 | { |
1da177e4 | 388 | struct inet_sock *inet = inet_sk(sk); |
f6d8bd05 | 389 | struct ip_options_rcu *inet_opt; |
b57ae01a | 390 | struct flowi4 *fl4; |
1da177e4 LT |
391 | struct rtable *rt; |
392 | struct iphdr *iph; | |
ab6e3feb | 393 | int res; |
1da177e4 LT |
394 | |
395 | /* Skip all of this if the packet is already routed, | |
396 | * f.e. by something like SCTP. | |
397 | */ | |
ab6e3feb | 398 | rcu_read_lock(); |
f6d8bd05 | 399 | inet_opt = rcu_dereference(inet->inet_opt); |
ea4fc0d6 | 400 | fl4 = &fl->u.ip4; |
511c3f92 | 401 | rt = skb_rtable(skb); |
00db4124 | 402 | if (rt) |
1da177e4 LT |
403 | goto packet_routed; |
404 | ||
405 | /* Make sure we can route this packet. */ | |
406 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
51456b29 | 407 | if (!rt) { |
3ca3c68e | 408 | __be32 daddr; |
1da177e4 LT |
409 | |
410 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 411 | daddr = inet->inet_daddr; |
f6d8bd05 ED |
412 | if (inet_opt && inet_opt->opt.srr) |
413 | daddr = inet_opt->opt.faddr; | |
1da177e4 | 414 | |
78fbfd8a DM |
415 | /* If this fails, retransmit mechanism of transport layer will |
416 | * keep trying until route appears or the connection times | |
417 | * itself out. | |
418 | */ | |
b57ae01a | 419 | rt = ip_route_output_ports(sock_net(sk), fl4, sk, |
78fbfd8a DM |
420 | daddr, inet->inet_saddr, |
421 | inet->inet_dport, | |
422 | inet->inet_sport, | |
423 | sk->sk_protocol, | |
424 | RT_CONN_FLAGS(sk), | |
425 | sk->sk_bound_dev_if); | |
426 | if (IS_ERR(rt)) | |
427 | goto no_route; | |
d8d1f30b | 428 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 429 | } |
d8d1f30b | 430 | skb_dst_set_noref(skb, &rt->dst); |
1da177e4 LT |
431 | |
432 | packet_routed: | |
155e8336 | 433 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) |
1da177e4 LT |
434 | goto no_route; |
435 | ||
436 | /* OK, we know where to send it, allocate and build IP header. */ | |
f6d8bd05 | 437 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); |
8856dfa3 | 438 | skb_reset_network_header(skb); |
eddc9ec5 | 439 | iph = ip_hdr(skb); |
714e85be | 440 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
60ff7467 | 441 | if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) |
1da177e4 LT |
442 | iph->frag_off = htons(IP_DF); |
443 | else | |
444 | iph->frag_off = 0; | |
d8d1f30b | 445 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 446 | iph->protocol = sk->sk_protocol; |
84f9307c ED |
447 | ip_copy_addrs(iph, fl4); |
448 | ||
1da177e4 LT |
449 | /* Transport layer set skb->h.foo itself. */ |
450 | ||
f6d8bd05 ED |
451 | if (inet_opt && inet_opt->opt.optlen) { |
452 | iph->ihl += inet_opt->opt.optlen >> 2; | |
453 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
1da177e4 LT |
454 | } |
455 | ||
b6a7719a HFS |
456 | ip_select_ident_segs(sock_net(sk), skb, sk, |
457 | skb_shinfo(skb)->gso_segs ?: 1); | |
1da177e4 | 458 | |
b0270e91 | 459 | /* TODO : should we use skb->sk here instead of sk ? */ |
1da177e4 | 460 | skb->priority = sk->sk_priority; |
4a19ec58 | 461 | skb->mark = sk->sk_mark; |
1da177e4 | 462 | |
850dcc4d | 463 | res = ip_local_out_sk(sk, skb); |
ab6e3feb ED |
464 | rcu_read_unlock(); |
465 | return res; | |
1da177e4 LT |
466 | |
467 | no_route: | |
ab6e3feb | 468 | rcu_read_unlock(); |
5e38e270 | 469 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
470 | kfree_skb(skb); |
471 | return -EHOSTUNREACH; | |
472 | } | |
4bc2f18b | 473 | EXPORT_SYMBOL(ip_queue_xmit); |
1da177e4 | 474 | |
1da177e4 LT |
475 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) |
476 | { | |
477 | to->pkt_type = from->pkt_type; | |
478 | to->priority = from->priority; | |
479 | to->protocol = from->protocol; | |
adf30907 | 480 | skb_dst_drop(to); |
fe76cda3 | 481 | skb_dst_copy(to, from); |
1da177e4 | 482 | to->dev = from->dev; |
82e91ffe | 483 | to->mark = from->mark; |
1da177e4 LT |
484 | |
485 | /* Copy the flags to each fragment. */ | |
486 | IPCB(to)->flags = IPCB(from)->flags; | |
487 | ||
488 | #ifdef CONFIG_NET_SCHED | |
489 | to->tc_index = from->tc_index; | |
490 | #endif | |
e7ac05f3 | 491 | nf_copy(to, from); |
c98d80ed JA |
492 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
493 | to->ipvs_property = from->ipvs_property; | |
1da177e4 | 494 | #endif |
984bc16c | 495 | skb_copy_secmark(to, from); |
1da177e4 LT |
496 | } |
497 | ||
694869b3 | 498 | static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
c5501eb3 | 499 | unsigned int mtu, |
694869b3 | 500 | int (*output)(struct net *, struct sock *, struct sk_buff *)) |
49d16b23 AZ |
501 | { |
502 | struct iphdr *iph = ip_hdr(skb); | |
49d16b23 | 503 | |
d6b915e2 | 504 | if ((iph->frag_off & htons(IP_DF)) == 0) |
694869b3 | 505 | return ip_do_fragment(net, sk, skb, output); |
d6b915e2 FW |
506 | |
507 | if (unlikely(!skb->ignore_df || | |
49d16b23 AZ |
508 | (IPCB(skb)->frag_max_size && |
509 | IPCB(skb)->frag_max_size > mtu))) { | |
9479b0af | 510 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
49d16b23 AZ |
511 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
512 | htonl(mtu)); | |
513 | kfree_skb(skb); | |
514 | return -EMSGSIZE; | |
515 | } | |
516 | ||
694869b3 | 517 | return ip_do_fragment(net, sk, skb, output); |
49d16b23 AZ |
518 | } |
519 | ||
1da177e4 LT |
520 | /* |
521 | * This IP datagram is too large to be sent in one piece. Break it up into | |
522 | * smaller pieces (each of size equal to IP header plus | |
523 | * a block of the data of the original IP data part) that will yet fit in a | |
524 | * single device frame, and queue such a frame for sending. | |
525 | */ | |
526 | ||
694869b3 EB |
