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 LT |
85 | |
86 | /* Generate a checksum for an outgoing IP datagram. */ | |
2fbd9679 | 87 | void ip_send_check(struct iphdr *iph) |
1da177e4 LT |
88 | { |
89 | iph->check = 0; | |
90 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
91 | } | |
4bc2f18b | 92 | EXPORT_SYMBOL(ip_send_check); |
1da177e4 | 93 | |
c439cb2e HX |
94 | int __ip_local_out(struct sk_buff *skb) |
95 | { | |
96 | struct iphdr *iph = ip_hdr(skb); | |
97 | ||
98 | iph->tot_len = htons(skb->len); | |
99 | ip_send_check(iph); | |
9bbc768a JE |
100 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL, |
101 | skb_dst(skb)->dev, dst_output); | |
c439cb2e HX |
102 | } |
103 | ||
aad88724 | 104 | int ip_local_out_sk(struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
105 | { |
106 | int err; | |
107 | ||
108 | err = __ip_local_out(skb); | |
109 | if (likely(err == 1)) | |
aad88724 | 110 | err = dst_output_sk(sk, skb); |
c439cb2e HX |
111 | |
112 | return err; | |
113 | } | |
aad88724 | 114 | EXPORT_SYMBOL_GPL(ip_local_out_sk); |
c439cb2e | 115 | |
1da177e4 LT |
116 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) |
117 | { | |
118 | int ttl = inet->uc_ttl; | |
119 | ||
120 | if (ttl < 0) | |
323e126f | 121 | ttl = ip4_dst_hoplimit(dst); |
1da177e4 LT |
122 | return ttl; |
123 | } | |
124 | ||
e905a9ed | 125 | /* |
1da177e4 LT |
126 | * Add an ip header to a skbuff and send it out. |
127 | * | |
128 | */ | |
129 | int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, | |
f6d8bd05 | 130 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) |
1da177e4 LT |
131 | { |
132 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 133 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
134 | struct iphdr *iph; |
135 | ||
136 | /* Build the IP header. */ | |
f6d8bd05 | 137 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); |
8856dfa3 | 138 | skb_reset_network_header(skb); |
eddc9ec5 | 139 | iph = ip_hdr(skb); |
1da177e4 LT |
140 | iph->version = 4; |
141 | iph->ihl = 5; | |
142 | iph->tos = inet->tos; | |
d8d1f30b | 143 | if (ip_dont_fragment(sk, &rt->dst)) |
1da177e4 LT |
144 | iph->frag_off = htons(IP_DF); |
145 | else | |
146 | iph->frag_off = 0; | |
d8d1f30b | 147 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
dd927a26 DM |
148 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); |
149 | iph->saddr = saddr; | |
1da177e4 | 150 | iph->protocol = sk->sk_protocol; |
73f156a6 | 151 | ip_select_ident(skb, sk); |
1da177e4 | 152 | |
f6d8bd05 ED |
153 | if (opt && opt->opt.optlen) { |
154 | iph->ihl += opt->opt.optlen>>2; | |
155 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
1da177e4 | 156 | } |
1da177e4 LT |
157 | |
158 | skb->priority = sk->sk_priority; | |
4a19ec58 | 159 | skb->mark = sk->sk_mark; |
1da177e4 LT |
160 | |
161 | /* Send it out. */ | |
c439cb2e | 162 | return ip_local_out(skb); |
1da177e4 | 163 | } |
d8c97a94 ACM |
164 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
165 | ||
1da177e4 LT |
166 | static inline int ip_finish_output2(struct sk_buff *skb) |
167 | { | |
adf30907 | 168 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 169 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 170 | struct net_device *dev = dst->dev; |
c2636b4d | 171 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
f6b72b62 | 172 | struct neighbour *neigh; |
a263b309 | 173 | u32 nexthop; |
1da177e4 | 174 | |
edf391ff NH |
175 | if (rt->rt_type == RTN_MULTICAST) { |
176 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); | |
177 | } else if (rt->rt_type == RTN_BROADCAST) | |
178 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); | |
80787ebc | 179 | |
1da177e4 | 180 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 181 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
182 | struct sk_buff *skb2; |
183 | ||
184 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
185 | if (skb2 == NULL) { | |
186 | kfree_skb(skb); | |
187 | return -ENOMEM; | |
188 | } | |
189 | if (skb->sk) | |
190 | skb_set_owner_w(skb2, skb->sk); | |
5d0ba55b | 191 | consume_skb(skb); |
1da177e4 LT |
192 | skb = skb2; |
193 | } | |
194 | ||
a263b309 | 195 | rcu_read_lock_bh(); |
155e8336 | 196 | nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); |
a263b309 DM |
197 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); |
198 | if (unlikely(!neigh)) | |
199 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
9871f1ad | 200 | if (!IS_ERR(neigh)) { |
5110effe | 201 | int res = dst_neigh_output(dst, neigh, skb); |
f2c31e32 | 202 | |
a263b309 | 203 | rcu_read_unlock_bh(); |
f2c31e32 ED |
204 | return res; |
205 | } | |
a263b309 | 206 | rcu_read_unlock_bh(); |
05e3aa09 | 207 | |
e87cc472 JP |
208 | net_dbg_ratelimited("%s: No header cache and no neighbour!\n", |
209 | __func__); | |
1da177e4 LT |
210 | kfree_skb(skb); |
211 | return -EINVAL; | |
212 | } | |
213 | ||
c7ba65d7 FW |
214 | static int ip_finish_output_gso(struct sk_buff *skb) |
215 | { | |
216 | netdev_features_t features; | |
217 | struct sk_buff *segs; | |
218 | int ret = 0; | |
219 | ||
220 | /* common case: locally created skb or seglen is <= mtu */ | |
221 | if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) || | |
222 | skb_gso_network_seglen(skb) <= ip_skb_dst_mtu(skb)) | |
223 | return ip_finish_output2(skb); | |
224 | ||
225 | /* Slowpath - GSO segment length is exceeding the dst MTU. | |
226 | * | |
227 | * This can happen in two cases: | |
228 | * 1) TCP GRO packet, DF bit not set | |
229 | * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly | |
230 | * from host network stack. | |
231 | */ | |
232 | features = netif_skb_features(skb); | |
233 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); | |
234 | if (IS_ERR(segs)) { | |
235 | kfree_skb(skb); | |
236 | return -ENOMEM; | |
237 | } | |
238 | ||
239 | consume_skb(skb); | |
240 | ||
241 | do { | |
242 | struct sk_buff *nskb = segs->next; | |
243 | int err; | |
244 | ||
245 | segs->next = NULL; | |
246 | err = ip_fragment(segs, ip_finish_output2); | |
247 | ||
248 | if (err && ret == 0) | |
249 | ret = err; | |
250 | segs = nskb; | |
251 | } while (segs); | |
252 | ||
253 | return ret; | |
254 | } | |
255 | ||
861d0486 | 256 | static int ip_finish_output(struct sk_buff *skb) |
1da177e4 | 257 | { |
5c901daa PM |
258 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
259 | /* Policy lookup after SNAT yielded a new policy */ | |
adf30907 | 260 | if (skb_dst(skb)->xfrm != NULL) { |
48d5cad8 PM |
261 | IPCB(skb)->flags |= IPSKB_REROUTED; |
262 | return dst_output(skb); | |
263 | } | |
5c901daa | 264 | #endif |
c7ba65d7 FW |
265 | if (skb_is_gso(skb)) |
266 | return ip_finish_output_gso(skb); | |
267 | ||
268 | if (skb->len > ip_skb_dst_mtu(skb)) | |
1bd9bef6 | 269 | return ip_fragment(skb, ip_finish_output2); |
c7ba65d7 FW |
270 | |
271 | return ip_finish_output2(skb); | |
