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> | |
46 | #include <asm/system.h> | |
47 | #include <linux/module.h> | |
48 | #include <linux/types.h> | |
49 | #include <linux/kernel.h> | |
1da177e4 LT |
50 | #include <linux/mm.h> |
51 | #include <linux/string.h> | |
52 | #include <linux/errno.h> | |
a1f8e7f7 | 53 | #include <linux/highmem.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
1da177e4 LT |
55 | |
56 | #include <linux/socket.h> | |
57 | #include <linux/sockios.h> | |
58 | #include <linux/in.h> | |
59 | #include <linux/inet.h> | |
60 | #include <linux/netdevice.h> | |
61 | #include <linux/etherdevice.h> | |
62 | #include <linux/proc_fs.h> | |
63 | #include <linux/stat.h> | |
64 | #include <linux/init.h> | |
65 | ||
66 | #include <net/snmp.h> | |
67 | #include <net/ip.h> | |
68 | #include <net/protocol.h> | |
69 | #include <net/route.h> | |
cfacb057 | 70 | #include <net/xfrm.h> |
1da177e4 LT |
71 | #include <linux/skbuff.h> |
72 | #include <net/sock.h> | |
73 | #include <net/arp.h> | |
74 | #include <net/icmp.h> | |
1da177e4 LT |
75 | #include <net/checksum.h> |
76 | #include <net/inetpeer.h> | |
1da177e4 LT |
77 | #include <linux/igmp.h> |
78 | #include <linux/netfilter_ipv4.h> | |
79 | #include <linux/netfilter_bridge.h> | |
80 | #include <linux/mroute.h> | |
81 | #include <linux/netlink.h> | |
6cbb0df7 | 82 | #include <linux/tcp.h> |
1da177e4 | 83 | |
ab32ea5d | 84 | int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; |
323e126f | 85 | EXPORT_SYMBOL(sysctl_ip_default_ttl); |
1da177e4 LT |
86 | |
87 | /* Generate a checksum for an outgoing IP datagram. */ | |
88 | __inline__ void ip_send_check(struct iphdr *iph) | |
89 | { | |
90 | iph->check = 0; | |
91 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
92 | } | |
4bc2f18b | 93 | EXPORT_SYMBOL(ip_send_check); |
1da177e4 | 94 | |
c439cb2e HX |
95 | int __ip_local_out(struct sk_buff *skb) |
96 | { | |
97 | struct iphdr *iph = ip_hdr(skb); | |
98 | ||
99 | iph->tot_len = htons(skb->len); | |
100 | ip_send_check(iph); | |
9bbc768a JE |
101 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL, |
102 | skb_dst(skb)->dev, dst_output); | |
c439cb2e HX |
103 | } |
104 | ||
105 | int ip_local_out(struct sk_buff *skb) | |
106 | { | |
107 | int err; | |
108 | ||
109 | err = __ip_local_out(skb); | |
110 | if (likely(err == 1)) | |
111 | err = dst_output(skb); | |
112 | ||
113 | return err; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(ip_local_out); | |
116 | ||
1da177e4 LT |
117 | /* dev_loopback_xmit for use with netfilter. */ |
118 | static int ip_dev_loopback_xmit(struct sk_buff *newskb) | |
119 | { | |
459a98ed | 120 | skb_reset_mac_header(newskb); |
bbe735e4 | 121 | __skb_pull(newskb, skb_network_offset(newskb)); |
1da177e4 LT |
122 | newskb->pkt_type = PACKET_LOOPBACK; |
123 | newskb->ip_summed = CHECKSUM_UNNECESSARY; | |
adf30907 | 124 | WARN_ON(!skb_dst(newskb)); |
e30b38c2 | 125 | netif_rx_ni(newskb); |
1da177e4 LT |
126 | return 0; |
127 | } | |
128 | ||
129 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) | |
130 | { | |
131 | int ttl = inet->uc_ttl; | |
132 | ||
133 | if (ttl < 0) | |
323e126f | 134 | ttl = ip4_dst_hoplimit(dst); |
1da177e4 LT |
135 | return ttl; |
136 | } | |
137 | ||
e905a9ed | 138 | /* |
1da177e4 LT |
139 | * Add an ip header to a skbuff and send it out. |
140 | * | |
141 | */ | |
142 | int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, | |
f6d8bd05 | 143 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) |
1da177e4 LT |
144 | { |
145 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 146 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
147 | struct iphdr *iph; |
148 | ||
149 | /* Build the IP header. */ | |
f6d8bd05 | 150 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); |
8856dfa3 | 151 | skb_reset_network_header(skb); |
eddc9ec5 | 152 | iph = ip_hdr(skb); |
1da177e4 LT |
153 | iph->version = 4; |
154 | iph->ihl = 5; | |
155 | iph->tos = inet->tos; | |
d8d1f30b | 156 | if (ip_dont_fragment(sk, &rt->dst)) |
1da177e4 LT |
157 | iph->frag_off = htons(IP_DF); |
158 | else | |
159 | iph->frag_off = 0; | |
d8d1f30b | 160 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
dd927a26 DM |
161 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); |
162 | iph->saddr = saddr; | |
1da177e4 | 163 | iph->protocol = sk->sk_protocol; |
d8d1f30b | 164 | ip_select_ident(iph, &rt->dst, sk); |
1da177e4 | 165 | |
f6d8bd05 ED |
166 | if (opt && opt->opt.optlen) { |
167 | iph->ihl += opt->opt.optlen>>2; | |
168 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
1da177e4 | 169 | } |
1da177e4 LT |
170 | |
171 | skb->priority = sk->sk_priority; | |
4a19ec58 | 172 | skb->mark = sk->sk_mark; |
1da177e4 LT |
173 | |
174 | /* Send it out. */ | |
c439cb2e | 175 | return ip_local_out(skb); |
1da177e4 | 176 | } |
d8c97a94 ACM |
177 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
178 | ||
1da177e4 LT |
179 | static inline int ip_finish_output2(struct sk_buff *skb) |
180 | { | |
adf30907 | 181 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 182 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 183 | struct net_device *dev = dst->dev; |
c2636b4d | 184 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
1da177e4 | 185 | |
edf391ff NH |
186 | if (rt->rt_type == RTN_MULTICAST) { |
187 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); | |
188 | } else if (rt->rt_type == RTN_BROADCAST) | |
189 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); | |
80787ebc | 190 | |
1da177e4 | 191 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 192 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
193 | struct sk_buff *skb2; |
194 | ||
195 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
196 | if (skb2 == NULL) { | |
197 | kfree_skb(skb); | |
198 | return -ENOMEM; | |
199 | } | |
200 | if (skb->sk) | |
201 | skb_set_owner_w(skb2, skb->sk); | |
202 | kfree_skb(skb); | |
203 | skb = skb2; | |
204 | } | |
205 | ||
3644f0ce SH |
206 | if (dst->hh) |
207 | return neigh_hh_output(dst->hh, skb); | |
208 | else if (dst->neighbour) | |
1da177e4 LT |
209 | return dst->neighbour->output(skb); |
210 | ||
211 | if (net_ratelimit()) | |
212 | printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n"); | |
213 | kfree_skb(skb); | |
214 | return -EINVAL; | |
215 | } | |
216 | ||
628a5c56 JH |
217 | static inline int ip_skb_dst_mtu(struct sk_buff *skb) |
218 | { | |
219 | struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL; | |
220 | ||
221 | return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ? | |
adf30907 | 222 | skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb)); |
628a5c56 JH |
223 | } |