527 | int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
528 | int (*output)(struct net *, struct sock *, struct sk_buff *)) | |
1da177e4 LT |
529 | { |
530 | struct iphdr *iph; | |
1da177e4 LT |
531 | int ptr; |
532 | struct net_device *dev; | |
533 | struct sk_buff *skb2; | |
c893b806 | 534 | unsigned int mtu, hlen, left, len, ll_rs; |
1da177e4 | 535 | int offset; |
76ab608d | 536 | __be16 not_last_frag; |
511c3f92 | 537 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
538 | int err = 0; |
539 | ||
d8d1f30b | 540 | dev = rt->dst.dev; |
1da177e4 LT |
541 | |
542 | /* | |
543 | * Point into the IP datagram header. | |
544 | */ | |
545 | ||
eddc9ec5 | 546 | iph = ip_hdr(skb); |
1da177e4 | 547 | |
69647ce4 | 548 | mtu = ip_skb_dst_mtu(skb); |
d6b915e2 FW |
549 | if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) |
550 | mtu = IPCB(skb)->frag_max_size; | |
1da177e4 LT |
551 | |
552 | /* | |
553 | * Setup starting values. | |
554 | */ | |
555 | ||
556 | hlen = iph->ihl * 4; | |
f87c10a8 | 557 | mtu = mtu - hlen; /* Size of data space */ |
89cee8b1 | 558 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
559 | |
560 | /* When frag_list is given, use it. First, check its validity: | |
561 | * some transformers could create wrong frag_list or break existing | |
562 | * one, it is not prohibited. In this case fall back to copying. | |
563 | * | |
564 | * LATER: this step can be merged to real generation of fragments, | |
565 | * we can switch to copy when see the first bad fragment. | |
566 | */ | |
21dc3301 | 567 | if (skb_has_frag_list(skb)) { |
3d13008e | 568 | struct sk_buff *frag, *frag2; |
1da177e4 LT |
569 | int first_len = skb_pagelen(skb); |
570 | ||
571 | if (first_len - hlen > mtu || | |
572 | ((first_len - hlen) & 7) || | |
56f8a75c | 573 | ip_is_fragment(iph) || |
1da177e4 LT |
574 | skb_cloned(skb)) |
575 | goto slow_path; | |
576 | ||
d7fcf1a5 | 577 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
578 | /* Correct geometry. */ |
579 | if (frag->len > mtu || | |
580 | ((frag->len & 7) && frag->next) || | |
581 | skb_headroom(frag) < hlen) | |
3d13008e | 582 | goto slow_path_clean; |
1da177e4 LT |
583 | |
584 | /* Partially cloned skb? */ | |
585 | if (skb_shared(frag)) | |
3d13008e | 586 | goto slow_path_clean; |
2fdba6b0 HX |
587 | |
588 | BUG_ON(frag->sk); | |
589 | if (skb->sk) { | |
2fdba6b0 HX |
590 | frag->sk = skb->sk; |
591 | frag->destructor = sock_wfree; | |
2fdba6b0 | 592 | } |
3d13008e | 593 | skb->truesize -= frag->truesize; |
1da177e4 LT |
594 | } |
595 | ||
596 | /* Everything is OK. Generate! */ | |
597 | ||
598 | err = 0; | |
599 | offset = 0; | |
600 | frag = skb_shinfo(skb)->frag_list; | |
d7fcf1a5 | 601 | skb_frag_list_init(skb); |
1da177e4 LT |
602 | skb->data_len = first_len - skb_headlen(skb); |
603 | skb->len = first_len; | |
604 | iph->tot_len = htons(first_len); | |
605 | iph->frag_off = htons(IP_MF); | |
606 | ip_send_check(iph); | |
607 | ||
608 | for (;;) { | |
609 | /* Prepare header of the next frame, | |
610 | * before previous one went down. */ | |
611 | if (frag) { | |
612 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 613 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
614 | __skb_push(frag, hlen); |
615 | skb_reset_network_header(frag); | |
d56f90a7 | 616 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 617 | iph = ip_hdr(frag); |
1da177e4 LT |
618 | iph->tot_len = htons(frag->len); |
619 | ip_copy_metadata(frag, skb); | |
620 | if (offset == 0) | |
621 | ip_options_fragment(frag); | |
622 | offset += skb->len - hlen; | |
623 | iph->frag_off = htons(offset>>3); | |
00db4124 | 624 | if (frag->next) |
1da177e4 LT |
625 | iph->frag_off |= htons(IP_MF); |
626 | /* Ready, complete checksum */ | |
627 | ip_send_check(iph); | |
628 | } | |
629 | ||
694869b3 | 630 | err = output(net, sk, skb); |
1da177e4 | 631 | |
dafee490 | 632 | if (!err) |
26a949db | 633 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
634 | if (err || !frag) |
635 | break; | |
636 | ||
637 | skb = frag; | |
638 | frag = skb->next; | |
639 | skb->next = NULL; | |
640 | } | |
641 | ||
642 | if (err == 0) { | |
26a949db | 643 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
644 | return 0; |
645 | } | |
646 | ||
647 | while (frag) { | |
648 | skb = frag->next; | |
649 | kfree_skb(frag); | |
650 | frag = skb; | |
651 | } | |
26a949db | 652 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 653 | return err; |
3d13008e ED |
654 | |
655 | slow_path_clean: | |
656 | skb_walk_frags(skb, frag2) { | |
657 | if (frag2 == frag) | |
658 | break; | |
659 | frag2->sk = NULL; | |
660 | frag2->destructor = NULL; | |
661 | skb->truesize += frag2->truesize; | |
662 | } | |
1da177e4 LT |
663 | } |
664 | ||
665 | slow_path: | |
fc70fb64 AD |
666 | /* for offloaded checksums cleanup checksum before fragmentation */ |
667 | if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb)) | |
668 | goto fail; | |
c9af6db4 | 669 | iph = ip_hdr(skb); |
fc70fb64 | 670 | |
1da177e4 | 671 | left = skb->len - hlen; /* Space per frame */ |
49085bd7 | 672 | ptr = hlen; /* Where to start from */ |
1da177e4 | 673 | |
8d045163 | 674 | ll_rs = LL_RESERVED_SPACE(rt->dst.dev); |
9bcfcaf5 | 675 | |
1da177e4 LT |
676 | /* |
677 | * Fragment the datagram. | |
678 | */ | |
679 | ||
680 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
681 | not_last_frag = iph->frag_off & htons(IP_MF); | |
682 | ||
683 | /* | |
684 | * Keep copying data until we run out. | |
685 | */ | |
686 | ||
132adf54 | 687 | while (left > 0) { |
1da177e4 LT |
688 | len = left; |
689 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
690 | if (len > mtu) | |
691 | len = mtu; | |