1da177e4 LT |
272 | } |
273 | ||
aad88724 | 274 | int ip_mc_output(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 275 | { |
511c3f92 | 276 | struct rtable *rt = skb_rtable(skb); |
d8d1f30b | 277 | struct net_device *dev = rt->dst.dev; |
1da177e4 LT |
278 | |
279 | /* | |
280 | * If the indicated interface is up and running, send the packet. | |
281 | */ | |
edf391ff | 282 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
283 | |
284 | skb->dev = dev; | |
285 | skb->protocol = htons(ETH_P_IP); | |
286 | ||
287 | /* | |
288 | * Multicasts are looped back for other local users | |
289 | */ | |
290 | ||
291 | if (rt->rt_flags&RTCF_MULTICAST) { | |
7ad6848c | 292 | if (sk_mc_loop(sk) |
1da177e4 LT |
293 | #ifdef CONFIG_IP_MROUTE |
294 | /* Small optimization: do not loopback not local frames, | |
295 | which returned after forwarding; they will be dropped | |
296 | by ip_mr_input in any case. | |
297 | Note, that local frames are looped back to be delivered | |
298 | to local recipients. | |
299 | ||
300 | This check is duplicated in ip_mr_input at the moment. | |
301 | */ | |
9d4fb27d JP |
302 | && |
303 | ((rt->rt_flags & RTCF_LOCAL) || | |
304 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 305 | #endif |
9d4fb27d | 306 | ) { |
1da177e4 LT |
307 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
308 | if (newskb) | |
9bbc768a JE |
309 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
310 | newskb, NULL, newskb->dev, | |
95603e22 | 311 | dev_loopback_xmit); |
1da177e4 LT |
312 | } |
313 | ||
314 | /* Multicasts with ttl 0 must not go beyond the host */ | |
315 | ||
eddc9ec5 | 316 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
317 | kfree_skb(skb); |
318 | return 0; | |
319 | } | |
320 | } | |
321 | ||
322 | if (rt->rt_flags&RTCF_BROADCAST) { | |
323 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
324 | if (newskb) | |
9bbc768a | 325 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb, |
95603e22 | 326 | NULL, newskb->dev, dev_loopback_xmit); |
1da177e4 LT |
327 | } |
328 | ||
9bbc768a JE |
329 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, |
330 | skb->dev, ip_finish_output, | |
48d5cad8 | 331 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
332 | } |
333 | ||
aad88724 | 334 | int ip_output(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 335 | { |
adf30907 | 336 | struct net_device *dev = skb_dst(skb)->dev; |
1bd9bef6 | 337 | |
edf391ff | 338 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 339 | |
1bd9bef6 PM |
340 | skb->dev = dev; |
341 | skb->protocol = htons(ETH_P_IP); | |
342 | ||
9bbc768a | 343 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev, |
e905a9ed | 344 | ip_finish_output, |
48d5cad8 | 345 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
346 | } |
347 | ||
84f9307c ED |
348 | /* |
349 | * copy saddr and daddr, possibly using 64bit load/stores | |
350 | * Equivalent to : | |
351 | * iph->saddr = fl4->saddr; | |
352 | * iph->daddr = fl4->daddr; | |
353 | */ | |
354 | static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) | |
355 | { | |
356 | BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != | |
357 | offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); | |
358 | memcpy(&iph->saddr, &fl4->saddr, | |
359 | sizeof(fl4->saddr) + sizeof(fl4->daddr)); | |
360 | } | |
361 | ||
b0270e91 ED |
362 | /* Note: skb->sk can be different from sk, in case of tunnels */ |
363 | int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) | |
1da177e4 | 364 | { |
1da177e4 | 365 | struct inet_sock *inet = inet_sk(sk); |
f6d8bd05 | 366 | struct ip_options_rcu *inet_opt; |
b57ae01a | 367 | struct flowi4 *fl4; |
1da177e4 LT |
368 | struct rtable *rt; |
369 | struct iphdr *iph; | |
ab6e3feb | 370 | int res; |
1da177e4 LT |
371 | |
372 | /* Skip all of this if the packet is already routed, | |
373 | * f.e. by something like SCTP. | |
374 | */ | |
ab6e3feb | 375 | rcu_read_lock(); |
f6d8bd05 | 376 | inet_opt = rcu_dereference(inet->inet_opt); |
ea4fc0d6 | 377 | fl4 = &fl->u.ip4; |
511c3f92 | 378 | rt = skb_rtable(skb); |
1da177e4 LT |
379 | if (rt != NULL) |
380 | goto packet_routed; | |
381 | ||
382 | /* Make sure we can route this packet. */ | |
383 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
384 | if (rt == NULL) { | |
3ca3c68e | 385 | __be32 daddr; |
1da177e4 LT |
386 | |
387 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 388 | daddr = inet->inet_daddr; |
f6d8bd05 ED |
389 | if (inet_opt && inet_opt->opt.srr) |
390 | daddr = inet_opt->opt.faddr; | |
1da177e4 | 391 | |
78fbfd8a DM |
392 | /* If this fails, retransmit mechanism of transport layer will |
393 | * keep trying until route appears or the connection times | |
394 | * itself out. | |
395 | */ | |
b57ae01a | 396 | rt = ip_route_output_ports(sock_net(sk), fl4, sk, |
78fbfd8a DM |
397 | daddr, inet->inet_saddr, |
398 | inet->inet_dport, | |
399 | inet->inet_sport, | |
400 | sk->sk_protocol, | |
401 | RT_CONN_FLAGS(sk), | |
402 | sk->sk_bound_dev_if); | |
403 | if (IS_ERR(rt)) | |
404 | goto no_route; | |
d8d1f30b | 405 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 406 | } |
d8d1f30b | 407 | skb_dst_set_noref(skb, &rt->dst); |
1da177e4 LT |
408 | |
409 | packet_routed: | |
155e8336 | 410 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) |
1da177e4 LT |
411 | goto no_route; |
412 | ||
413 | /* OK, we know where to send it, allocate and build IP header. */ | |
f6d8bd05 | 414 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); |
8856dfa3 | 415 | skb_reset_network_header(skb); |
eddc9ec5 | 416 | iph = ip_hdr(skb); |
714e85be | 417 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
60ff7467 | 418 | if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) |
1da177e4 LT |
419 | iph->frag_off = htons(IP_DF); |
420 | else | |
421 | iph->frag_off = 0; | |
d8d1f30b | 422 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 423 | iph->protocol = sk->sk_protocol; |
84f9307c ED |
424 | ip_copy_addrs(iph, fl4); |
425 | ||
1da177e4 LT |
426 | /* Transport layer set skb->h.foo itself. */ |
427 | ||
f6d8bd05 ED |
428 | if (inet_opt && inet_opt->opt.optlen) { |
429 | iph->ihl += inet_opt->opt.optlen >> 2; | |
430 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
1da177e4 LT |
431 | } |
432 | ||
73f156a6 | 433 | ip_select_ident_segs(skb, sk, skb_shinfo(skb)->gso_segs ?: 1); |
1da177e4 | 434 | |
b0270e91 | 435 | /* TODO : should we use skb->sk here instead of sk ? */ |
1da177e4 | 436 | skb->priority = sk->sk_priority; |
4a19ec58 | 437 | skb->mark = sk->sk_mark; |
1da177e4 | 438 | |
ab6e3feb ED |
439 | res = ip_local_out(skb); |
440 | rcu_read_unlock(); | |
441 | return res; | |
1da177e4 LT |
442 | |
443 | no_route: | |
ab6e3feb | 444 | rcu_read_unlock(); |
5e38e270 | 445 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
446 | kfree_skb(skb); |
447 | return -EHOSTUNREACH; | |