224 | ||
861d0486 | 225 | static int ip_finish_output(struct sk_buff *skb) |
1da177e4 | 226 | { |
5c901daa PM |
227 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
228 | /* Policy lookup after SNAT yielded a new policy */ | |
adf30907 | 229 | if (skb_dst(skb)->xfrm != NULL) { |
48d5cad8 PM |
230 | IPCB(skb)->flags |= IPSKB_REROUTED; |
231 | return dst_output(skb); | |
232 | } | |
5c901daa | 233 | #endif |
628a5c56 | 234 | if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb)) |
1bd9bef6 PM |
235 | return ip_fragment(skb, ip_finish_output2); |
236 | else | |
237 | return ip_finish_output2(skb); | |
1da177e4 LT |
238 | } |
239 | ||
240 | int ip_mc_output(struct sk_buff *skb) | |
241 | { | |
242 | struct sock *sk = skb->sk; | |
511c3f92 | 243 | struct rtable *rt = skb_rtable(skb); |
d8d1f30b | 244 | struct net_device *dev = rt->dst.dev; |
1da177e4 LT |
245 | |
246 | /* | |
247 | * If the indicated interface is up and running, send the packet. | |
248 | */ | |
edf391ff | 249 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
250 | |
251 | skb->dev = dev; | |
252 | skb->protocol = htons(ETH_P_IP); | |
253 | ||
254 | /* | |
255 | * Multicasts are looped back for other local users | |
256 | */ | |
257 | ||
258 | if (rt->rt_flags&RTCF_MULTICAST) { | |
7ad6848c | 259 | if (sk_mc_loop(sk) |
1da177e4 LT |
260 | #ifdef CONFIG_IP_MROUTE |
261 | /* Small optimization: do not loopback not local frames, | |
262 | which returned after forwarding; they will be dropped | |
263 | by ip_mr_input in any case. | |
264 | Note, that local frames are looped back to be delivered | |
265 | to local recipients. | |
266 | ||
267 | This check is duplicated in ip_mr_input at the moment. | |
268 | */ | |
9d4fb27d JP |
269 | && |
270 | ((rt->rt_flags & RTCF_LOCAL) || | |
271 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 272 | #endif |
9d4fb27d | 273 | ) { |
1da177e4 LT |
274 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
275 | if (newskb) | |
9bbc768a JE |
276 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
277 | newskb, NULL, newskb->dev, | |
1da177e4 LT |
278 | ip_dev_loopback_xmit); |
279 | } | |
280 | ||
281 | /* Multicasts with ttl 0 must not go beyond the host */ | |
282 | ||
eddc9ec5 | 283 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
284 | kfree_skb(skb); |
285 | return 0; | |
286 | } | |
287 | } | |
288 | ||
289 | if (rt->rt_flags&RTCF_BROADCAST) { | |
290 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
291 | if (newskb) | |
9bbc768a JE |
292 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, |
293 | NULL, newskb->dev, ip_dev_loopback_xmit); | |
1da177e4 LT |
294 | } |
295 | ||
9bbc768a JE |
296 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, |
297 | skb->dev, ip_finish_output, | |
48d5cad8 | 298 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
299 | } |
300 | ||
301 | int ip_output(struct sk_buff *skb) | |
302 | { | |
adf30907 | 303 | struct net_device *dev = skb_dst(skb)->dev; |
1bd9bef6 | 304 | |
edf391ff | 305 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 306 | |
1bd9bef6 PM |
307 | skb->dev = dev; |
308 | skb->protocol = htons(ETH_P_IP); | |
309 | ||
9bbc768a | 310 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, |
e905a9ed | 311 | ip_finish_output, |
48d5cad8 | 312 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
313 | } |
314 | ||
4e15ed4d | 315 | int ip_queue_xmit(struct sk_buff *skb) |
1da177e4 | 316 | { |
e89862f4 | 317 | struct sock *sk = skb->sk; |
1da177e4 | 318 | struct inet_sock *inet = inet_sk(sk); |
f6d8bd05 | 319 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
320 | struct rtable *rt; |
321 | struct iphdr *iph; | |
ab6e3feb | 322 | int res; |
1da177e4 LT |
323 | |
324 | /* Skip all of this if the packet is already routed, | |
325 | * f.e. by something like SCTP. | |
326 | */ | |
ab6e3feb | 327 | rcu_read_lock(); |
f6d8bd05 | 328 | inet_opt = rcu_dereference(inet->inet_opt); |
511c3f92 | 329 | rt = skb_rtable(skb); |
1da177e4 LT |
330 | if (rt != NULL) |
331 | goto packet_routed; | |
332 | ||
333 | /* Make sure we can route this packet. */ | |
334 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
335 | if (rt == NULL) { | |
31e4543d | 336 | struct flowi4 fl4; |
3ca3c68e | 337 | __be32 daddr; |
1da177e4 LT |
338 | |
339 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 340 | daddr = inet->inet_daddr; |
f6d8bd05 ED |
341 | if (inet_opt && inet_opt->opt.srr) |
342 | daddr = inet_opt->opt.faddr; | |
1da177e4 | 343 | |
78fbfd8a DM |
344 | /* If this fails, retransmit mechanism of transport layer will |
345 | * keep trying until route appears or the connection times | |
346 | * itself out. | |
347 | */ | |
31e4543d | 348 | rt = ip_route_output_ports(sock_net(sk), &fl4, sk, |
78fbfd8a DM |
349 | daddr, inet->inet_saddr, |
350 | inet->inet_dport, | |
351 | inet->inet_sport, | |
352 | sk->sk_protocol, | |
353 | RT_CONN_FLAGS(sk), | |
354 | sk->sk_bound_dev_if); | |
355 | if (IS_ERR(rt)) | |
356 | goto no_route; | |
d8d1f30b | 357 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 358 | } |
d8d1f30b | 359 | skb_dst_set_noref(skb, &rt->dst); |
1da177e4 LT |
360 | |
361 | packet_routed: | |
f6d8bd05 | 362 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_dst != rt->rt_gateway) |
1da177e4 LT |
363 | goto no_route; |
364 | ||
365 | /* OK, we know where to send it, allocate and build IP header. */ | |
f6d8bd05 | 366 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); |
8856dfa3 | 367 | skb_reset_network_header(skb); |
eddc9ec5 | 368 | iph = ip_hdr(skb); |
714e85be | 369 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
d8d1f30b | 370 | if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df) |
1da177e4 LT |
371 | iph->frag_off = htons(IP_DF); |
372 | else | |
373 | iph->frag_off = 0; | |
d8d1f30b | 374 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 LT |
375 | iph->protocol = sk->sk_protocol; |
376 | iph->saddr = rt->rt_src; | |
377 | iph->daddr = rt->rt_dst; | |
1da177e4 LT |
378 | /* Transport layer set skb->h.foo itself. */ |
379 | ||
f6d8bd05 ED |
380 | if (inet_opt && inet_opt->opt.optlen) { |
381 | iph->ihl += inet_opt->opt.optlen >> 2; | |
382 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
1da177e4 LT |
383 | } |
384 | ||
d8d1f30b | 385 | ip_select_ident_more(iph, &rt->dst, sk, |
7967168c | 386 | (skb_shinfo(skb)->gso_segs ?: 1) - 1); |
1da177e4 | 387 | |
1da177e4 | 388 | skb->priority = sk->sk_priority; |
4a19ec58 | 389 | skb->mark = sk->sk_mark; |
1da177e4 | 390 | |
ab6e3feb ED |
391 | res = ip_local_out(skb); |
392 | rcu_read_unlock(); | |
393 | return res; | |
1da177e4 LT |
394 | |
395 | no_route: | |