25985edc | 692 | /* IF: we are not sending up to and including the packet end |
1da177e4 LT |
693 | then align the next start on an eight byte boundary */ |
694 | if (len < left) { | |
695 | len &= ~7; | |
696 | } | |
1da177e4 | 697 | |
cbffccc9 JP |
698 | /* Allocate buffer */ |
699 | skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); | |
700 | if (!skb2) { | |
1da177e4 LT |
701 | err = -ENOMEM; |
702 | goto fail; | |
703 | } | |
704 | ||
705 | /* | |
706 | * Set up data on packet | |
707 | */ | |
708 | ||
709 | ip_copy_metadata(skb2, skb); | |
710 | skb_reserve(skb2, ll_rs); | |
711 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 712 | skb_reset_network_header(skb2); |
b0e380b1 | 713 | skb2->transport_header = skb2->network_header + hlen; |
1da177e4 LT |
714 | |
715 | /* | |
716 | * Charge the memory for the fragment to any owner | |
717 | * it might possess | |
718 | */ | |
719 | ||
720 | if (skb->sk) | |
721 | skb_set_owner_w(skb2, skb->sk); | |
722 | ||
723 | /* | |
724 | * Copy the packet header into the new buffer. | |
725 | */ | |
726 | ||
d626f62b | 727 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); |
1da177e4 LT |
728 | |
729 | /* | |
730 | * Copy a block of the IP datagram. | |
731 | */ | |
bff9b61c | 732 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) |
1da177e4 LT |
733 | BUG(); |
734 | left -= len; | |
735 | ||
736 | /* | |
737 | * Fill in the new header fields. | |
738 | */ | |
eddc9ec5 | 739 | iph = ip_hdr(skb2); |
1da177e4 LT |
740 | iph->frag_off = htons((offset >> 3)); |
741 | ||
d6b915e2 FW |
742 | if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) |
743 | iph->frag_off |= htons(IP_DF); | |
744 | ||
1da177e4 LT |
745 | /* ANK: dirty, but effective trick. Upgrade options only if |
746 | * the segment to be fragmented was THE FIRST (otherwise, | |
747 | * options are already fixed) and make it ONCE | |
748 | * on the initial skb, so that all the following fragments | |
749 | * will inherit fixed options. | |
750 | */ | |
751 | if (offset == 0) | |
752 | ip_options_fragment(skb); | |
753 | ||
754 | /* | |
755 | * Added AC : If we are fragmenting a fragment that's not the | |
756 | * last fragment then keep MF on each bit | |
757 | */ | |
758 | if (left > 0 || not_last_frag) | |
759 | iph->frag_off |= htons(IP_MF); | |
760 | ptr += len; | |
761 | offset += len; | |
762 | ||
763 | /* | |
764 | * Put this fragment into the sending queue. | |
765 | */ | |
1da177e4 LT |
766 | iph->tot_len = htons(len + hlen); |
767 | ||
768 | ip_send_check(iph); | |
769 | ||
694869b3 | 770 | err = output(net, sk, skb2); |
1da177e4 LT |
771 | if (err) |
772 | goto fail; | |
dafee490 | 773 | |
26a949db | 774 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
1da177e4 | 775 | } |
5d0ba55b | 776 | consume_skb(skb); |
26a949db | 777 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
778 | return err; |
779 | ||
780 | fail: | |
e905a9ed | 781 | kfree_skb(skb); |
26a949db | 782 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
783 | return err; |
784 | } | |
49d16b23 | 785 | EXPORT_SYMBOL(ip_do_fragment); |
2e2f7aef | 786 | |
1da177e4 LT |
787 | int |
788 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
789 | { | |
f69e6d13 | 790 | struct msghdr *msg = from; |
1da177e4 | 791 | |
84fa7933 | 792 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
21226abb | 793 | if (copy_from_iter(to, len, &msg->msg_iter) != len) |
1da177e4 LT |
794 | return -EFAULT; |
795 | } else { | |
44bb9363 | 796 | __wsum csum = 0; |
21226abb | 797 | if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len) |
1da177e4 LT |
798 | return -EFAULT; |
799 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
800 | } | |
801 | return 0; | |
802 | } | |
4bc2f18b | 803 | EXPORT_SYMBOL(ip_generic_getfrag); |
1da177e4 | 804 | |
44bb9363 | 805 | static inline __wsum |
1da177e4 LT |
806 | csum_page(struct page *page, int offset, int copy) |
807 | { | |
808 | char *kaddr; | |
44bb9363 | 809 | __wsum csum; |
1da177e4 LT |
810 | kaddr = kmap(page); |
811 | csum = csum_partial(kaddr + offset, copy, 0); | |
812 | kunmap(page); | |
813 | return csum; | |
814 | } | |
815 | ||
4b30b1c6 | 816 | static inline int ip_ufo_append_data(struct sock *sk, |
1470ddf7 | 817 | struct sk_buff_head *queue, |
e89e9cf5 AR |
818 | int getfrag(void *from, char *to, int offset, int len, |
819 | int odd, struct sk_buff *skb), | |
820 | void *from, int length, int hh_len, int fragheaderlen, | |
d9be4f7a | 821 | int transhdrlen, int maxfraglen, unsigned int flags) |
e89e9cf5 AR |
822 | { |
823 | struct sk_buff *skb; | |
824 | int err; | |
825 | ||
826 | /* There is support for UDP fragmentation offload by network | |
827 | * device, so create one single skb packet containing complete | |
828 | * udp datagram | |
829 | */ | |
51456b29 IM |
830 | skb = skb_peek_tail(queue); |
831 | if (!skb) { | |
e89e9cf5 AR |
832 | skb = sock_alloc_send_skb(sk, |
833 | hh_len + fragheaderlen + transhdrlen + 20, | |
834 | (flags & MSG_DONTWAIT), &err); | |
835 | ||
51456b29 | 836 | if (!skb) |
e89e9cf5 AR |
837 | return err; |
838 | ||
839 | /* reserve space for Hardware header */ | |
840 | skb_reserve(skb, hh_len); | |
841 | ||
842 | /* create space for UDP/IP header */ | |
d9319100 | 843 | skb_put(skb, fragheaderlen + transhdrlen); |
e89e9cf5 AR |
844 | |
845 | /* initialize network header pointer */ | |
c1d2bbe1 | 846 | skb_reset_network_header(skb); |
e89e9cf5 AR |
847 | |
848 | /* initialize protocol header pointer */ | |
b0e380b1 | 849 | skb->transport_header = skb->network_header + fragheaderlen; |
e89e9cf5 | 850 | |
e89e9cf5 | 851 | skb->csum = 0; |
e89e9cf5 | 852 | |