448 | } | |
4bc2f18b | 449 | EXPORT_SYMBOL(ip_queue_xmit); |
1da177e4 LT |
450 | |
451 | ||
452 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
453 | { | |
454 | to->pkt_type = from->pkt_type; | |
455 | to->priority = from->priority; | |
456 | to->protocol = from->protocol; | |
adf30907 | 457 | skb_dst_drop(to); |
fe76cda3 | 458 | skb_dst_copy(to, from); |
1da177e4 | 459 | to->dev = from->dev; |
82e91ffe | 460 | to->mark = from->mark; |
1da177e4 LT |
461 | |
462 | /* Copy the flags to each fragment. */ | |
463 | IPCB(to)->flags = IPCB(from)->flags; | |
464 | ||
465 | #ifdef CONFIG_NET_SCHED | |
466 | to->tc_index = from->tc_index; | |
467 | #endif | |
e7ac05f3 | 468 | nf_copy(to, from); |
c98d80ed JA |
469 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
470 | to->ipvs_property = from->ipvs_property; | |
1da177e4 | 471 | #endif |
984bc16c | 472 | skb_copy_secmark(to, from); |
1da177e4 LT |
473 | } |
474 | ||
475 | /* | |
476 | * This IP datagram is too large to be sent in one piece. Break it up into | |
477 | * smaller pieces (each of size equal to IP header plus | |
478 | * a block of the data of the original IP data part) that will yet fit in a | |
479 | * single device frame, and queue such a frame for sending. | |
480 | */ | |
481 | ||
d9319100 | 482 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) |
1da177e4 LT |
483 | { |
484 | struct iphdr *iph; | |
1da177e4 LT |
485 | int ptr; |
486 | struct net_device *dev; | |
487 | struct sk_buff *skb2; | |
c893b806 | 488 | unsigned int mtu, hlen, left, len, ll_rs; |
1da177e4 | 489 | int offset; |
76ab608d | 490 | __be16 not_last_frag; |
511c3f92 | 491 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
492 | int err = 0; |
493 | ||
d8d1f30b | 494 | dev = rt->dst.dev; |
1da177e4 LT |
495 | |
496 | /* | |
497 | * Point into the IP datagram header. | |
498 | */ | |
499 | ||
eddc9ec5 | 500 | iph = ip_hdr(skb); |
1da177e4 | 501 | |
69647ce4 | 502 | mtu = ip_skb_dst_mtu(skb); |
60ff7467 | 503 | if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) || |
5f2d04f1 | 504 | (IPCB(skb)->frag_max_size && |
f87c10a8 | 505 | IPCB(skb)->frag_max_size > mtu))) { |
5e38e270 | 506 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 507 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
f87c10a8 | 508 | htonl(mtu)); |
1da177e4 LT |
509 | kfree_skb(skb); |
510 | return -EMSGSIZE; | |
511 | } | |
512 | ||
513 | /* | |
514 | * Setup starting values. | |
515 | */ | |
516 | ||
517 | hlen = iph->ihl * 4; | |
f87c10a8 | 518 | mtu = mtu - hlen; /* Size of data space */ |
6c79bf0f BDS |
519 | #ifdef CONFIG_BRIDGE_NETFILTER |
520 | if (skb->nf_bridge) | |
521 | mtu -= nf_bridge_mtu_reduction(skb); | |
522 | #endif | |
89cee8b1 | 523 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
524 | |
525 | /* When frag_list is given, use it. First, check its validity: | |
526 | * some transformers could create wrong frag_list or break existing | |
527 | * one, it is not prohibited. In this case fall back to copying. | |
528 | * | |
529 | * LATER: this step can be merged to real generation of fragments, | |
530 | * we can switch to copy when see the first bad fragment. | |
531 | */ | |
21dc3301 | 532 | if (skb_has_frag_list(skb)) { |
3d13008e | 533 | struct sk_buff *frag, *frag2; |
1da177e4 LT |
534 | int first_len = skb_pagelen(skb); |
535 | ||
536 | if (first_len - hlen > mtu || | |
537 | ((first_len - hlen) & 7) || | |
56f8a75c | 538 | ip_is_fragment(iph) || |
1da177e4 LT |
539 | skb_cloned(skb)) |
540 | goto slow_path; | |
541 | ||
d7fcf1a5 | 542 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
543 | /* Correct geometry. */ |
544 | if (frag->len > mtu || | |
545 | ((frag->len & 7) && frag->next) || | |
546 | skb_headroom(frag) < hlen) | |
3d13008e | 547 | goto slow_path_clean; |
1da177e4 LT |
548 | |
549 | /* Partially cloned skb? */ | |
550 | if (skb_shared(frag)) | |
3d13008e | 551 | goto slow_path_clean; |
2fdba6b0 HX |
552 | |
553 | BUG_ON(frag->sk); | |
554 | if (skb->sk) { | |
2fdba6b0 HX |
555 | frag->sk = skb->sk; |
556 | frag->destructor = sock_wfree; | |
2fdba6b0 | 557 | } |
3d13008e | 558 | skb->truesize -= frag->truesize; |
1da177e4 LT |
559 | } |
560 | ||
561 | /* Everything is OK. Generate! */ | |
562 | ||
563 | err = 0; | |
564 | offset = 0; | |
565 | frag = skb_shinfo(skb)->frag_list; | |
d7fcf1a5 | 566 | skb_frag_list_init(skb); |
1da177e4 LT |
567 | skb->data_len = first_len - skb_headlen(skb); |
568 | skb->len = first_len; | |
569 | iph->tot_len = htons(first_len); | |
570 | iph->frag_off = htons(IP_MF); | |
571 | ip_send_check(iph); | |
572 | ||
573 | for (;;) { | |
574 | /* Prepare header of the next frame, | |
575 | * before previous one went down. */ | |
576 | if (frag) { | |
577 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 578 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
579 | __skb_push(frag, hlen); |
580 | skb_reset_network_header(frag); | |
d56f90a7 | 581 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 582 | iph = ip_hdr(frag); |
1da177e4 LT |
583 | iph->tot_len = htons(frag->len); |
584 | ip_copy_metadata(frag, skb); | |
585 | if (offset == 0) | |
586 | ip_options_fragment(frag); | |
587 | offset += skb->len - hlen; | |
588 | iph->frag_off = htons(offset>>3); | |
589 | if (frag->next != NULL) | |
590 | iph->frag_off |= htons(IP_MF); | |
591 | /* Ready, complete checksum */ | |
592 | ip_send_check(iph); | |
593 | } | |
594 | ||
595 | err = output(skb); | |
596 | ||
dafee490 | 597 | if (!err) |
5e38e270 | 598 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
599 | if (err || !frag) |
600 | break; | |
601 | ||
602 | skb = frag; | |
603 | frag = skb->next; | |
604 | skb->next = NULL; | |
605 | } | |
606 | ||
607 | if (err == 0) { | |
5e38e270 | 608 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
609 | return 0; |
610 | } | |
611 | ||
612 | while (frag) { | |
613 | skb = frag->next; | |
614 | kfree_skb(frag); | |
615 | frag = skb; | |
616 | } | |
5e38e270 | 617 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 618 | return err; |
3d13008e ED |
619 | |
620 | slow_path_clean: | |
621 | skb_walk_frags(skb, frag2) { | |
622 | if (frag2 == frag) | |
623 | break; | |
624 | frag2->sk = NULL; | |
625 | frag2->destructor = NULL; | |
626 | skb->truesize += frag2->truesize; | |
627 | } | |
1da177e4 LT |
628 | } |
629 | ||
630 | slow_path: | |
fc70fb64 AD |
631 | /* for offloaded checksums cleanup checksum before fragmentation */ |
632 | if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb)) | |
633 | goto fail; | |
c9af6db4 | 634 | iph = ip_hdr(skb); |
fc70fb64 | 635 | |
1da177e4 | 636 | left = skb->len - hlen; /* Space per frame */ |
49085bd7 | 637 | ptr = hlen; /* Where to start from */ |
1da177e4 | 638 | |
1da177e4 | 639 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, |
9bcfcaf5 SH |
640 | * we need to make room for the encapsulating header |