ab6e3feb | 396 | rcu_read_unlock(); |
5e38e270 | 397 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
398 | kfree_skb(skb); |
399 | return -EHOSTUNREACH; | |
400 | } | |
4bc2f18b | 401 | EXPORT_SYMBOL(ip_queue_xmit); |
1da177e4 LT |
402 | |
403 | ||
404 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
405 | { | |
406 | to->pkt_type = from->pkt_type; | |
407 | to->priority = from->priority; | |
408 | to->protocol = from->protocol; | |
adf30907 | 409 | skb_dst_drop(to); |
fe76cda3 | 410 | skb_dst_copy(to, from); |
1da177e4 | 411 | to->dev = from->dev; |
82e91ffe | 412 | to->mark = from->mark; |
1da177e4 LT |
413 | |
414 | /* Copy the flags to each fragment. */ | |
415 | IPCB(to)->flags = IPCB(from)->flags; | |
416 | ||
417 | #ifdef CONFIG_NET_SCHED | |
418 | to->tc_index = from->tc_index; | |
419 | #endif | |
e7ac05f3 | 420 | nf_copy(to, from); |
ba9dda3a JK |
421 | #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ |
422 | defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) | |
423 | to->nf_trace = from->nf_trace; | |
424 | #endif | |
c98d80ed JA |
425 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
426 | to->ipvs_property = from->ipvs_property; | |
1da177e4 | 427 | #endif |
984bc16c | 428 | skb_copy_secmark(to, from); |
1da177e4 LT |
429 | } |
430 | ||
431 | /* | |
432 | * This IP datagram is too large to be sent in one piece. Break it up into | |
433 | * smaller pieces (each of size equal to IP header plus | |
434 | * a block of the data of the original IP data part) that will yet fit in a | |
435 | * single device frame, and queue such a frame for sending. | |
436 | */ | |
437 | ||
d9319100 | 438 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) |
1da177e4 LT |
439 | { |
440 | struct iphdr *iph; | |
1da177e4 LT |
441 | int ptr; |
442 | struct net_device *dev; | |
443 | struct sk_buff *skb2; | |
c893b806 | 444 | unsigned int mtu, hlen, left, len, ll_rs; |
1da177e4 | 445 | int offset; |
76ab608d | 446 | __be16 not_last_frag; |
511c3f92 | 447 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
448 | int err = 0; |
449 | ||
d8d1f30b | 450 | dev = rt->dst.dev; |
1da177e4 LT |
451 | |
452 | /* | |
453 | * Point into the IP datagram header. | |
454 | */ | |
455 | ||
eddc9ec5 | 456 | iph = ip_hdr(skb); |
1da177e4 LT |
457 | |
458 | if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { | |
5e38e270 | 459 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 460 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
628a5c56 | 461 | htonl(ip_skb_dst_mtu(skb))); |
1da177e4 LT |
462 | kfree_skb(skb); |
463 | return -EMSGSIZE; | |
464 | } | |
465 | ||
466 | /* | |
467 | * Setup starting values. | |
468 | */ | |
469 | ||
470 | hlen = iph->ihl * 4; | |
d8d1f30b | 471 | mtu = dst_mtu(&rt->dst) - hlen; /* Size of data space */ |
6c79bf0f BDS |
472 | #ifdef CONFIG_BRIDGE_NETFILTER |
473 | if (skb->nf_bridge) | |
474 | mtu -= nf_bridge_mtu_reduction(skb); | |
475 | #endif | |
89cee8b1 | 476 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
477 | |
478 | /* When frag_list is given, use it. First, check its validity: | |
479 | * some transformers could create wrong frag_list or break existing | |
480 | * one, it is not prohibited. In this case fall back to copying. | |
481 | * | |
482 | * LATER: this step can be merged to real generation of fragments, | |
483 | * we can switch to copy when see the first bad fragment. | |
484 | */ | |
21dc3301 | 485 | if (skb_has_frag_list(skb)) { |
3d13008e | 486 | struct sk_buff *frag, *frag2; |
1da177e4 LT |
487 | int first_len = skb_pagelen(skb); |
488 | ||
489 | if (first_len - hlen > mtu || | |
490 | ((first_len - hlen) & 7) || | |
491 | (iph->frag_off & htons(IP_MF|IP_OFFSET)) || | |
492 | skb_cloned(skb)) | |
493 | goto slow_path; | |
494 | ||
d7fcf1a5 | 495 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
496 | /* Correct geometry. */ |
497 | if (frag->len > mtu || | |
498 | ((frag->len & 7) && frag->next) || | |
499 | skb_headroom(frag) < hlen) | |
3d13008e | 500 | goto slow_path_clean; |
1da177e4 LT |
501 | |
502 | /* Partially cloned skb? */ | |
503 | if (skb_shared(frag)) | |
3d13008e | 504 | goto slow_path_clean; |
2fdba6b0 HX |
505 | |
506 | BUG_ON(frag->sk); | |
507 | if (skb->sk) { | |
2fdba6b0 HX |
508 | frag->sk = skb->sk; |
509 | frag->destructor = sock_wfree; | |
2fdba6b0 | 510 | } |
3d13008e | 511 | skb->truesize -= frag->truesize; |
1da177e4 LT |
512 | } |
513 | ||
514 | /* Everything is OK. Generate! */ | |
515 | ||
516 | err = 0; | |
517 | offset = 0; | |
518 | frag = skb_shinfo(skb)->frag_list; | |
d7fcf1a5 | 519 | skb_frag_list_init(skb); |
1da177e4 LT |
520 | skb->data_len = first_len - skb_headlen(skb); |
521 | skb->len = first_len; | |
522 | iph->tot_len = htons(first_len); | |
523 | iph->frag_off = htons(IP_MF); | |
524 | ip_send_check(iph); | |
525 | ||
526 | for (;;) { | |
527 | /* Prepare header of the next frame, | |
528 | * before previous one went down. */ | |
529 | if (frag) { | |
530 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 531 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
532 | __skb_push(frag, hlen); |
533 | skb_reset_network_header(frag); | |
d56f90a7 | 534 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 535 | iph = ip_hdr(frag); |
1da177e4 LT |
536 | iph->tot_len = htons(frag->len); |
537 | ip_copy_metadata(frag, skb); | |
538 | if (offset == 0) | |
539 | ip_options_fragment(frag); | |
540 | offset += skb->len - hlen; | |
541 | iph->frag_off = htons(offset>>3); | |
542 | if (frag->next != NULL) | |
543 | iph->frag_off |= htons(IP_MF); | |
544 | /* Ready, complete checksum */ | |
545 | ip_send_check(iph); | |
546 | } | |
547 | ||
548 | err = output(skb); | |
549 | ||
dafee490 | 550 | if (!err) |
5e38e270 | 551 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
552 | if (err || !frag) |
553 | break; | |
554 | ||
555 | skb = frag; | |
556 | frag = skb->next; | |
557 | skb->next = NULL; | |
558 | } | |
559 | ||
560 | if (err == 0) { | |
5e38e270 | 561 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
562 | return 0; |
563 | } | |
564 | ||
565 | while (frag) { | |
566 | skb = frag->next; | |
567 | kfree_skb(frag); | |
568 | frag = skb; | |
569 | } | |
5e38e270 | 570 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 571 | return err; |
3d13008e ED |
572 | |
573 | slow_path_clean: | |
574 | skb_walk_frags(skb, frag2) { | |
575 | if (frag2 == frag) | |
576 | break; | |
577 | frag2->sk = NULL; | |
578 | frag2->destructor = NULL; | |
579 | skb->truesize += frag2->truesize; | |
580 | } | |
1da177e4 LT |
581 | } |
582 | ||
583 | slow_path: | |
584 | left = skb->len - hlen; /* Space per frame */ | |