1470ddf7 | 853 | __skb_queue_tail(queue, skb); |
e93b7d74 JP |
854 | } else if (skb_is_gso(skb)) { |
855 | goto append; | |
e89e9cf5 | 856 | } |
be9164e7 | 857 | |
e93b7d74 JP |
858 | skb->ip_summed = CHECKSUM_PARTIAL; |
859 | /* specify the length of each IP datagram fragment */ | |
860 | skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; | |
861 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
862 | ||
863 | append: | |
be9164e7 K |
864 | return skb_append_datato_frags(sk, skb, getfrag, from, |
865 | (length - transhdrlen)); | |
e89e9cf5 AR |
866 | } |
867 | ||
f5fca608 DM |
868 | static int __ip_append_data(struct sock *sk, |
869 | struct flowi4 *fl4, | |
870 | struct sk_buff_head *queue, | |
1470ddf7 | 871 | struct inet_cork *cork, |
5640f768 | 872 | struct page_frag *pfrag, |
1470ddf7 HX |
873 | int getfrag(void *from, char *to, int offset, |
874 | int len, int odd, struct sk_buff *skb), | |
875 | void *from, int length, int transhdrlen, | |
876 | unsigned int flags) | |
1da177e4 LT |
877 | { |
878 | struct inet_sock *inet = inet_sk(sk); | |
879 | struct sk_buff *skb; | |
880 | ||
07df5294 | 881 | struct ip_options *opt = cork->opt; |
1da177e4 LT |
882 | int hh_len; |
883 | int exthdrlen; | |
884 | int mtu; | |
885 | int copy; | |
886 | int err; | |
887 | int offset = 0; | |
daba287b | 888 | unsigned int maxfraglen, fragheaderlen, maxnonfragsize; |
1da177e4 | 889 | int csummode = CHECKSUM_NONE; |
1470ddf7 | 890 | struct rtable *rt = (struct rtable *)cork->dst; |
09c2d251 | 891 | u32 tskey = 0; |
1da177e4 | 892 | |
96d7303e SK |
893 | skb = skb_peek_tail(queue); |
894 | ||
895 | exthdrlen = !skb ? rt->dst.header_len : 0; | |
07df5294 | 896 | mtu = cork->fragsize; |
09c2d251 WB |
897 | if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && |
898 | sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) | |
899 | tskey = sk->sk_tskey++; | |
1da177e4 | 900 | |
d8d1f30b | 901 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
902 | |
903 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
904 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 905 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 906 | |
daba287b | 907 | if (cork->length + length > maxnonfragsize - fragheaderlen) { |
f5fca608 | 908 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
61e7f09d | 909 | mtu - (opt ? opt->optlen : 0)); |
1da177e4 LT |
910 | return -EMSGSIZE; |
911 | } | |
912 | ||
913 | /* | |
914 | * transhdrlen > 0 means that this is the first fragment and we wish | |
915 | * it won't be fragmented in the future. | |
916 | */ | |
917 | if (transhdrlen && | |
918 | length + fragheaderlen <= mtu && | |
d8d1f30b | 919 | rt->dst.dev->features & NETIF_F_V4_CSUM && |
1da177e4 | 920 | !exthdrlen) |
84fa7933 | 921 | csummode = CHECKSUM_PARTIAL; |
1da177e4 | 922 | |
1470ddf7 | 923 | cork->length += length; |
26cde9f7 | 924 | if (((length > mtu) || (skb && skb_is_gso(skb))) && |
be9164e7 | 925 | (sk->sk_protocol == IPPROTO_UDP) && |
acf8dd0a MK |
926 | (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len && |
927 | (sk->sk_type == SOCK_DGRAM)) { | |
1470ddf7 HX |
928 | err = ip_ufo_append_data(sk, queue, getfrag, from, length, |
929 | hh_len, fragheaderlen, transhdrlen, | |
d9be4f7a | 930 | maxfraglen, flags); |
baa829d8 | 931 | if (err) |
e89e9cf5 | 932 | goto error; |
e89e9cf5 AR |
933 | return 0; |
934 | } | |
1da177e4 LT |
935 | |
936 | /* So, what's going on in the loop below? | |
937 | * | |
938 | * We use calculated fragment length to generate chained skb, | |
939 | * each of segments is IP fragment ready for sending to network after | |
940 | * adding appropriate IP header. | |
941 | */ | |
942 | ||
26cde9f7 | 943 | if (!skb) |
1da177e4 LT |
944 | goto alloc_new_skb; |
945 | ||
946 | while (length > 0) { | |
947 | /* Check if the remaining data fits into current packet. */ | |
948 | copy = mtu - skb->len; | |
949 | if (copy < length) | |
950 | copy = maxfraglen - skb->len; | |
951 | if (copy <= 0) { | |
952 | char *data; | |
953 | unsigned int datalen; | |
954 | unsigned int fraglen; | |
955 | unsigned int fraggap; | |
956 | unsigned int alloclen; | |
957 | struct sk_buff *skb_prev; | |
958 | alloc_new_skb: | |
959 | skb_prev = skb; | |
960 | if (skb_prev) | |
961 | fraggap = skb_prev->len - maxfraglen; | |
962 | else | |
963 | fraggap = 0; | |
964 | ||
965 | /* | |
966 | * If remaining data exceeds the mtu, | |
967 | * we know we need more fragment(s). | |
968 | */ | |
969 | datalen = length + fraggap; | |
970 | if (datalen > mtu - fragheaderlen) | |
971 | datalen = maxfraglen - fragheaderlen; | |
972 | fraglen = datalen + fragheaderlen; | |
973 | ||
e905a9ed | 974 | if ((flags & MSG_MORE) && |
d8d1f30b | 975 | !(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
976 | alloclen = mtu; |
977 | else | |
59104f06 | 978 | alloclen = fraglen; |
1da177e4 | 979 | |
353e5c9a SK |
980 | alloclen += exthdrlen; |
981 | ||
1da177e4 LT |
982 | /* The last fragment gets additional space at tail. |
983 | * Note, with MSG_MORE we overallocate on fragments, | |
984 | * because we have no idea what fragment will be | |
985 | * the last. | |
986 | */ | |
33f99dc7 | 987 | if (datalen == length + fraggap) |
d8d1f30b | 988 | alloclen += rt->dst.trailer_len; |
33f99dc7 | 989 | |
1da177e4 | 990 | if (transhdrlen) { |
e905a9ed | 991 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
992 | alloclen + hh_len + 15, |
993 | (flags & MSG_DONTWAIT), &err); | |
994 | } else { | |
995 | skb = NULL; | |
996 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
997 | 2 * sk->sk_sndbuf) | |
e905a9ed | 998 | skb = sock_wmalloc(sk, |
1da177e4 LT |