641 | */ | |
c893b806 | 642 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb)); |
9bcfcaf5 | 643 | |
1da177e4 LT |
644 | /* |
645 | * Fragment the datagram. | |
646 | */ | |
647 | ||
648 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
649 | not_last_frag = iph->frag_off & htons(IP_MF); | |
650 | ||
651 | /* | |
652 | * Keep copying data until we run out. | |
653 | */ | |
654 | ||
132adf54 | 655 | while (left > 0) { |
1da177e4 LT |
656 | len = left; |
657 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
658 | if (len > mtu) | |
659 | len = mtu; | |
25985edc | 660 | /* IF: we are not sending up to and including the packet end |
1da177e4 LT |
661 | then align the next start on an eight byte boundary */ |
662 | if (len < left) { | |
663 | len &= ~7; | |
664 | } | |
665 | /* | |
666 | * Allocate buffer. | |
667 | */ | |
668 | ||
669 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 670 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
671 | err = -ENOMEM; |
672 | goto fail; | |
673 | } | |
674 | ||
675 | /* | |
676 | * Set up data on packet | |
677 | */ | |
678 | ||
679 | ip_copy_metadata(skb2, skb); | |
680 | skb_reserve(skb2, ll_rs); | |
681 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 682 | skb_reset_network_header(skb2); |
b0e380b1 | 683 | skb2->transport_header = skb2->network_header + hlen; |
1da177e4 LT |
684 | |
685 | /* | |
686 | * Charge the memory for the fragment to any owner | |
687 | * it might possess | |
688 | */ | |
689 | ||
690 | if (skb->sk) | |
691 | skb_set_owner_w(skb2, skb->sk); | |
692 | ||
693 | /* | |
694 | * Copy the packet header into the new buffer. | |
695 | */ | |
696 | ||
d626f62b | 697 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); |
1da177e4 LT |
698 | |
699 | /* | |
700 | * Copy a block of the IP datagram. | |
701 | */ | |
bff9b61c | 702 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) |
1da177e4 LT |
703 | BUG(); |
704 | left -= len; | |
705 | ||
706 | /* | |
707 | * Fill in the new header fields. | |
708 | */ | |
eddc9ec5 | 709 | iph = ip_hdr(skb2); |
1da177e4 LT |
710 | iph->frag_off = htons((offset >> 3)); |
711 | ||
712 | /* ANK: dirty, but effective trick. Upgrade options only if | |
713 | * the segment to be fragmented was THE FIRST (otherwise, | |
714 | * options are already fixed) and make it ONCE | |
715 | * on the initial skb, so that all the following fragments | |
716 | * will inherit fixed options. | |
717 | */ | |
718 | if (offset == 0) | |
719 | ip_options_fragment(skb); | |
720 | ||
721 | /* | |
722 | * Added AC : If we are fragmenting a fragment that's not the | |
723 | * last fragment then keep MF on each bit | |
724 | */ | |
725 | if (left > 0 || not_last_frag) | |
726 | iph->frag_off |= htons(IP_MF); | |
727 | ptr += len; | |
728 | offset += len; | |
729 | ||
730 | /* | |
731 | * Put this fragment into the sending queue. | |
732 | */ | |
1da177e4 LT |
733 | iph->tot_len = htons(len + hlen); |
734 | ||
735 | ip_send_check(iph); | |
736 | ||
737 | err = output(skb2); | |
738 | if (err) | |
739 | goto fail; | |
dafee490 | 740 | |
5e38e270 | 741 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 | 742 | } |
5d0ba55b | 743 | consume_skb(skb); |
5e38e270 | 744 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
745 | return err; |
746 | ||
747 | fail: | |
e905a9ed | 748 | kfree_skb(skb); |
5e38e270 | 749 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
750 | return err; |
751 | } | |
2e2f7aef PM |
752 | EXPORT_SYMBOL(ip_fragment); |
753 | ||
1da177e4 LT |
754 | int |
755 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
756 | { | |
757 | struct iovec *iov = from; | |
758 | ||
84fa7933 | 759 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1da177e4 LT |
760 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) |
761 | return -EFAULT; | |
762 | } else { | |
44bb9363 | 763 | __wsum csum = 0; |
1da177e4 LT |
764 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) |
765 | return -EFAULT; | |
766 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
767 | } | |
768 | return 0; | |
769 | } | |
4bc2f18b | 770 | EXPORT_SYMBOL(ip_generic_getfrag); |
1da177e4 | 771 | |
44bb9363 | 772 | static inline __wsum |
1da177e4 LT |
773 | csum_page(struct page *page, int offset, int copy) |
774 | { | |
775 | char *kaddr; | |
44bb9363 | 776 | __wsum csum; |
1da177e4 LT |
777 | kaddr = kmap(page); |
778 | csum = csum_partial(kaddr + offset, copy, 0); | |
779 | kunmap(page); | |
780 | return csum; | |
781 | } | |
782 | ||
4b30b1c6 | 783 | static inline int ip_ufo_append_data(struct sock *sk, |
1470ddf7 | 784 | struct sk_buff_head *queue, |
e89e9cf5 AR |
785 | int getfrag(void *from, char *to, int offset, int len, |
786 | int odd, struct sk_buff *skb), | |
787 | void *from, int length, int hh_len, int fragheaderlen, | |
d9be4f7a | 788 | int transhdrlen, int maxfraglen, unsigned int flags) |
e89e9cf5 AR |
789 | { |
790 | struct sk_buff *skb; | |
791 | int err; | |
792 | ||
793 | /* There is support for UDP fragmentation offload by network | |
794 | * device, so create one single skb packet containing complete | |
795 | * udp datagram | |
796 | */ | |
1470ddf7 | 797 | if ((skb = skb_peek_tail(queue)) == NULL) { |
e89e9cf5 AR |
798 | skb = sock_alloc_send_skb(sk, |
799 | hh_len + fragheaderlen + transhdrlen + 20, | |
800 | (flags & MSG_DONTWAIT), &err); | |
801 | ||
802 | if (skb == NULL) | |
803 | return err; | |
804 | ||
805 | /* reserve space for Hardware header */ | |
806 | skb_reserve(skb, hh_len); | |
807 | ||
808 | /* create space for UDP/IP header */ | |
d9319100 | 809 | skb_put(skb, fragheaderlen + transhdrlen); |
e89e9cf5 AR |
810 | |
811 | /* initialize network header pointer */ | |
c1d2bbe1 | 812 | skb_reset_network_header(skb); |
e89e9cf5 AR |
813 | |
814 | /* initialize protocol header pointer */ | |
b0e380b1 | 815 | skb->transport_header = skb->network_header + fragheaderlen; |
e89e9cf5 | 816 | |
e89e9cf5 | 817 | skb->csum = 0; |
e89e9cf5 | 818 | |
e93b7d74 | 819 | |
1470ddf7 | 820 | __skb_queue_tail(queue, skb); |
e93b7d74 JP |
821 | } else if (skb_is_gso(skb)) { |
822 | goto append; | |
e89e9cf5 | 823 | } |
be9164e7 | 824 | |
e93b7d74 JP |
825 | skb->ip_summed = CHECKSUM_PARTIAL; |
826 | /* specify the length of each IP datagram fragment */ | |
827 | skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; | |
828 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
829 | ||
830 | append: | |
be9164e7 K |
831 | return skb_append_datato_frags(sk, skb, getfrag, from, |
832 | (length - transhdrlen)); | |
e89e9cf5 AR |
833 | } |
834 | ||
f5fca608 DM |
835 | static int __ip_append_data(struct sock *sk, |
836 | struct flowi4 *fl4, | |
837 | struct sk_buff_head *queue, | |
1470ddf7 | 838 | struct inet_cork *cork, |
5640f768 | 839 | struct page_frag *pfrag, |
1470ddf7 HX |
840 | int getfrag(void *from, char *to, int offset, |