49085bd7 | 585 | ptr = hlen; /* Where to start from */ |
1da177e4 | 586 | |
1da177e4 | 587 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, |
9bcfcaf5 SH |
588 | * we need to make room for the encapsulating header |
589 | */ | |
c893b806 | 590 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); |
9bcfcaf5 | 591 | |
1da177e4 LT |
592 | /* |
593 | * Fragment the datagram. | |
594 | */ | |
595 | ||
596 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
597 | not_last_frag = iph->frag_off & htons(IP_MF); | |
598 | ||
599 | /* | |
600 | * Keep copying data until we run out. | |
601 | */ | |
602 | ||
132adf54 | 603 | while (left > 0) { |
1da177e4 LT |
604 | len = left; |
605 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
606 | if (len > mtu) | |
607 | len = mtu; | |
25985edc | 608 | /* IF: we are not sending up to and including the packet end |
1da177e4 LT |
609 | then align the next start on an eight byte boundary */ |
610 | if (len < left) { | |
611 | len &= ~7; | |
612 | } | |
613 | /* | |
614 | * Allocate buffer. | |
615 | */ | |
616 | ||
617 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 618 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
619 | err = -ENOMEM; |
620 | goto fail; | |
621 | } | |
622 | ||
623 | /* | |
624 | * Set up data on packet | |
625 | */ | |
626 | ||
627 | ip_copy_metadata(skb2, skb); | |
628 | skb_reserve(skb2, ll_rs); | |
629 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 630 | skb_reset_network_header(skb2); |
b0e380b1 | 631 | skb2->transport_header = skb2->network_header + hlen; |
1da177e4 LT |
632 | |
633 | /* | |
634 | * Charge the memory for the fragment to any owner | |
635 | * it might possess | |
636 | */ | |
637 | ||
638 | if (skb->sk) | |
639 | skb_set_owner_w(skb2, skb->sk); | |
640 | ||
641 | /* | |
642 | * Copy the packet header into the new buffer. | |
643 | */ | |
644 | ||
d626f62b | 645 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); |
1da177e4 LT |
646 | |
647 | /* | |
648 | * Copy a block of the IP datagram. | |
649 | */ | |
bff9b61c | 650 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) |
1da177e4 LT |
651 | BUG(); |
652 | left -= len; | |
653 | ||
654 | /* | |
655 | * Fill in the new header fields. | |
656 | */ | |
eddc9ec5 | 657 | iph = ip_hdr(skb2); |
1da177e4 LT |
658 | iph->frag_off = htons((offset >> 3)); |
659 | ||
660 | /* ANK: dirty, but effective trick. Upgrade options only if | |
661 | * the segment to be fragmented was THE FIRST (otherwise, | |
662 | * options are already fixed) and make it ONCE | |
663 | * on the initial skb, so that all the following fragments | |
664 | * will inherit fixed options. | |
665 | */ | |
666 | if (offset == 0) | |
667 | ip_options_fragment(skb); | |
668 | ||
669 | /* | |
670 | * Added AC : If we are fragmenting a fragment that's not the | |
671 | * last fragment then keep MF on each bit | |
672 | */ | |
673 | if (left > 0 || not_last_frag) | |
674 | iph->frag_off |= htons(IP_MF); | |
675 | ptr += len; | |
676 | offset += len; | |
677 | ||
678 | /* | |
679 | * Put this fragment into the sending queue. | |
680 | */ | |
1da177e4 LT |
681 | iph->tot_len = htons(len + hlen); |
682 | ||
683 | ip_send_check(iph); | |
684 | ||
685 | err = output(skb2); | |
686 | if (err) | |
687 | goto fail; | |
dafee490 | 688 | |
5e38e270 | 689 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
690 | } |
691 | kfree_skb(skb); | |
5e38e270 | 692 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
693 | return err; |
694 | ||
695 | fail: | |
e905a9ed | 696 | kfree_skb(skb); |
5e38e270 | 697 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
698 | return err; |
699 | } | |
2e2f7aef PM |
700 | EXPORT_SYMBOL(ip_fragment); |
701 | ||
1da177e4 LT |
702 | int |
703 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
704 | { | |
705 | struct iovec *iov = from; | |
706 | ||
84fa7933 | 707 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1da177e4 LT |
708 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) |
709 | return -EFAULT; | |
710 | } else { | |
44bb9363 | 711 | __wsum csum = 0; |
1da177e4 LT |
712 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) |
713 | return -EFAULT; | |
714 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
715 | } | |
716 | return 0; | |
717 | } | |
4bc2f18b | 718 | EXPORT_SYMBOL(ip_generic_getfrag); |
1da177e4 | 719 | |
44bb9363 | 720 | static inline __wsum |
1da177e4 LT |
721 | csum_page(struct page *page, int offset, int copy) |
722 | { | |
723 | char *kaddr; | |
44bb9363 | 724 | __wsum csum; |
1da177e4 LT |
725 | kaddr = kmap(page); |
726 | csum = csum_partial(kaddr + offset, copy, 0); | |
727 | kunmap(page); | |
728 | return csum; | |
729 | } | |
730 | ||
4b30b1c6 | 731 | static inline int ip_ufo_append_data(struct sock *sk, |
1470ddf7 | 732 | struct sk_buff_head *queue, |
e89e9cf5 AR |
733 | int getfrag(void *from, char *to, int offset, int len, |
734 | int odd, struct sk_buff *skb), | |
735 | void *from, int length, int hh_len, int fragheaderlen, | |
d9319100 | 736 | int transhdrlen, int mtu, unsigned int flags) |
e89e9cf5 AR |
737 | { |
738 | struct sk_buff *skb; | |
739 | int err; | |
740 | ||
741 | /* There is support for UDP fragmentation offload by network | |
742 | * device, so create one single skb packet containing complete | |
743 | * udp datagram | |
744 | */ | |
1470ddf7 | 745 | if ((skb = skb_peek_tail(queue)) == NULL) { |
e89e9cf5 AR |
746 | skb = sock_alloc_send_skb(sk, |
747 | hh_len + fragheaderlen + transhdrlen + 20, | |
748 | (flags & MSG_DONTWAIT), &err); | |
749 | ||
750 | if (skb == NULL) | |
751 | return err; | |
752 | ||
753 | /* reserve space for Hardware header */ | |
754 | skb_reserve(skb, hh_len); | |
755 | ||
756 | /* create space for UDP/IP header */ | |
d9319100 | 757 | skb_put(skb, fragheaderlen + transhdrlen); |
e89e9cf5 AR |
758 | |
759 | /* initialize network header pointer */ | |
c1d2bbe1 | 760 | skb_reset_network_header(skb); |
e89e9cf5 AR |
761 | |
762 | /* initialize protocol header pointer */ | |
b0e380b1 | 763 | skb->transport_header = skb->network_header + fragheaderlen; |
e89e9cf5 | 764 | |
84fa7933 | 765 | skb->ip_summed = CHECKSUM_PARTIAL; |
e89e9cf5 | 766 | skb->csum = 0; |
e89e9cf5 | 767 | |
be9164e7 | 768 | /* specify the length of each IP datagram fragment */ |
7967168c | 769 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 770 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
1470ddf7 | 771 | __skb_queue_tail(queue, skb); |
e89e9cf5 | 772 | } |
be9164e7 K |
773 | |
774 | return skb_append_datato_frags(sk, skb, getfrag, from, | |
775 | (length - transhdrlen)); | |
e89e9cf5 AR |
776 | } |
777 | ||
1470ddf7 HX |