999 | alloclen + hh_len + 15, 1, |
1000 | sk->sk_allocation); | |
51456b29 | 1001 | if (unlikely(!skb)) |
1da177e4 LT |
1002 | err = -ENOBUFS; |
1003 | } | |
51456b29 | 1004 | if (!skb) |
1da177e4 LT |
1005 | goto error; |
1006 | ||
1007 | /* | |
1008 | * Fill in the control structures | |
1009 | */ | |
1010 | skb->ip_summed = csummode; | |
1011 | skb->csum = 0; | |
1012 | skb_reserve(skb, hh_len); | |
11878b40 WB |
1013 | |
1014 | /* only the initial fragment is time stamped */ | |
1470ddf7 | 1015 | skb_shinfo(skb)->tx_flags = cork->tx_flags; |
11878b40 | 1016 | cork->tx_flags = 0; |
09c2d251 WB |
1017 | skb_shinfo(skb)->tskey = tskey; |
1018 | tskey = 0; | |
1da177e4 LT |
1019 | |
1020 | /* | |
1021 | * Find where to start putting bytes. | |
1022 | */ | |
353e5c9a | 1023 | data = skb_put(skb, fraglen + exthdrlen); |
c14d2450 | 1024 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
1025 | skb->transport_header = (skb->network_header + |
1026 | fragheaderlen); | |
353e5c9a | 1027 | data += fragheaderlen + exthdrlen; |
1da177e4 LT |
1028 | |
1029 | if (fraggap) { | |
1030 | skb->csum = skb_copy_and_csum_bits( | |
1031 | skb_prev, maxfraglen, | |
1032 | data + transhdrlen, fraggap, 0); | |
1033 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1034 | skb->csum); | |
1035 | data += fraggap; | |
e9fa4f7b | 1036 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1037 | } |
1038 | ||
1039 | copy = datalen - transhdrlen - fraggap; | |
1040 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
1041 | err = -EFAULT; | |
1042 | kfree_skb(skb); | |
1043 | goto error; | |
1044 | } | |
1045 | ||
1046 | offset += copy; | |
1047 | length -= datalen - fraggap; | |
1048 | transhdrlen = 0; | |
1049 | exthdrlen = 0; | |
1050 | csummode = CHECKSUM_NONE; | |
1051 | ||
1052 | /* | |
1053 | * Put the packet on the pending queue. | |
1054 | */ | |
1470ddf7 | 1055 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
1056 | continue; |
1057 | } | |
1058 | ||
1059 | if (copy > length) | |
1060 | copy = length; | |
1061 | ||
d8d1f30b | 1062 | if (!(rt->dst.dev->features&NETIF_F_SG)) { |
1da177e4 LT |
1063 | unsigned int off; |
1064 | ||
1065 | off = skb->len; | |
e905a9ed | 1066 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
1067 | offset, copy, off, skb) < 0) { |
1068 | __skb_trim(skb, off); | |
1069 | err = -EFAULT; | |
1070 | goto error; | |
1071 | } | |
1072 | } else { | |
1073 | int i = skb_shinfo(skb)->nr_frags; | |
1da177e4 | 1074 | |
5640f768 ED |
1075 | err = -ENOMEM; |
1076 | if (!sk_page_frag_refill(sk, pfrag)) | |
1da177e4 | 1077 | goto error; |
5640f768 ED |
1078 | |
1079 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1080 | pfrag->offset)) { | |
1081 | err = -EMSGSIZE; | |
1082 | if (i == MAX_SKB_FRAGS) | |
1083 | goto error; | |
1084 | ||
1085 | __skb_fill_page_desc(skb, i, pfrag->page, | |
1086 | pfrag->offset, 0); | |
1087 | skb_shinfo(skb)->nr_frags = ++i; | |
1088 | get_page(pfrag->page); | |
1da177e4 | 1089 | } |
5640f768 ED |
1090 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1091 | if (getfrag(from, | |
1092 | page_address(pfrag->page) + pfrag->offset, | |
1093 | offset, copy, skb->len, skb) < 0) | |
1094 | goto error_efault; | |
1095 | ||
1096 | pfrag->offset += copy; | |
1097 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1da177e4 LT |
1098 | skb->len += copy; |
1099 | skb->data_len += copy; | |
f945fa7a HX |
1100 | skb->truesize += copy; |
1101 | atomic_add(copy, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1102 | } |
1103 | offset += copy; | |
1104 | length -= copy; | |
1105 | } | |
1106 | ||
1107 | return 0; | |
1108 | ||
5640f768 ED |
1109 | error_efault: |
1110 | err = -EFAULT; | |
1da177e4 | 1111 | error: |
1470ddf7 | 1112 | cork->length -= length; |
5e38e270 | 1113 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
e905a9ed | 1114 | return err; |
1da177e4 LT |
1115 | } |
1116 | ||
1470ddf7 HX |
1117 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, |
1118 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1119 | { | |
f6d8bd05 | 1120 | struct ip_options_rcu *opt; |
1470ddf7 HX |
1121 | struct rtable *rt; |
1122 | ||
1123 | /* | |
1124 | * setup for corking. | |
1125 | */ | |
1126 | opt = ipc->opt; | |
1127 | if (opt) { | |
51456b29 | 1128 | if (!cork->opt) { |
1470ddf7 HX |
1129 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, |
1130 | sk->sk_allocation); | |
51456b29 | 1131 | if (unlikely(!cork->opt)) |
1470ddf7 HX |
1132 | return -ENOBUFS; |
1133 | } | |
f6d8bd05 | 1134 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); |
1470ddf7 HX |
1135 | cork->flags |= IPCORK_OPT; |
1136 | cork->addr = ipc->addr; | |
1137 | } | |
1138 | rt = *rtp; | |
1139 | if (unlikely(!rt)) | |
1140 | return -EFAULT; | |
1141 | /* | |
1142 | * We steal reference to this route, caller should not release it | |
1143 | */ | |
1144 | *rtp = NULL; | |
482fc609 HFS |
1145 | cork->fragsize = ip_sk_use_pmtu(sk) ? |
1146 | dst_mtu(&rt->dst) : rt->dst.dev->mtu; | |
1470ddf7 HX |
1147 | cork->dst = &rt->dst; |
1148 | cork->length = 0; | |
aa661581 FF |
1149 | cork->ttl = ipc->ttl; |
1150 | cork->tos = ipc->tos; | |
1151 | cork->priority = ipc->priority; | |
1470ddf7 | 1152 | cork->tx_flags = ipc->tx_flags; |
1470ddf7 HX |
1153 | |
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1159 | * from many pieces of data. Each pieces will be holded on the socket | |
1160 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1161 | * or non-page data. | |
1162 | * | |
1163 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1164 | * this interface potentially. | |
1165 | * | |
1166 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1167 | */ | |