841 | int len, int odd, struct sk_buff *skb), | |
842 | void *from, int length, int transhdrlen, | |
843 | unsigned int flags) | |
1da177e4 LT |
844 | { |
845 | struct inet_sock *inet = inet_sk(sk); | |
846 | struct sk_buff *skb; | |
847 | ||
07df5294 | 848 | struct ip_options *opt = cork->opt; |
1da177e4 LT |
849 | int hh_len; |
850 | int exthdrlen; | |
851 | int mtu; | |
852 | int copy; | |
853 | int err; | |
854 | int offset = 0; | |
daba287b | 855 | unsigned int maxfraglen, fragheaderlen, maxnonfragsize; |
1da177e4 | 856 | int csummode = CHECKSUM_NONE; |
1470ddf7 | 857 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 858 | |
96d7303e SK |
859 | skb = skb_peek_tail(queue); |
860 | ||
861 | exthdrlen = !skb ? rt->dst.header_len : 0; | |
07df5294 | 862 | mtu = cork->fragsize; |
1da177e4 | 863 | |
d8d1f30b | 864 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
865 | |
866 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
867 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 868 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 869 | |
daba287b | 870 | if (cork->length + length > maxnonfragsize - fragheaderlen) { |
f5fca608 | 871 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
61e7f09d | 872 | mtu - (opt ? opt->optlen : 0)); |
1da177e4 LT |
873 | return -EMSGSIZE; |
874 | } | |
875 | ||
876 | /* | |
877 | * transhdrlen > 0 means that this is the first fragment and we wish | |
878 | * it won't be fragmented in the future. | |
879 | */ | |
880 | if (transhdrlen && | |
881 | length + fragheaderlen <= mtu && | |
d8d1f30b | 882 | rt->dst.dev->features & NETIF_F_V4_CSUM && |
1da177e4 | 883 | !exthdrlen) |
84fa7933 | 884 | csummode = CHECKSUM_PARTIAL; |
1da177e4 | 885 | |
1470ddf7 | 886 | cork->length += length; |
26cde9f7 | 887 | if (((length > mtu) || (skb && skb_is_gso(skb))) && |
be9164e7 | 888 | (sk->sk_protocol == IPPROTO_UDP) && |
c146066a | 889 | (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) { |
1470ddf7 HX |
890 | err = ip_ufo_append_data(sk, queue, getfrag, from, length, |
891 | hh_len, fragheaderlen, transhdrlen, | |
d9be4f7a | 892 | maxfraglen, flags); |
baa829d8 | 893 | if (err) |
e89e9cf5 | 894 | goto error; |
e89e9cf5 AR |
895 | return 0; |
896 | } | |
1da177e4 LT |
897 | |
898 | /* So, what's going on in the loop below? | |
899 | * | |
900 | * We use calculated fragment length to generate chained skb, | |
901 | * each of segments is IP fragment ready for sending to network after | |
902 | * adding appropriate IP header. | |
903 | */ | |
904 | ||
26cde9f7 | 905 | if (!skb) |
1da177e4 LT |
906 | goto alloc_new_skb; |
907 | ||
908 | while (length > 0) { | |
909 | /* Check if the remaining data fits into current packet. */ | |
910 | copy = mtu - skb->len; | |
911 | if (copy < length) | |
912 | copy = maxfraglen - skb->len; | |
913 | if (copy <= 0) { | |
914 | char *data; | |
915 | unsigned int datalen; | |
916 | unsigned int fraglen; | |
917 | unsigned int fraggap; | |
918 | unsigned int alloclen; | |
919 | struct sk_buff *skb_prev; | |
920 | alloc_new_skb: | |
921 | skb_prev = skb; | |
922 | if (skb_prev) | |
923 | fraggap = skb_prev->len - maxfraglen; | |
924 | else | |
925 | fraggap = 0; | |
926 | ||
927 | /* | |
928 | * If remaining data exceeds the mtu, | |
929 | * we know we need more fragment(s). | |
930 | */ | |
931 | datalen = length + fraggap; | |
932 | if (datalen > mtu - fragheaderlen) | |
933 | datalen = maxfraglen - fragheaderlen; | |
934 | fraglen = datalen + fragheaderlen; | |
935 | ||
e905a9ed | 936 | if ((flags & MSG_MORE) && |
d8d1f30b | 937 | !(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
938 | alloclen = mtu; |
939 | else | |
59104f06 | 940 | alloclen = fraglen; |
1da177e4 | 941 | |
353e5c9a SK |
942 | alloclen += exthdrlen; |
943 | ||
1da177e4 LT |
944 | /* The last fragment gets additional space at tail. |
945 | * Note, with MSG_MORE we overallocate on fragments, | |
946 | * because we have no idea what fragment will be | |
947 | * the last. | |
948 | */ | |
33f99dc7 | 949 | if (datalen == length + fraggap) |
d8d1f30b | 950 | alloclen += rt->dst.trailer_len; |
33f99dc7 | 951 | |
1da177e4 | 952 | if (transhdrlen) { |
e905a9ed | 953 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
954 | alloclen + hh_len + 15, |
955 | (flags & MSG_DONTWAIT), &err); | |
956 | } else { | |
957 | skb = NULL; | |
958 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
959 | 2 * sk->sk_sndbuf) | |
e905a9ed | 960 | skb = sock_wmalloc(sk, |
1da177e4 LT |
961 | alloclen + hh_len + 15, 1, |
962 | sk->sk_allocation); | |
963 | if (unlikely(skb == NULL)) | |
964 | err = -ENOBUFS; | |
51f31cab PO |
965 | else |
966 | /* only the initial fragment is | |
967 | time stamped */ | |
1470ddf7 | 968 | cork->tx_flags = 0; |
1da177e4 LT |
969 | } |
970 | if (skb == NULL) | |
971 | goto error; | |
972 | ||
973 | /* | |
974 | * Fill in the control structures | |
975 | */ | |
976 | skb->ip_summed = csummode; | |
977 | skb->csum = 0; | |
978 | skb_reserve(skb, hh_len); | |
1470ddf7 | 979 | skb_shinfo(skb)->tx_flags = cork->tx_flags; |
1da177e4 LT |
980 | |
981 | /* | |
982 | * Find where to start putting bytes. | |
983 | */ | |
353e5c9a | 984 | data = skb_put(skb, fraglen + exthdrlen); |
c14d2450 | 985 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
986 | skb->transport_header = (skb->network_header + |
987 | fragheaderlen); | |
353e5c9a | 988 | data += fragheaderlen + exthdrlen; |
1da177e4 LT |
989 | |
990 | if (fraggap) { | |
991 | skb->csum = skb_copy_and_csum_bits( | |
992 | skb_prev, maxfraglen, | |
993 | data + transhdrlen, fraggap, 0); | |
994 | skb_prev->csum = csum_sub(skb_prev->csum, | |
995 | skb->csum); | |
996 | data += fraggap; | |
e9fa4f7b | 997 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
998 | } |
999 | ||
1000 | copy = datalen - transhdrlen - fraggap; | |
1001 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
1002 | err = -EFAULT; | |
1003 | kfree_skb(skb); | |
1004 | goto error; | |
1005 | } | |
1006 | ||
1007 | offset += copy; | |
1008 | length -= datalen - fraggap; | |
1009 | transhdrlen = 0; | |
1010 | exthdrlen = 0; | |
1011 | csummode = CHECKSUM_NONE; | |
1012 | ||
1013 | /* | |
1014 | * Put the packet on the pending queue. | |
1015 | */ | |
1470ddf7 | 1016 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
1017 | continue; |
1018 | } | |
1019 | ||
1020 | if (copy > length) | |
1021 | copy = length; | |
1022 | ||
d8d1f30b | 1023 | if (!(rt->dst.dev->features&NETIF_F_SG)) { |
1da177e4 LT |
1024 | unsigned int off; |
1025 | ||
1026 | off = skb->len; | |
e905a9ed | 1027 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
1028 | offset, copy, off, skb) < 0) { |
1029 | __skb_trim(skb, off); | |
1030 | err = -EFAULT; | |
1031 | goto error; | |
1032 | } | |
1033 | } else { | |
1034 | int i = skb_shinfo(skb)->nr_frags; | |
1da177e4 | 1035 | |
5640f768 ED |
1036 | err = -ENOMEM; |