778 | static int __ip_append_data(struct sock *sk, struct sk_buff_head *queue, |
779 | struct inet_cork *cork, | |
780 | int getfrag(void *from, char *to, int offset, | |
781 | int len, int odd, struct sk_buff *skb), | |
782 | void *from, int length, int transhdrlen, | |
783 | unsigned int flags) | |
1da177e4 LT |
784 | { |
785 | struct inet_sock *inet = inet_sk(sk); | |
786 | struct sk_buff *skb; | |
787 | ||
07df5294 | 788 | struct ip_options *opt = cork->opt; |
1da177e4 LT |
789 | int hh_len; |
790 | int exthdrlen; | |
791 | int mtu; | |
792 | int copy; | |
793 | int err; | |
794 | int offset = 0; | |
795 | unsigned int maxfraglen, fragheaderlen; | |
796 | int csummode = CHECKSUM_NONE; | |
1470ddf7 | 797 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 798 | |
1470ddf7 HX |
799 | exthdrlen = transhdrlen ? rt->dst.header_len : 0; |
800 | length += exthdrlen; | |
801 | transhdrlen += exthdrlen; | |
07df5294 | 802 | mtu = cork->fragsize; |
1da177e4 | 803 | |
d8d1f30b | 804 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
805 | |
806 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
807 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
808 | ||
1470ddf7 | 809 | if (cork->length + length > 0xFFFF - fragheaderlen) { |
c720c7e8 ED |
810 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->inet_dport, |
811 | mtu-exthdrlen); | |
1da177e4 LT |
812 | return -EMSGSIZE; |
813 | } | |
814 | ||
815 | /* | |
816 | * transhdrlen > 0 means that this is the first fragment and we wish | |
817 | * it won't be fragmented in the future. | |
818 | */ | |
819 | if (transhdrlen && | |
820 | length + fragheaderlen <= mtu && | |
d8d1f30b | 821 | rt->dst.dev->features & NETIF_F_V4_CSUM && |
1da177e4 | 822 | !exthdrlen) |
84fa7933 | 823 | csummode = CHECKSUM_PARTIAL; |
1da177e4 | 824 | |
1470ddf7 | 825 | skb = skb_peek_tail(queue); |
26cde9f7 | 826 | |
1470ddf7 | 827 | cork->length += length; |
26cde9f7 | 828 | if (((length > mtu) || (skb && skb_is_gso(skb))) && |
be9164e7 | 829 | (sk->sk_protocol == IPPROTO_UDP) && |
d8d1f30b | 830 | (rt->dst.dev->features & NETIF_F_UFO)) { |
1470ddf7 HX |
831 | err = ip_ufo_append_data(sk, queue, getfrag, from, length, |
832 | hh_len, fragheaderlen, transhdrlen, | |
833 | mtu, flags); | |
baa829d8 | 834 | if (err) |
e89e9cf5 | 835 | goto error; |
e89e9cf5 AR |
836 | return 0; |
837 | } | |
1da177e4 LT |
838 | |
839 | /* So, what's going on in the loop below? | |
840 | * | |
841 | * We use calculated fragment length to generate chained skb, | |
842 | * each of segments is IP fragment ready for sending to network after | |
843 | * adding appropriate IP header. | |
844 | */ | |
845 | ||
26cde9f7 | 846 | if (!skb) |
1da177e4 LT |
847 | goto alloc_new_skb; |
848 | ||
849 | while (length > 0) { | |
850 | /* Check if the remaining data fits into current packet. */ | |
851 | copy = mtu - skb->len; | |
852 | if (copy < length) | |
853 | copy = maxfraglen - skb->len; | |
854 | if (copy <= 0) { | |
855 | char *data; | |
856 | unsigned int datalen; | |
857 | unsigned int fraglen; | |
858 | unsigned int fraggap; | |
859 | unsigned int alloclen; | |
860 | struct sk_buff *skb_prev; | |
861 | alloc_new_skb: | |
862 | skb_prev = skb; | |
863 | if (skb_prev) | |
864 | fraggap = skb_prev->len - maxfraglen; | |
865 | else | |
866 | fraggap = 0; | |
867 | ||
868 | /* | |
869 | * If remaining data exceeds the mtu, | |
870 | * we know we need more fragment(s). | |
871 | */ | |
872 | datalen = length + fraggap; | |
873 | if (datalen > mtu - fragheaderlen) | |
874 | datalen = maxfraglen - fragheaderlen; | |
875 | fraglen = datalen + fragheaderlen; | |
876 | ||
e905a9ed | 877 | if ((flags & MSG_MORE) && |
d8d1f30b | 878 | !(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
879 | alloclen = mtu; |
880 | else | |
59104f06 | 881 | alloclen = fraglen; |
1da177e4 LT |
882 | |
883 | /* The last fragment gets additional space at tail. | |
884 | * Note, with MSG_MORE we overallocate on fragments, | |
885 | * because we have no idea what fragment will be | |
886 | * the last. | |
887 | */ | |
59104f06 | 888 | if (datalen == length + fraggap) { |
d8d1f30b | 889 | alloclen += rt->dst.trailer_len; |
59104f06 ED |
890 | /* make sure mtu is not reached */ |
891 | if (datalen > mtu - fragheaderlen - rt->dst.trailer_len) | |
892 | datalen -= ALIGN(rt->dst.trailer_len, 8); | |
893 | } | |
1da177e4 | 894 | if (transhdrlen) { |
e905a9ed | 895 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
896 | alloclen + hh_len + 15, |
897 | (flags & MSG_DONTWAIT), &err); | |
898 | } else { | |
899 | skb = NULL; | |
900 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
901 | 2 * sk->sk_sndbuf) | |
e905a9ed | 902 | skb = sock_wmalloc(sk, |
1da177e4 LT |
903 | alloclen + hh_len + 15, 1, |
904 | sk->sk_allocation); | |
905 | if (unlikely(skb == NULL)) | |
906 | err = -ENOBUFS; | |
51f31cab PO |
907 | else |
908 | /* only the initial fragment is | |
909 | time stamped */ | |
1470ddf7 | 910 | cork->tx_flags = 0; |
1da177e4 LT |
911 | } |
912 | if (skb == NULL) | |
913 | goto error; | |
914 | ||
915 | /* | |
916 | * Fill in the control structures | |
917 | */ | |
918 | skb->ip_summed = csummode; | |
919 | skb->csum = 0; | |
920 | skb_reserve(skb, hh_len); | |
1470ddf7 | 921 | skb_shinfo(skb)->tx_flags = cork->tx_flags; |
1da177e4 LT |
922 | |
923 | /* | |
924 | * Find where to start putting bytes. | |
925 | */ | |
926 | data = skb_put(skb, fraglen); | |
c14d2450 | 927 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
928 | skb->transport_header = (skb->network_header + |
929 | fragheaderlen); | |
1da177e4 | 930 | data += fragheaderlen; |
1da177e4 LT |
931 | |
932 | if (fraggap) { | |
933 | skb->csum = skb_copy_and_csum_bits( | |
934 | skb_prev, maxfraglen, | |
935 | data + transhdrlen, fraggap, 0); | |
936 | skb_prev->csum = csum_sub(skb_prev->csum, | |
937 | skb->csum); | |
938 | data += fraggap; | |
e9fa4f7b | 939 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
940 | } |
941 | ||
942 | copy = datalen - transhdrlen - fraggap; | |
943 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
944 | err = -EFAULT; | |
945 | kfree_skb(skb); | |
946 | goto error; | |
947 | } | |
948 | ||
949 | offset += copy; | |
950 | length -= datalen - fraggap; | |
951 | transhdrlen = 0; | |
952 | exthdrlen = 0; | |
953 | csummode = CHECKSUM_NONE; | |
954 | ||
955 | /* | |
956 | * Put the packet on the pending queue. | |
957 | */ | |
1470ddf7 | 958 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
959 | continue; |
960 | } | |
961 | ||
962 | if (copy > length) | |
963 | copy = length; | |
964 | ||
d8d1f30b | 965 | if (!(rt->dst.dev->features&NETIF_F_SG)) { |
1da177e4 LT |