f5fca608 | 1168 | int ip_append_data(struct sock *sk, struct flowi4 *fl4, |
1470ddf7 HX |
1169 | int getfrag(void *from, char *to, int offset, int len, |
1170 | int odd, struct sk_buff *skb), | |
1171 | void *from, int length, int transhdrlen, | |
1172 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1173 | unsigned int flags) | |
1174 | { | |
1175 | struct inet_sock *inet = inet_sk(sk); | |
1176 | int err; | |
1177 | ||
1178 | if (flags&MSG_PROBE) | |
1179 | return 0; | |
1180 | ||
1181 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
bdc712b4 | 1182 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); |
1470ddf7 HX |
1183 | if (err) |
1184 | return err; | |
1185 | } else { | |
1186 | transhdrlen = 0; | |
1187 | } | |
1188 | ||
5640f768 ED |
1189 | return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, |
1190 | sk_page_frag(sk), getfrag, | |
1470ddf7 HX |
1191 | from, length, transhdrlen, flags); |
1192 | } | |
1193 | ||
f5fca608 | 1194 | ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, |
1da177e4 LT |
1195 | int offset, size_t size, int flags) |
1196 | { | |
1197 | struct inet_sock *inet = inet_sk(sk); | |
1198 | struct sk_buff *skb; | |
1199 | struct rtable *rt; | |
1200 | struct ip_options *opt = NULL; | |
bdc712b4 | 1201 | struct inet_cork *cork; |
1da177e4 LT |
1202 | int hh_len; |
1203 | int mtu; | |
1204 | int len; | |
1205 | int err; | |
daba287b | 1206 | unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; |
1da177e4 LT |
1207 | |
1208 | if (inet->hdrincl) | |
1209 | return -EPERM; | |
1210 | ||
1211 | if (flags&MSG_PROBE) | |
1212 | return 0; | |
1213 | ||
1214 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1215 | return -EINVAL; | |
1216 | ||
bdc712b4 DM |
1217 | cork = &inet->cork.base; |
1218 | rt = (struct rtable *)cork->dst; | |
1219 | if (cork->flags & IPCORK_OPT) | |
1220 | opt = cork->opt; | |
1da177e4 | 1221 | |
d8d1f30b | 1222 | if (!(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
1223 | return -EOPNOTSUPP; |
1224 | ||
d8d1f30b | 1225 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
bdc712b4 | 1226 | mtu = cork->fragsize; |
1da177e4 LT |
1227 | |
1228 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1229 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 1230 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 1231 | |
daba287b | 1232 | if (cork->length + size > maxnonfragsize - fragheaderlen) { |
61e7f09d HFS |
1233 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
1234 | mtu - (opt ? opt->optlen : 0)); | |
1da177e4 LT |
1235 | return -EMSGSIZE; |
1236 | } | |
1237 | ||
51456b29 IM |
1238 | skb = skb_peek_tail(&sk->sk_write_queue); |
1239 | if (!skb) | |
1da177e4 LT |
1240 | return -EINVAL; |
1241 | ||
bdc712b4 | 1242 | cork->length += size; |
26cde9f7 HX |
1243 | if ((size + skb->len > mtu) && |
1244 | (sk->sk_protocol == IPPROTO_UDP) && | |
d8d1f30b | 1245 | (rt->dst.dev->features & NETIF_F_UFO)) { |
7967168c | 1246 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 1247 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1248 | } |
e89e9cf5 | 1249 | |
1da177e4 | 1250 | while (size > 0) { |
be12a1fe | 1251 | if (skb_is_gso(skb)) { |
e89e9cf5 | 1252 | len = size; |
be12a1fe | 1253 | } else { |
e89e9cf5 AR |
1254 | |
1255 | /* Check if the remaining data fits into current packet. */ | |
1256 | len = mtu - skb->len; | |
1257 | if (len < size) | |
1258 | len = maxfraglen - skb->len; | |
1259 | } | |
1da177e4 LT |
1260 | if (len <= 0) { |
1261 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1262 | int alloclen; |
1263 | ||
1264 | skb_prev = skb; | |
0d0d2bba | 1265 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1266 | |
1267 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1268 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1269 | if (unlikely(!skb)) { | |
1270 | err = -ENOBUFS; | |
1271 | goto error; | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * Fill in the control structures | |
1276 | */ | |
1277 | skb->ip_summed = CHECKSUM_NONE; | |
1278 | skb->csum = 0; | |
1279 | skb_reserve(skb, hh_len); | |
1280 | ||
1281 | /* | |
1282 | * Find where to start putting bytes. | |
1283 | */ | |
967b05f6 | 1284 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1285 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1286 | skb->transport_header = (skb->network_header + |
1287 | fragheaderlen); | |
1da177e4 | 1288 | if (fraggap) { |
967b05f6 ACM |
1289 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1290 | maxfraglen, | |
9c70220b | 1291 | skb_transport_header(skb), |
967b05f6 | 1292 | fraggap, 0); |
1da177e4 LT |
1293 | skb_prev->csum = csum_sub(skb_prev->csum, |
1294 | skb->csum); | |
e9fa4f7b | 1295 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1296 | } |
1297 | ||
1298 | /* | |
1299 | * Put the packet on the pending queue. | |
1300 | */ | |
1301 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1302 | continue; | |
1303 | } | |
1304 | ||
1da177e4 LT |
1305 | if (len > size) |
1306 | len = size; | |
be12a1fe HFS |
1307 | |
1308 | if (skb_append_pagefrags(skb, page, offset, len)) { | |
1da177e4 LT |
1309 | err = -EMSGSIZE; |
1310 | goto error; | |
1311 | } | |
1312 | ||
1313 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1314 | __wsum csum; |
1da177e4 LT |
1315 | csum = csum_page(page, offset, len); |
1316 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1317 | } | |
1318 | ||
1319 | skb->len += len; | |
1320 | skb->data_len += len; | |
1e34a11d DM |
1321 | skb->truesize += len; |
1322 | atomic_add(len, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1323 | offset += len; |