1037 | if (!sk_page_frag_refill(sk, pfrag)) | |
1da177e4 | 1038 | goto error; |
5640f768 ED |
1039 | |
1040 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1041 | pfrag->offset)) { | |
1042 | err = -EMSGSIZE; | |
1043 | if (i == MAX_SKB_FRAGS) | |
1044 | goto error; | |
1045 | ||
1046 | __skb_fill_page_desc(skb, i, pfrag->page, | |
1047 | pfrag->offset, 0); | |
1048 | skb_shinfo(skb)->nr_frags = ++i; | |
1049 | get_page(pfrag->page); | |
1da177e4 | 1050 | } |
5640f768 ED |
1051 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1052 | if (getfrag(from, | |
1053 | page_address(pfrag->page) + pfrag->offset, | |
1054 | offset, copy, skb->len, skb) < 0) | |
1055 | goto error_efault; | |
1056 | ||
1057 | pfrag->offset += copy; | |
1058 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1da177e4 LT |
1059 | skb->len += copy; |
1060 | skb->data_len += copy; | |
f945fa7a HX |
1061 | skb->truesize += copy; |
1062 | atomic_add(copy, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1063 | } |
1064 | offset += copy; | |
1065 | length -= copy; | |
1066 | } | |
1067 | ||
1068 | return 0; | |
1069 | ||
5640f768 ED |
1070 | error_efault: |
1071 | err = -EFAULT; | |
1da177e4 | 1072 | error: |
1470ddf7 | 1073 | cork->length -= length; |
5e38e270 | 1074 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
e905a9ed | 1075 | return err; |
1da177e4 LT |
1076 | } |
1077 | ||
1470ddf7 HX |
1078 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, |
1079 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1080 | { | |
f6d8bd05 | 1081 | struct ip_options_rcu *opt; |
1470ddf7 HX |
1082 | struct rtable *rt; |
1083 | ||
1084 | /* | |
1085 | * setup for corking. | |
1086 | */ | |
1087 | opt = ipc->opt; | |
1088 | if (opt) { | |
1089 | if (cork->opt == NULL) { | |
1090 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, | |
1091 | sk->sk_allocation); | |
1092 | if (unlikely(cork->opt == NULL)) | |
1093 | return -ENOBUFS; | |
1094 | } | |
f6d8bd05 | 1095 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); |
1470ddf7 HX |
1096 | cork->flags |= IPCORK_OPT; |
1097 | cork->addr = ipc->addr; | |
1098 | } | |
1099 | rt = *rtp; | |
1100 | if (unlikely(!rt)) | |
1101 | return -EFAULT; | |
1102 | /* | |
1103 | * We steal reference to this route, caller should not release it | |
1104 | */ | |
1105 | *rtp = NULL; | |
482fc609 HFS |
1106 | cork->fragsize = ip_sk_use_pmtu(sk) ? |
1107 | dst_mtu(&rt->dst) : rt->dst.dev->mtu; | |
1470ddf7 HX |
1108 | cork->dst = &rt->dst; |
1109 | cork->length = 0; | |
aa661581 FF |
1110 | cork->ttl = ipc->ttl; |
1111 | cork->tos = ipc->tos; | |
1112 | cork->priority = ipc->priority; | |
1470ddf7 | 1113 | cork->tx_flags = ipc->tx_flags; |
1470ddf7 HX |
1114 | |
1115 | return 0; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1120 | * from many pieces of data. Each pieces will be holded on the socket | |
1121 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1122 | * or non-page data. | |
1123 | * | |
1124 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1125 | * this interface potentially. | |
1126 | * | |
1127 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1128 | */ | |
f5fca608 | 1129 | int ip_append_data(struct sock *sk, struct flowi4 *fl4, |
1470ddf7 HX |
1130 | int getfrag(void *from, char *to, int offset, int len, |
1131 | int odd, struct sk_buff *skb), | |
1132 | void *from, int length, int transhdrlen, | |
1133 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1134 | unsigned int flags) | |
1135 | { | |
1136 | struct inet_sock *inet = inet_sk(sk); | |
1137 | int err; | |
1138 | ||
1139 | if (flags&MSG_PROBE) | |
1140 | return 0; | |
1141 | ||
1142 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
bdc712b4 | 1143 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); |
1470ddf7 HX |
1144 | if (err) |
1145 | return err; | |
1146 | } else { | |
1147 | transhdrlen = 0; | |
1148 | } | |
1149 | ||
5640f768 ED |
1150 | return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, |
1151 | sk_page_frag(sk), getfrag, | |
1470ddf7 HX |
1152 | from, length, transhdrlen, flags); |
1153 | } | |
1154 | ||
f5fca608 | 1155 | ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, |
1da177e4 LT |
1156 | int offset, size_t size, int flags) |
1157 | { | |
1158 | struct inet_sock *inet = inet_sk(sk); | |
1159 | struct sk_buff *skb; | |
1160 | struct rtable *rt; | |
1161 | struct ip_options *opt = NULL; | |
bdc712b4 | 1162 | struct inet_cork *cork; |
1da177e4 LT |
1163 | int hh_len; |
1164 | int mtu; | |
1165 | int len; | |
1166 | int err; | |
daba287b | 1167 | unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; |
1da177e4 LT |
1168 | |
1169 | if (inet->hdrincl) | |
1170 | return -EPERM; | |
1171 | ||
1172 | if (flags&MSG_PROBE) | |
1173 | return 0; | |
1174 | ||
1175 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1176 | return -EINVAL; | |
1177 | ||
bdc712b4 DM |
1178 | cork = &inet->cork.base; |
1179 | rt = (struct rtable *)cork->dst; | |
1180 | if (cork->flags & IPCORK_OPT) | |
1181 | opt = cork->opt; | |
1da177e4 | 1182 | |
d8d1f30b | 1183 | if (!(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
1184 | return -EOPNOTSUPP; |
1185 | ||
d8d1f30b | 1186 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
bdc712b4 | 1187 | mtu = cork->fragsize; |
1da177e4 LT |
1188 | |
1189 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1190 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 1191 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 1192 | |
daba287b | 1193 | if (cork->length + size > maxnonfragsize - fragheaderlen) { |
61e7f09d HFS |
1194 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
1195 | mtu - (opt ? opt->optlen : 0)); | |
1da177e4 LT |
1196 | return -EMSGSIZE; |
1197 | } | |
1198 | ||
1199 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1200 | return -EINVAL; | |
1201 | ||
bdc712b4 | 1202 | cork->length += size; |
26cde9f7 HX |
1203 | if ((size + skb->len > mtu) && |
1204 | (sk->sk_protocol == IPPROTO_UDP) && | |
d8d1f30b | 1205 | (rt->dst.dev->features & NETIF_F_UFO)) { |
7967168c | 1206 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 1207 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1208 | } |
e89e9cf5 | 1209 | |
1da177e4 LT |
1210 | |
1211 | while (size > 0) { | |
1212 | int i; | |
1213 | ||
89114afd | 1214 | if (skb_is_gso(skb)) |
e89e9cf5 AR |
1215 | len = size; |
1216 | else { | |
1217 | ||
1218 | /* Check if the remaining data fits into current packet. */ | |
1219 | len = mtu - skb->len; | |
1220 | if (len < size) | |
1221 | len = maxfraglen - skb->len; | |
1222 | } | |
1da177e4 LT |
1223 | if (len <= 0) { |
1224 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1225 | int alloclen; |
1226 | ||
1227 | skb_prev = skb; | |
0d0d2bba | 1228 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1229 | |