966 | unsigned int off; |
967 | ||
968 | off = skb->len; | |
e905a9ed | 969 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
970 | offset, copy, off, skb) < 0) { |
971 | __skb_trim(skb, off); | |
972 | err = -EFAULT; | |
973 | goto error; | |
974 | } | |
975 | } else { | |
976 | int i = skb_shinfo(skb)->nr_frags; | |
977 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; | |
1470ddf7 HX |
978 | struct page *page = cork->page; |
979 | int off = cork->off; | |
1da177e4 LT |
980 | unsigned int left; |
981 | ||
982 | if (page && (left = PAGE_SIZE - off) > 0) { | |
983 | if (copy >= left) | |
984 | copy = left; | |
985 | if (page != frag->page) { | |
986 | if (i == MAX_SKB_FRAGS) { | |
987 | err = -EMSGSIZE; | |
988 | goto error; | |
989 | } | |
990 | get_page(page); | |
1470ddf7 | 991 | skb_fill_page_desc(skb, i, page, off, 0); |
1da177e4 LT |
992 | frag = &skb_shinfo(skb)->frags[i]; |
993 | } | |
994 | } else if (i < MAX_SKB_FRAGS) { | |
995 | if (copy > PAGE_SIZE) | |
996 | copy = PAGE_SIZE; | |
997 | page = alloc_pages(sk->sk_allocation, 0); | |
998 | if (page == NULL) { | |
999 | err = -ENOMEM; | |
1000 | goto error; | |
1001 | } | |
1470ddf7 HX |
1002 | cork->page = page; |
1003 | cork->off = 0; | |
1da177e4 LT |
1004 | |
1005 | skb_fill_page_desc(skb, i, page, 0, 0); | |
1006 | frag = &skb_shinfo(skb)->frags[i]; | |
1da177e4 LT |
1007 | } else { |
1008 | err = -EMSGSIZE; | |
1009 | goto error; | |
1010 | } | |
1011 | if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { | |
1012 | err = -EFAULT; | |
1013 | goto error; | |
1014 | } | |
1470ddf7 | 1015 | cork->off += copy; |
1da177e4 LT |
1016 | frag->size += copy; |
1017 | skb->len += copy; | |
1018 | skb->data_len += copy; | |
f945fa7a HX |
1019 | skb->truesize += copy; |
1020 | atomic_add(copy, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1021 | } |
1022 | offset += copy; | |
1023 | length -= copy; | |
1024 | } | |
1025 | ||
1026 | return 0; | |
1027 | ||
1028 | error: | |
1470ddf7 | 1029 | cork->length -= length; |
5e38e270 | 1030 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
e905a9ed | 1031 | return err; |
1da177e4 LT |
1032 | } |
1033 | ||
1470ddf7 HX |
1034 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, |
1035 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1036 | { | |
1037 | struct inet_sock *inet = inet_sk(sk); | |
f6d8bd05 | 1038 | struct ip_options_rcu *opt; |
1470ddf7 HX |
1039 | struct rtable *rt; |
1040 | ||
1041 | /* | |
1042 | * setup for corking. | |
1043 | */ | |
1044 | opt = ipc->opt; | |
1045 | if (opt) { | |
1046 | if (cork->opt == NULL) { | |
1047 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, | |
1048 | sk->sk_allocation); | |
1049 | if (unlikely(cork->opt == NULL)) | |
1050 | return -ENOBUFS; | |
1051 | } | |
f6d8bd05 | 1052 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); |
1470ddf7 HX |
1053 | cork->flags |= IPCORK_OPT; |
1054 | cork->addr = ipc->addr; | |
1055 | } | |
1056 | rt = *rtp; | |
1057 | if (unlikely(!rt)) | |
1058 | return -EFAULT; | |
1059 | /* | |
1060 | * We steal reference to this route, caller should not release it | |
1061 | */ | |
1062 | *rtp = NULL; | |
1063 | cork->fragsize = inet->pmtudisc == IP_PMTUDISC_PROBE ? | |
1064 | rt->dst.dev->mtu : dst_mtu(rt->dst.path); | |
1065 | cork->dst = &rt->dst; | |
1066 | cork->length = 0; | |
1067 | cork->tx_flags = ipc->tx_flags; | |
1068 | cork->page = NULL; | |
1069 | cork->off = 0; | |
1070 | ||
1071 | return 0; | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1076 | * from many pieces of data. Each pieces will be holded on the socket | |
1077 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1078 | * or non-page data. | |
1079 | * | |
1080 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1081 | * this interface potentially. | |
1082 | * | |
1083 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1084 | */ | |
1085 | int ip_append_data(struct sock *sk, | |
1086 | int getfrag(void *from, char *to, int offset, int len, | |
1087 | int odd, struct sk_buff *skb), | |
1088 | void *from, int length, int transhdrlen, | |
1089 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1090 | unsigned int flags) | |
1091 | { | |
1092 | struct inet_sock *inet = inet_sk(sk); | |
1093 | int err; | |
1094 | ||
1095 | if (flags&MSG_PROBE) | |
1096 | return 0; | |
1097 | ||
1098 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
bdc712b4 | 1099 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); |
1470ddf7 HX |
1100 | if (err) |
1101 | return err; | |
1102 | } else { | |
1103 | transhdrlen = 0; | |
1104 | } | |
1105 | ||
bdc712b4 | 1106 | return __ip_append_data(sk, &sk->sk_write_queue, &inet->cork.base, getfrag, |
1470ddf7 HX |
1107 | from, length, transhdrlen, flags); |
1108 | } | |
1109 | ||
1da177e4 LT |
1110 | ssize_t ip_append_page(struct sock *sk, struct page *page, |
1111 | int offset, size_t size, int flags) | |
1112 | { | |
1113 | struct inet_sock *inet = inet_sk(sk); | |
1114 | struct sk_buff *skb; | |
1115 | struct rtable *rt; | |
1116 | struct ip_options *opt = NULL; | |
bdc712b4 | 1117 | struct inet_cork *cork; |
1da177e4 LT |
1118 | int hh_len; |
1119 | int mtu; | |
1120 | int len; | |
1121 | int err; | |
1122 | unsigned int maxfraglen, fragheaderlen, fraggap; | |
1123 | ||
1124 | if (inet->hdrincl) | |
1125 | return -EPERM; | |
1126 | ||
1127 | if (flags&MSG_PROBE) | |
1128 | return 0; | |
1129 | ||
1130 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1131 | return -EINVAL; | |
1132 | ||
bdc712b4 DM |
1133 | cork = &inet->cork.base; |
1134 | rt = (struct rtable *)cork->dst; | |
1135 | if (cork->flags & IPCORK_OPT) | |
1136 | opt = cork->opt; | |
1da177e4 | 1137 | |
d8d1f30b | 1138 | if (!(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
1139 | return -EOPNOTSUPP; |
1140 | ||
d8d1f30b | 1141 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
bdc712b4 | 1142 | mtu = cork->fragsize; |
1da177e4 LT |
1143 | |
1144 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1145 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1146 | ||
bdc712b4 | 1147 | if (cork->length + size > 0xFFFF - fragheaderlen) { |
c720c7e8 | 1148 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->inet_dport, mtu); |
1da177e4 LT |
1149 | return -EMSGSIZE; |
1150 | } | |
1151 | ||
1152 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1153 | return -EINVAL; | |
1154 | ||
bdc712b4 | 1155 | cork->length += size; |
26cde9f7 HX |
1156 | if ((size + skb->len > mtu) && |
1157 | (sk->sk_protocol == IPPROTO_UDP) && | |
d8d1f30b | 1158 | (rt->dst.dev->features & NETIF_F_UFO)) { |
7967168c | 1159 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 1160 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1161 | } |