1324 | size -= len; | |
1325 | } | |
1326 | return 0; | |
1327 | ||
1328 | error: | |
bdc712b4 | 1329 | cork->length -= size; |
5e38e270 | 1330 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1331 | return err; |
1332 | } | |
1333 | ||
1470ddf7 | 1334 | static void ip_cork_release(struct inet_cork *cork) |
429f08e9 | 1335 | { |
1470ddf7 HX |
1336 | cork->flags &= ~IPCORK_OPT; |
1337 | kfree(cork->opt); | |
1338 | cork->opt = NULL; | |
1339 | dst_release(cork->dst); | |
1340 | cork->dst = NULL; | |
429f08e9 PE |
1341 | } |
1342 | ||
1da177e4 LT |
1343 | /* |
1344 | * Combined all pending IP fragments on the socket as one IP datagram | |
1345 | * and push them out. | |
1346 | */ | |
1c32c5ad | 1347 | struct sk_buff *__ip_make_skb(struct sock *sk, |
77968b78 | 1348 | struct flowi4 *fl4, |
1c32c5ad HX |
1349 | struct sk_buff_head *queue, |
1350 | struct inet_cork *cork) | |
1da177e4 LT |
1351 | { |
1352 | struct sk_buff *skb, *tmp_skb; | |
1353 | struct sk_buff **tail_skb; | |
1354 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1355 | struct net *net = sock_net(sk); |
1da177e4 | 1356 | struct ip_options *opt = NULL; |
1470ddf7 | 1357 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 1358 | struct iphdr *iph; |
76ab608d | 1359 | __be16 df = 0; |
1da177e4 | 1360 | __u8 ttl; |
1da177e4 | 1361 | |
51456b29 IM |
1362 | skb = __skb_dequeue(queue); |
1363 | if (!skb) | |
1da177e4 LT |
1364 | goto out; |
1365 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1366 | ||
1367 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1368 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1369 | __skb_pull(skb, skb_network_offset(skb)); |
1470ddf7 | 1370 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { |
cfe1fc77 | 1371 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1372 | *tail_skb = tmp_skb; |
1373 | tail_skb = &(tmp_skb->next); | |
1374 | skb->len += tmp_skb->len; | |
1375 | skb->data_len += tmp_skb->len; | |
1376 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1377 | tmp_skb->destructor = NULL; |
1378 | tmp_skb->sk = NULL; | |
1379 | } | |
1380 | ||
1381 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1382 | * to fragment the frame generated here. No matter, what transforms | |
1383 | * how transforms change size of the packet, it will come out. | |
1384 | */ | |
60ff7467 | 1385 | skb->ignore_df = ip_sk_ignore_df(sk); |
1da177e4 LT |
1386 | |
1387 | /* DF bit is set when we want to see DF on outgoing frames. | |
60ff7467 | 1388 | * If ignore_df is set too, we still allow to fragment this frame |
1da177e4 | 1389 | * locally. */ |
482fc609 HFS |
1390 | if (inet->pmtudisc == IP_PMTUDISC_DO || |
1391 | inet->pmtudisc == IP_PMTUDISC_PROBE || | |
d8d1f30b CG |
1392 | (skb->len <= dst_mtu(&rt->dst) && |
1393 | ip_dont_fragment(sk, &rt->dst))) | |
1da177e4 LT |
1394 | df = htons(IP_DF); |
1395 | ||
1470ddf7 HX |
1396 | if (cork->flags & IPCORK_OPT) |
1397 | opt = cork->opt; | |
1da177e4 | 1398 | |
aa661581 FF |
1399 | if (cork->ttl != 0) |
1400 | ttl = cork->ttl; | |
1401 | else if (rt->rt_type == RTN_MULTICAST) | |
1da177e4 LT |
1402 | ttl = inet->mc_ttl; |
1403 | else | |
d8d1f30b | 1404 | ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 1405 | |
749154aa | 1406 | iph = ip_hdr(skb); |
1da177e4 LT |
1407 | iph->version = 4; |
1408 | iph->ihl = 5; | |
aa661581 | 1409 | iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; |
1da177e4 | 1410 | iph->frag_off = df; |
1da177e4 LT |
1411 | iph->ttl = ttl; |
1412 | iph->protocol = sk->sk_protocol; | |
84f9307c | 1413 | ip_copy_addrs(iph, fl4); |
b6a7719a | 1414 | ip_select_ident(net, skb, sk); |
1da177e4 | 1415 | |
22f728f8 DM |
1416 | if (opt) { |
1417 | iph->ihl += opt->optlen>>2; | |
1418 | ip_options_build(skb, opt, cork->addr, rt, 0); | |
1419 | } | |
1420 | ||
aa661581 | 1421 | skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; |
4a19ec58 | 1422 | skb->mark = sk->sk_mark; |
a21bba94 ED |
1423 | /* |
1424 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1425 | * on dst refcount | |
1426 | */ | |
1470ddf7 | 1427 | cork->dst = NULL; |
d8d1f30b | 1428 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 1429 | |
96793b48 | 1430 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1431 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1432 | skb_transport_header(skb))->type); |
1433 | ||
1c32c5ad HX |
1434 | ip_cork_release(cork); |
1435 | out: | |
1436 | return skb; | |
1437 | } | |
1438 | ||
b5ec8eea | 1439 | int ip_send_skb(struct net *net, struct sk_buff *skb) |
1c32c5ad | 1440 | { |
1c32c5ad HX |
1441 | int err; |
1442 | ||
c439cb2e | 1443 | err = ip_local_out(skb); |
1da177e4 LT |
1444 | if (err) { |
1445 | if (err > 0) | |
6ce9e7b5 | 1446 | err = net_xmit_errno(err); |
1da177e4 | 1447 | if (err) |
1c32c5ad | 1448 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1449 | } |
1450 | ||
1da177e4 | 1451 | return err; |
1da177e4 LT |
1452 | } |
1453 | ||
77968b78 | 1454 | int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) |
1470ddf7 | 1455 | { |
1c32c5ad HX |
1456 | struct sk_buff *skb; |
1457 | ||
77968b78 | 1458 | skb = ip_finish_skb(sk, fl4); |
1c32c5ad HX |
1459 | if (!skb) |
1460 | return 0; | |
1461 | ||
1462 | /* Netfilter gets whole the not fragmented skb. */ | |
b5ec8eea | 1463 | return ip_send_skb(sock_net(sk), skb); |
1470ddf7 HX |
1464 | } |
1465 | ||
1da177e4 LT |
1466 | /* |
1467 | * Throw away all pending data on the socket. | |
1468 | */ | |
1470ddf7 HX |
1469 | static void __ip_flush_pending_frames(struct sock *sk, |
1470 | struct sk_buff_head *queue, | |
1471 | struct inet_cork *cork) | |
1da177e4 | 1472 | { |
1da177e4 LT |