1230 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1231 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1232 | if (unlikely(!skb)) { | |
1233 | err = -ENOBUFS; | |
1234 | goto error; | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Fill in the control structures | |
1239 | */ | |
1240 | skb->ip_summed = CHECKSUM_NONE; | |
1241 | skb->csum = 0; | |
1242 | skb_reserve(skb, hh_len); | |
1243 | ||
1244 | /* | |
1245 | * Find where to start putting bytes. | |
1246 | */ | |
967b05f6 | 1247 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1248 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1249 | skb->transport_header = (skb->network_header + |
1250 | fragheaderlen); | |
1da177e4 | 1251 | if (fraggap) { |
967b05f6 ACM |
1252 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1253 | maxfraglen, | |
9c70220b | 1254 | skb_transport_header(skb), |
967b05f6 | 1255 | fraggap, 0); |
1da177e4 LT |
1256 | skb_prev->csum = csum_sub(skb_prev->csum, |
1257 | skb->csum); | |
e9fa4f7b | 1258 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1259 | } |
1260 | ||
1261 | /* | |
1262 | * Put the packet on the pending queue. | |
1263 | */ | |
1264 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1265 | continue; | |
1266 | } | |
1267 | ||
1268 | i = skb_shinfo(skb)->nr_frags; | |
1269 | if (len > size) | |
1270 | len = size; | |
1271 | if (skb_can_coalesce(skb, i, page, offset)) { | |
9e903e08 | 1272 | skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len); |
1da177e4 LT |
1273 | } else if (i < MAX_SKB_FRAGS) { |
1274 | get_page(page); | |
1275 | skb_fill_page_desc(skb, i, page, offset, len); | |
1276 | } else { | |
1277 | err = -EMSGSIZE; | |
1278 | goto error; | |
1279 | } | |
1280 | ||
1281 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1282 | __wsum csum; |
1da177e4 LT |
1283 | csum = csum_page(page, offset, len); |
1284 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1285 | } | |
1286 | ||
1287 | skb->len += len; | |
1288 | skb->data_len += len; | |
1e34a11d DM |
1289 | skb->truesize += len; |
1290 | atomic_add(len, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1291 | offset += len; |
1292 | size -= len; | |
1293 | } | |
1294 | return 0; | |
1295 | ||
1296 | error: | |
bdc712b4 | 1297 | cork->length -= size; |
5e38e270 | 1298 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1299 | return err; |
1300 | } | |
1301 | ||
1470ddf7 | 1302 | static void ip_cork_release(struct inet_cork *cork) |
429f08e9 | 1303 | { |
1470ddf7 HX |
1304 | cork->flags &= ~IPCORK_OPT; |
1305 | kfree(cork->opt); | |
1306 | cork->opt = NULL; | |
1307 | dst_release(cork->dst); | |
1308 | cork->dst = NULL; | |
429f08e9 PE |
1309 | } |
1310 | ||
1da177e4 LT |
1311 | /* |
1312 | * Combined all pending IP fragments on the socket as one IP datagram | |
1313 | * and push them out. | |
1314 | */ | |
1c32c5ad | 1315 | struct sk_buff *__ip_make_skb(struct sock *sk, |
77968b78 | 1316 | struct flowi4 *fl4, |
1c32c5ad HX |
1317 | struct sk_buff_head *queue, |
1318 | struct inet_cork *cork) | |
1da177e4 LT |
1319 | { |
1320 | struct sk_buff *skb, *tmp_skb; | |
1321 | struct sk_buff **tail_skb; | |
1322 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1323 | struct net *net = sock_net(sk); |
1da177e4 | 1324 | struct ip_options *opt = NULL; |
1470ddf7 | 1325 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 1326 | struct iphdr *iph; |
76ab608d | 1327 | __be16 df = 0; |
1da177e4 | 1328 | __u8 ttl; |
1da177e4 | 1329 | |
1470ddf7 | 1330 | if ((skb = __skb_dequeue(queue)) == NULL) |
1da177e4 LT |
1331 | goto out; |
1332 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1333 | ||
1334 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1335 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1336 | __skb_pull(skb, skb_network_offset(skb)); |
1470ddf7 | 1337 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { |
cfe1fc77 | 1338 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1339 | *tail_skb = tmp_skb; |
1340 | tail_skb = &(tmp_skb->next); | |
1341 | skb->len += tmp_skb->len; | |
1342 | skb->data_len += tmp_skb->len; | |
1343 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1344 | tmp_skb->destructor = NULL; |
1345 | tmp_skb->sk = NULL; | |
1346 | } | |
1347 | ||
1348 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1349 | * to fragment the frame generated here. No matter, what transforms | |
1350 | * how transforms change size of the packet, it will come out. | |
1351 | */ | |
60ff7467 | 1352 | skb->ignore_df = ip_sk_ignore_df(sk); |
1da177e4 LT |
1353 | |
1354 | /* DF bit is set when we want to see DF on outgoing frames. | |
60ff7467 | 1355 | * If ignore_df is set too, we still allow to fragment this frame |
1da177e4 | 1356 | * locally. */ |
482fc609 HFS |
1357 | if (inet->pmtudisc == IP_PMTUDISC_DO || |
1358 | inet->pmtudisc == IP_PMTUDISC_PROBE || | |
d8d1f30b CG |
1359 | (skb->len <= dst_mtu(&rt->dst) && |
1360 | ip_dont_fragment(sk, &rt->dst))) | |
1da177e4 LT |
1361 | df = htons(IP_DF); |
1362 | ||
1470ddf7 HX |
1363 | if (cork->flags & IPCORK_OPT) |
1364 | opt = cork->opt; | |
1da177e4 | 1365 | |
aa661581 FF |
1366 | if (cork->ttl != 0) |
1367 | ttl = cork->ttl; | |
1368 | else if (rt->rt_type == RTN_MULTICAST) | |
1da177e4 LT |
1369 | ttl = inet->mc_ttl; |
1370 | else | |
d8d1f30b | 1371 | ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 1372 | |
749154aa | 1373 | iph = ip_hdr(skb); |
1da177e4 LT |
1374 | iph->version = 4; |
1375 | iph->ihl = 5; | |
aa661581 | 1376 | iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; |
1da177e4 | 1377 | iph->frag_off = df; |
1da177e4 LT |
1378 | iph->ttl = ttl; |
1379 | iph->protocol = sk->sk_protocol; | |
84f9307c | 1380 | ip_copy_addrs(iph, fl4); |
73f156a6 | 1381 | ip_select_ident(skb, sk); |
1da177e4 | 1382 | |
22f728f8 DM |
1383 | if (opt) { |
1384 | iph->ihl += opt->optlen>>2; | |
1385 | ip_options_build(skb, opt, cork->addr, rt, 0); | |
1386 | } | |
1387 | ||
aa661581 | 1388 | skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; |
4a19ec58 | 1389 | skb->mark = sk->sk_mark; |
a21bba94 ED |
1390 | /* |
1391 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1392 | * on dst refcount | |
1393 | */ | |
1470ddf7 | 1394 | cork->dst = NULL; |
d8d1f30b | 1395 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 1396 | |
96793b48 | 1397 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1398 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1399 | skb_transport_header(skb))->type); |
1400 | ||
1c32c5ad HX |
1401 | ip_cork_release(cork); |
1402 | out: | |
1403 | return skb; | |
1404 | } | |
1405 | ||
b5ec8eea | 1406 | int ip_send_skb(struct net *net, struct sk_buff *skb) |
1c32c5ad | 1407 | { |
1c32c5ad HX |
1408 | int err; |
1409 | ||
c439cb2e | 1410 | err = ip_local_out(skb); |
1da177e4 LT |
1411 | if (err) { |
1412 | if (err > 0) | |
6ce9e7b5 | 1413 | err = net_xmit_errno(err); |
1da177e4 | 1414 | if (err) |