e89e9cf5 | 1162 | |
1da177e4 LT |
1163 | |
1164 | while (size > 0) { | |
1165 | int i; | |
1166 | ||
89114afd | 1167 | if (skb_is_gso(skb)) |
e89e9cf5 AR |
1168 | len = size; |
1169 | else { | |
1170 | ||
1171 | /* Check if the remaining data fits into current packet. */ | |
1172 | len = mtu - skb->len; | |
1173 | if (len < size) | |
1174 | len = maxfraglen - skb->len; | |
1175 | } | |
1da177e4 LT |
1176 | if (len <= 0) { |
1177 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1178 | int alloclen; |
1179 | ||
1180 | skb_prev = skb; | |
0d0d2bba | 1181 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1182 | |
1183 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1184 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1185 | if (unlikely(!skb)) { | |
1186 | err = -ENOBUFS; | |
1187 | goto error; | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * Fill in the control structures | |
1192 | */ | |
1193 | skb->ip_summed = CHECKSUM_NONE; | |
1194 | skb->csum = 0; | |
1195 | skb_reserve(skb, hh_len); | |
1196 | ||
1197 | /* | |
1198 | * Find where to start putting bytes. | |
1199 | */ | |
967b05f6 | 1200 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1201 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1202 | skb->transport_header = (skb->network_header + |
1203 | fragheaderlen); | |
1da177e4 | 1204 | if (fraggap) { |
967b05f6 ACM |
1205 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1206 | maxfraglen, | |
9c70220b | 1207 | skb_transport_header(skb), |
967b05f6 | 1208 | fraggap, 0); |
1da177e4 LT |
1209 | skb_prev->csum = csum_sub(skb_prev->csum, |
1210 | skb->csum); | |
e9fa4f7b | 1211 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1212 | } |
1213 | ||
1214 | /* | |
1215 | * Put the packet on the pending queue. | |
1216 | */ | |
1217 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1218 | continue; | |
1219 | } | |
1220 | ||
1221 | i = skb_shinfo(skb)->nr_frags; | |
1222 | if (len > size) | |
1223 | len = size; | |
1224 | if (skb_can_coalesce(skb, i, page, offset)) { | |
1225 | skb_shinfo(skb)->frags[i-1].size += len; | |
1226 | } else if (i < MAX_SKB_FRAGS) { | |
1227 | get_page(page); | |
1228 | skb_fill_page_desc(skb, i, page, offset, len); | |
1229 | } else { | |
1230 | err = -EMSGSIZE; | |
1231 | goto error; | |
1232 | } | |
1233 | ||
1234 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1235 | __wsum csum; |
1da177e4 LT |
1236 | csum = csum_page(page, offset, len); |
1237 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1238 | } | |
1239 | ||
1240 | skb->len += len; | |
1241 | skb->data_len += len; | |
1e34a11d DM |
1242 | skb->truesize += len; |
1243 | atomic_add(len, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1244 | offset += len; |
1245 | size -= len; | |
1246 | } | |
1247 | return 0; | |
1248 | ||
1249 | error: | |
bdc712b4 | 1250 | cork->length -= size; |
5e38e270 | 1251 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1252 | return err; |
1253 | } | |
1254 | ||
1470ddf7 | 1255 | static void ip_cork_release(struct inet_cork *cork) |
429f08e9 | 1256 | { |
1470ddf7 HX |
1257 | cork->flags &= ~IPCORK_OPT; |
1258 | kfree(cork->opt); | |
1259 | cork->opt = NULL; | |
1260 | dst_release(cork->dst); | |
1261 | cork->dst = NULL; | |
429f08e9 PE |
1262 | } |
1263 | ||
1da177e4 LT |
1264 | /* |
1265 | * Combined all pending IP fragments on the socket as one IP datagram | |
1266 | * and push them out. | |
1267 | */ | |
1c32c5ad HX |
1268 | struct sk_buff *__ip_make_skb(struct sock *sk, |
1269 | struct sk_buff_head *queue, | |
1270 | struct inet_cork *cork) | |
1da177e4 LT |
1271 | { |
1272 | struct sk_buff *skb, *tmp_skb; | |
1273 | struct sk_buff **tail_skb; | |
1274 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1275 | struct net *net = sock_net(sk); |
1da177e4 | 1276 | struct ip_options *opt = NULL; |
1470ddf7 | 1277 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 1278 | struct iphdr *iph; |
76ab608d | 1279 | __be16 df = 0; |
1da177e4 | 1280 | __u8 ttl; |
1da177e4 | 1281 | |
1470ddf7 | 1282 | if ((skb = __skb_dequeue(queue)) == NULL) |
1da177e4 LT |
1283 | goto out; |
1284 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1285 | ||
1286 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1287 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1288 | __skb_pull(skb, skb_network_offset(skb)); |
1470ddf7 | 1289 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { |
cfe1fc77 | 1290 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1291 | *tail_skb = tmp_skb; |
1292 | tail_skb = &(tmp_skb->next); | |
1293 | skb->len += tmp_skb->len; | |
1294 | skb->data_len += tmp_skb->len; | |
1295 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1296 | tmp_skb->destructor = NULL; |
1297 | tmp_skb->sk = NULL; | |
1298 | } | |
1299 | ||
1300 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1301 | * to fragment the frame generated here. No matter, what transforms | |
1302 | * how transforms change size of the packet, it will come out. | |
1303 | */ | |
628a5c56 | 1304 | if (inet->pmtudisc < IP_PMTUDISC_DO) |
1da177e4 LT |
1305 | skb->local_df = 1; |
1306 | ||
1307 | /* DF bit is set when we want to see DF on outgoing frames. | |
1308 | * If local_df is set too, we still allow to fragment this frame | |
1309 | * locally. */ | |
628a5c56 | 1310 | if (inet->pmtudisc >= IP_PMTUDISC_DO || |
d8d1f30b CG |
1311 | (skb->len <= dst_mtu(&rt->dst) && |
1312 | ip_dont_fragment(sk, &rt->dst))) | |
1da177e4 LT |
1313 | df = htons(IP_DF); |
1314 | ||
1470ddf7 HX |
1315 | if (cork->flags & IPCORK_OPT) |
1316 | opt = cork->opt; | |
1da177e4 LT |
1317 | |
1318 | if (rt->rt_type == RTN_MULTICAST) | |
1319 | ttl = inet->mc_ttl; | |
1320 | else | |
d8d1f30b | 1321 | ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 LT |
1322 | |
1323 | iph = (struct iphdr *)skb->data; | |
1324 | iph->version = 4; | |
1325 | iph->ihl = 5; | |
1326 | if (opt) { | |
1327 | iph->ihl += opt->optlen>>2; | |
1470ddf7 | 1328 | ip_options_build(skb, opt, cork->addr, rt, 0); |
1da177e4 LT |
1329 | } |
1330 | iph->tos = inet->tos; | |
1da177e4 | 1331 | iph->frag_off = df; |
d8d1f30b | 1332 | ip_select_ident(iph, &rt->dst, sk); |
1da177e4 LT |
1333 | iph->ttl = ttl; |
1334 | iph->protocol = sk->sk_protocol; | |
1335 | iph->saddr = rt->rt_src; | |
1336 | iph->daddr = rt->rt_dst; | |
1da177e4 LT |
1337 | |
1338 | skb->priority = sk->sk_priority; | |
4a19ec58 | 1339 | skb->mark = sk->sk_mark; |
a21bba94 ED |
1340 | /* |
1341 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1342 | * on dst refcount | |
1343 | */ | |
1470ddf7 | 1344 | cork->dst = NULL; |