1473 | struct sk_buff *skb; |
1474 | ||
1470ddf7 | 1475 | while ((skb = __skb_dequeue_tail(queue)) != NULL) |
1da177e4 LT |
1476 | kfree_skb(skb); |
1477 | ||
1470ddf7 HX |
1478 | ip_cork_release(cork); |
1479 | } | |
1480 | ||
1481 | void ip_flush_pending_frames(struct sock *sk) | |
1482 | { | |
bdc712b4 | 1483 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); |
1da177e4 LT |
1484 | } |
1485 | ||
1c32c5ad | 1486 | struct sk_buff *ip_make_skb(struct sock *sk, |
77968b78 | 1487 | struct flowi4 *fl4, |
1c32c5ad HX |
1488 | int getfrag(void *from, char *to, int offset, |
1489 | int len, int odd, struct sk_buff *skb), | |
1490 | void *from, int length, int transhdrlen, | |
1491 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1492 | unsigned int flags) | |
1493 | { | |
b80d7226 | 1494 | struct inet_cork cork; |
1c32c5ad HX |
1495 | struct sk_buff_head queue; |
1496 | int err; | |
1497 | ||
1498 | if (flags & MSG_PROBE) | |
1499 | return NULL; | |
1500 | ||
1501 | __skb_queue_head_init(&queue); | |
1502 | ||
b80d7226 DM |
1503 | cork.flags = 0; |
1504 | cork.addr = 0; | |
70652728 | 1505 | cork.opt = NULL; |
1c32c5ad HX |
1506 | err = ip_setup_cork(sk, &cork, ipc, rtp); |
1507 | if (err) | |
1508 | return ERR_PTR(err); | |
1509 | ||
5640f768 ED |
1510 | err = __ip_append_data(sk, fl4, &queue, &cork, |
1511 | ¤t->task_frag, getfrag, | |
1c32c5ad HX |
1512 | from, length, transhdrlen, flags); |
1513 | if (err) { | |
1514 | __ip_flush_pending_frames(sk, &queue, &cork); | |
1515 | return ERR_PTR(err); | |
1516 | } | |
1517 | ||
77968b78 | 1518 | return __ip_make_skb(sk, fl4, &queue, &cork); |
1c32c5ad | 1519 | } |
1da177e4 LT |
1520 | |
1521 | /* | |
1522 | * Fetch data from kernel space and fill in checksum if needed. | |
1523 | */ | |
e905a9ed | 1524 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1525 | int len, int odd, struct sk_buff *skb) |
1526 | { | |
5084205f | 1527 | __wsum csum; |
1da177e4 LT |
1528 | |
1529 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1530 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1531 | return 0; |
1da177e4 LT |
1532 | } |
1533 | ||
e905a9ed | 1534 | /* |
1da177e4 | 1535 | * Generic function to send a packet as reply to another packet. |
be9f4a44 | 1536 | * Used to send some TCP resets/acks so far. |
1da177e4 | 1537 | */ |
bdbbb852 | 1538 | void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, |
24a2d43d ED |
1539 | const struct ip_options *sopt, |
1540 | __be32 daddr, __be32 saddr, | |
1541 | const struct ip_reply_arg *arg, | |
70e73416 | 1542 | unsigned int len) |
1da177e4 | 1543 | { |
f6d8bd05 | 1544 | struct ip_options_data replyopts; |
1da177e4 | 1545 | struct ipcm_cookie ipc; |
77968b78 | 1546 | struct flowi4 fl4; |
511c3f92 | 1547 | struct rtable *rt = skb_rtable(skb); |
bdbbb852 | 1548 | struct net *net = sock_net(sk); |
be9f4a44 | 1549 | struct sk_buff *nskb; |
4062090e | 1550 | int err; |
f7ba868b | 1551 | int oif; |
1da177e4 | 1552 | |
24a2d43d | 1553 | if (__ip_options_echo(&replyopts.opt.opt, skb, sopt)) |
1da177e4 LT |
1554 | return; |
1555 | ||
0a5ebb80 | 1556 | ipc.addr = daddr; |
1da177e4 | 1557 | ipc.opt = NULL; |
2244d07b | 1558 | ipc.tx_flags = 0; |
aa661581 FF |
1559 | ipc.ttl = 0; |
1560 | ipc.tos = -1; | |
1da177e4 | 1561 | |
f6d8bd05 | 1562 | if (replyopts.opt.opt.optlen) { |
1da177e4 LT |
1563 | ipc.opt = &replyopts.opt; |
1564 | ||
f6d8bd05 ED |
1565 | if (replyopts.opt.opt.srr) |
1566 | daddr = replyopts.opt.opt.faddr; | |
1da177e4 LT |
1567 | } |
1568 | ||
f7ba868b | 1569 | oif = arg->bound_dev_if; |
007979ea | 1570 | if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) |
f7ba868b DA |
1571 | oif = skb->skb_iif; |
1572 | ||
1573 | flowi4_init_output(&fl4, oif, | |
e110861f | 1574 | IP4_REPLY_MARK(net, skb->mark), |
66b13d99 | 1575 | RT_TOS(arg->tos), |
be9f4a44 | 1576 | RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, |
77968b78 | 1577 | ip_reply_arg_flowi_flags(arg), |
70e73416 | 1578 | daddr, saddr, |
77968b78 DM |
1579 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest); |
1580 | security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); | |
be9f4a44 | 1581 | rt = ip_route_output_key(net, &fl4); |
77968b78 DM |
1582 | if (IS_ERR(rt)) |
1583 | return; | |
1da177e4 | 1584 | |
bdbbb852 | 1585 | inet_sk(sk)->tos = arg->tos; |
1da177e4 | 1586 | |
1da177e4 | 1587 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1588 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1589 | sk->sk_bound_dev_if = arg->bound_dev_if; |
be9f4a44 | 1590 | sk->sk_sndbuf = sysctl_wmem_default; |
4062090e VA |
1591 | err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, |
1592 | len, 0, &ipc, &rt, MSG_DONTWAIT); | |
1593 | if (unlikely(err)) { | |
1594 | ip_flush_pending_frames(sk); | |
1595 | goto out; | |
1596 | } | |
1597 | ||
be9f4a44 ED |
1598 | nskb = skb_peek(&sk->sk_write_queue); |
1599 | if (nskb) { | |
1da177e4 | 1600 | if (arg->csumoffset >= 0) |
be9f4a44 ED |
1601 | *((__sum16 *)skb_transport_header(nskb) + |
1602 | arg->csumoffset) = csum_fold(csum_add(nskb->csum, | |
9c70220b | 1603 | arg->csum)); |
be9f4a44 ED |
1604 | nskb->ip_summed = CHECKSUM_NONE; |
1605 | skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb)); | |
77968b78 | 1606 | ip_push_pending_frames(sk, &fl4); |
1da177e4 | 1607 | } |
4062090e | 1608 | out: |
1da177e4 LT |
1609 | ip_rt_put(rt); |
1610 | } | |
1611 | ||
1da177e4 LT |
1612 | void __init ip_init(void) |
1613 | { | |
1da177e4 LT |
1614 | ip_rt_init(); |
1615 | inet_initpeers(); | |
1616 | ||
72c1d3bd WC |
1617 | #if defined(CONFIG_IP_MULTICAST) |
1618 | igmp_mc_init(); | |
1da177e4 LT |
1619 | #endif |
1620 | } |