d8d1f30b | 1345 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 1346 | |
96793b48 | 1347 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1348 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1349 | skb_transport_header(skb))->type); |
1350 | ||
1c32c5ad HX |
1351 | ip_cork_release(cork); |
1352 | out: | |
1353 | return skb; | |
1354 | } | |
1355 | ||
1356 | int ip_send_skb(struct sk_buff *skb) | |
1357 | { | |
1358 | struct net *net = sock_net(skb->sk); | |
1359 | int err; | |
1360 | ||
c439cb2e | 1361 | err = ip_local_out(skb); |
1da177e4 LT |
1362 | if (err) { |
1363 | if (err > 0) | |
6ce9e7b5 | 1364 | err = net_xmit_errno(err); |
1da177e4 | 1365 | if (err) |
1c32c5ad | 1366 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1367 | } |
1368 | ||
1da177e4 | 1369 | return err; |
1da177e4 LT |
1370 | } |
1371 | ||
1470ddf7 HX |
1372 | int ip_push_pending_frames(struct sock *sk) |
1373 | { | |
1c32c5ad HX |
1374 | struct sk_buff *skb; |
1375 | ||
1376 | skb = ip_finish_skb(sk); | |
1377 | if (!skb) | |
1378 | return 0; | |
1379 | ||
1380 | /* Netfilter gets whole the not fragmented skb. */ | |
1381 | return ip_send_skb(skb); | |
1470ddf7 HX |
1382 | } |
1383 | ||
1da177e4 LT |
1384 | /* |
1385 | * Throw away all pending data on the socket. | |
1386 | */ | |
1470ddf7 HX |
1387 | static void __ip_flush_pending_frames(struct sock *sk, |
1388 | struct sk_buff_head *queue, | |
1389 | struct inet_cork *cork) | |
1da177e4 | 1390 | { |
1da177e4 LT |
1391 | struct sk_buff *skb; |
1392 | ||
1470ddf7 | 1393 | while ((skb = __skb_dequeue_tail(queue)) != NULL) |
1da177e4 LT |
1394 | kfree_skb(skb); |
1395 | ||
1470ddf7 HX |
1396 | ip_cork_release(cork); |
1397 | } | |
1398 | ||
1399 | void ip_flush_pending_frames(struct sock *sk) | |
1400 | { | |
bdc712b4 | 1401 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); |
1da177e4 LT |
1402 | } |
1403 | ||
1c32c5ad HX |
1404 | struct sk_buff *ip_make_skb(struct sock *sk, |
1405 | int getfrag(void *from, char *to, int offset, | |
1406 | int len, int odd, struct sk_buff *skb), | |
1407 | void *from, int length, int transhdrlen, | |
1408 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1409 | unsigned int flags) | |
1410 | { | |
b80d7226 | 1411 | struct inet_cork cork; |
1c32c5ad HX |
1412 | struct sk_buff_head queue; |
1413 | int err; | |
1414 | ||
1415 | if (flags & MSG_PROBE) | |
1416 | return NULL; | |
1417 | ||
1418 | __skb_queue_head_init(&queue); | |
1419 | ||
b80d7226 DM |
1420 | cork.flags = 0; |
1421 | cork.addr = 0; | |
1422 | cork.opt = 0; | |
1c32c5ad HX |
1423 | err = ip_setup_cork(sk, &cork, ipc, rtp); |
1424 | if (err) | |
1425 | return ERR_PTR(err); | |
1426 | ||
1427 | err = __ip_append_data(sk, &queue, &cork, getfrag, | |
1428 | from, length, transhdrlen, flags); | |
1429 | if (err) { | |
1430 | __ip_flush_pending_frames(sk, &queue, &cork); | |
1431 | return ERR_PTR(err); | |
1432 | } | |
1433 | ||
1434 | return __ip_make_skb(sk, &queue, &cork); | |
1435 | } | |
1da177e4 LT |
1436 | |
1437 | /* | |
1438 | * Fetch data from kernel space and fill in checksum if needed. | |
1439 | */ | |
e905a9ed | 1440 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1441 | int len, int odd, struct sk_buff *skb) |
1442 | { | |
5084205f | 1443 | __wsum csum; |
1da177e4 LT |
1444 | |
1445 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1446 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1447 | return 0; |
1da177e4 LT |
1448 | } |
1449 | ||
e905a9ed | 1450 | /* |
1da177e4 LT |
1451 | * Generic function to send a packet as reply to another packet. |
1452 | * Used to send TCP resets so far. ICMP should use this function too. | |
1453 | * | |
e905a9ed | 1454 | * Should run single threaded per socket because it uses the sock |
1da177e4 | 1455 | * structure to pass arguments. |
1da177e4 LT |
1456 | */ |
1457 | void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg, | |
1458 | unsigned int len) | |
1459 | { | |
1460 | struct inet_sock *inet = inet_sk(sk); | |
f6d8bd05 | 1461 | struct ip_options_data replyopts; |
1da177e4 | 1462 | struct ipcm_cookie ipc; |
3ca3c68e | 1463 | __be32 daddr; |
511c3f92 | 1464 | struct rtable *rt = skb_rtable(skb); |
1da177e4 | 1465 | |
f6d8bd05 | 1466 | if (ip_options_echo(&replyopts.opt.opt, skb)) |
1da177e4 LT |
1467 | return; |
1468 | ||
1469 | daddr = ipc.addr = rt->rt_src; | |
1470 | ipc.opt = NULL; | |
2244d07b | 1471 | ipc.tx_flags = 0; |
1da177e4 | 1472 | |
f6d8bd05 | 1473 | if (replyopts.opt.opt.optlen) { |
1da177e4 LT |
1474 | ipc.opt = &replyopts.opt; |
1475 | ||
f6d8bd05 ED |
1476 | if (replyopts.opt.opt.srr) |
1477 | daddr = replyopts.opt.opt.faddr; | |
1da177e4 LT |
1478 | } |
1479 | ||
1480 | { | |
538de0e0 DM |
1481 | struct flowi4 fl4; |
1482 | ||
1483 | flowi4_init_output(&fl4, arg->bound_dev_if, 0, | |
1484 | RT_TOS(ip_hdr(skb)->tos), | |
1485 | RT_SCOPE_UNIVERSE, sk->sk_protocol, | |
1486 | ip_reply_arg_flowi_flags(arg), | |
1487 | daddr, rt->rt_spec_dst, | |
1488 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest); | |
9d6ec938 DM |
1489 | security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); |
1490 | rt = ip_route_output_key(sock_net(sk), &fl4); | |
b23dd4fe | 1491 | if (IS_ERR(rt)) |
1da177e4 LT |
1492 | return; |
1493 | } | |
1494 | ||
1495 | /* And let IP do all the hard work. | |
1496 | ||
1497 | This chunk is not reenterable, hence spinlock. | |
1498 | Note that it uses the fact, that this function is called | |
1499 | with locally disabled BH and that sk cannot be already spinlocked. | |
1500 | */ | |
1501 | bh_lock_sock(sk); | |
eddc9ec5 | 1502 | inet->tos = ip_hdr(skb)->tos; |
1da177e4 | 1503 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1504 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1505 | sk->sk_bound_dev_if = arg->bound_dev_if; |
1da177e4 | 1506 | ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0, |
2e77d89b | 1507 | &ipc, &rt, MSG_DONTWAIT); |
1da177e4 LT |
1508 | if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { |
1509 | if (arg->csumoffset >= 0) | |
9c70220b ACM |
1510 | *((__sum16 *)skb_transport_header(skb) + |
1511 | arg->csumoffset) = csum_fold(csum_add(skb->csum, | |
1512 | arg->csum)); | |
1da177e4 LT |
1513 | skb->ip_summed = CHECKSUM_NONE; |
1514 | ip_push_pending_frames(sk); | |
1515 | } | |
1516 | ||
1517 | bh_unlock_sock(sk); | |
1518 | ||
1519 | ip_rt_put(rt); | |
1520 | } | |
1521 | ||
1da177e4 LT |
1522 | void __init ip_init(void) |
1523 | { | |
1da177e4 LT |
1524 | ip_rt_init(); |
1525 | inet_initpeers(); | |
1526 | ||
1527 | #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) | |
1528 | igmp_mc_proc_init(); | |
1529 | #endif | |
1530 | } |