1c32c5ad | 1415 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1416 | } |
1417 | ||
1da177e4 | 1418 | return err; |
1da177e4 LT |
1419 | } |
1420 | ||
77968b78 | 1421 | int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) |
1470ddf7 | 1422 | { |
1c32c5ad HX |
1423 | struct sk_buff *skb; |
1424 | ||
77968b78 | 1425 | skb = ip_finish_skb(sk, fl4); |
1c32c5ad HX |
1426 | if (!skb) |
1427 | return 0; | |
1428 | ||
1429 | /* Netfilter gets whole the not fragmented skb. */ | |
b5ec8eea | 1430 | return ip_send_skb(sock_net(sk), skb); |
1470ddf7 HX |
1431 | } |
1432 | ||
1da177e4 LT |
1433 | /* |
1434 | * Throw away all pending data on the socket. | |
1435 | */ | |
1470ddf7 HX |
1436 | static void __ip_flush_pending_frames(struct sock *sk, |
1437 | struct sk_buff_head *queue, | |
1438 | struct inet_cork *cork) | |
1da177e4 | 1439 | { |
1da177e4 LT |
1440 | struct sk_buff *skb; |
1441 | ||
1470ddf7 | 1442 | while ((skb = __skb_dequeue_tail(queue)) != NULL) |
1da177e4 LT |
1443 | kfree_skb(skb); |
1444 | ||
1470ddf7 HX |
1445 | ip_cork_release(cork); |
1446 | } | |
1447 | ||
1448 | void ip_flush_pending_frames(struct sock *sk) | |
1449 | { | |
bdc712b4 | 1450 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); |
1da177e4 LT |
1451 | } |
1452 | ||
1c32c5ad | 1453 | struct sk_buff *ip_make_skb(struct sock *sk, |
77968b78 | 1454 | struct flowi4 *fl4, |
1c32c5ad HX |
1455 | int getfrag(void *from, char *to, int offset, |
1456 | int len, int odd, struct sk_buff *skb), | |
1457 | void *from, int length, int transhdrlen, | |
1458 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1459 | unsigned int flags) | |
1460 | { | |
b80d7226 | 1461 | struct inet_cork cork; |
1c32c5ad HX |
1462 | struct sk_buff_head queue; |
1463 | int err; | |
1464 | ||
1465 | if (flags & MSG_PROBE) | |
1466 | return NULL; | |
1467 | ||
1468 | __skb_queue_head_init(&queue); | |
1469 | ||
b80d7226 DM |
1470 | cork.flags = 0; |
1471 | cork.addr = 0; | |
70652728 | 1472 | cork.opt = NULL; |
1c32c5ad HX |
1473 | err = ip_setup_cork(sk, &cork, ipc, rtp); |
1474 | if (err) | |
1475 | return ERR_PTR(err); | |
1476 | ||
5640f768 ED |
1477 | err = __ip_append_data(sk, fl4, &queue, &cork, |
1478 | ¤t->task_frag, getfrag, | |
1c32c5ad HX |
1479 | from, length, transhdrlen, flags); |
1480 | if (err) { | |
1481 | __ip_flush_pending_frames(sk, &queue, &cork); | |
1482 | return ERR_PTR(err); | |
1483 | } | |
1484 | ||
77968b78 | 1485 | return __ip_make_skb(sk, fl4, &queue, &cork); |
1c32c5ad | 1486 | } |
1da177e4 LT |
1487 | |
1488 | /* | |
1489 | * Fetch data from kernel space and fill in checksum if needed. | |
1490 | */ | |
e905a9ed | 1491 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1492 | int len, int odd, struct sk_buff *skb) |
1493 | { | |
5084205f | 1494 | __wsum csum; |
1da177e4 LT |
1495 | |
1496 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1497 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1498 | return 0; |
1da177e4 LT |
1499 | } |
1500 | ||
e905a9ed | 1501 | /* |
1da177e4 | 1502 | * Generic function to send a packet as reply to another packet. |
be9f4a44 | 1503 | * Used to send some TCP resets/acks so far. |
1da177e4 | 1504 | * |
be9f4a44 | 1505 | * Use a fake percpu inet socket to avoid false sharing and contention. |
1da177e4 | 1506 | */ |
be9f4a44 ED |
1507 | static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = { |
1508 | .sk = { | |
1509 | .__sk_common = { | |
1510 | .skc_refcnt = ATOMIC_INIT(1), | |
1511 | }, | |
1512 | .sk_wmem_alloc = ATOMIC_INIT(1), | |
1513 | .sk_allocation = GFP_ATOMIC, | |
1514 | .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE), | |
1515 | }, | |
0980e56e ED |
1516 | .pmtudisc = IP_PMTUDISC_WANT, |
1517 | .uc_ttl = -1, | |
be9f4a44 ED |
1518 | }; |
1519 | ||
1520 | void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr, | |
70e73416 DM |
1521 | __be32 saddr, const struct ip_reply_arg *arg, |
1522 | unsigned int len) | |
1da177e4 | 1523 | { |
f6d8bd05 | 1524 | struct ip_options_data replyopts; |
1da177e4 | 1525 | struct ipcm_cookie ipc; |
77968b78 | 1526 | struct flowi4 fl4; |
511c3f92 | 1527 | struct rtable *rt = skb_rtable(skb); |
be9f4a44 ED |
1528 | struct sk_buff *nskb; |
1529 | struct sock *sk; | |
1530 | struct inet_sock *inet; | |
1da177e4 | 1531 | |
f6d8bd05 | 1532 | if (ip_options_echo(&replyopts.opt.opt, skb)) |
1da177e4 LT |
1533 | return; |
1534 | ||
0a5ebb80 | 1535 | ipc.addr = daddr; |
1da177e4 | 1536 | ipc.opt = NULL; |
2244d07b | 1537 | ipc.tx_flags = 0; |
aa661581 FF |
1538 | ipc.ttl = 0; |
1539 | ipc.tos = -1; | |
1da177e4 | 1540 | |
f6d8bd05 | 1541 | if (replyopts.opt.opt.optlen) { |
1da177e4 LT |
1542 | ipc.opt = &replyopts.opt; |
1543 | ||
f6d8bd05 ED |
1544 | if (replyopts.opt.opt.srr) |
1545 | daddr = replyopts.opt.opt.faddr; | |
1da177e4 LT |
1546 | } |
1547 | ||
e110861f LC |
1548 | flowi4_init_output(&fl4, arg->bound_dev_if, |
1549 | IP4_REPLY_MARK(net, skb->mark), | |
66b13d99 | 1550 | RT_TOS(arg->tos), |
be9f4a44 | 1551 | RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, |
77968b78 | 1552 | ip_reply_arg_flowi_flags(arg), |
70e73416 | 1553 | daddr, saddr, |
77968b78 DM |
1554 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest); |
1555 | security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); | |
be9f4a44 | 1556 | rt = ip_route_output_key(net, &fl4); |
77968b78 DM |
1557 | if (IS_ERR(rt)) |
1558 | return; | |
1da177e4 | 1559 | |
be9f4a44 | 1560 | inet = &get_cpu_var(unicast_sock); |
1da177e4 | 1561 | |
66b13d99 | 1562 | inet->tos = arg->tos; |
be9f4a44 | 1563 | sk = &inet->sk; |
1da177e4 | 1564 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1565 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1566 | sk->sk_bound_dev_if = arg->bound_dev_if; |
be9f4a44 ED |
1567 | sock_net_set(sk, net); |
1568 | __skb_queue_head_init(&sk->sk_write_queue); | |
1569 | sk->sk_sndbuf = sysctl_wmem_default; | |
f5fca608 | 1570 | ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0, |
2e77d89b | 1571 | &ipc, &rt, MSG_DONTWAIT); |
be9f4a44 ED |
1572 | nskb = skb_peek(&sk->sk_write_queue); |
1573 | if (nskb) { | |
1da177e4 | 1574 | if (arg->csumoffset >= 0) |
be9f4a44 ED |
1575 | *((__sum16 *)skb_transport_header(nskb) + |
1576 | arg->csumoffset) = csum_fold(csum_add(nskb->csum, | |
9c70220b | 1577 | arg->csum)); |
be9f4a44 | 1578 | nskb->ip_summed = CHECKSUM_NONE; |
3a7c384f | 1579 | skb_orphan(nskb); |
be9f4a44 | 1580 | skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb)); |
77968b78 | 1581 | ip_push_pending_frames(sk, &fl4); |
1da177e4 LT |
1582 | } |
1583 | ||
be9f4a44 | 1584 | put_cpu_var(unicast_sock); |
1da177e4 LT |
1585 | |
1586 | ip_rt_put(rt); | |
1587 | } | |
1588 | ||
1da177e4 LT |
1589 | void __init ip_init(void) |
1590 | { | |
1da177e4 LT |
1591 | ip_rt_init(); |
1592 | inet_initpeers(); | |
1593 | ||
72c1d3bd WC |
1594 | #if defined(CONFIG_IP_MULTICAST) |
1595 | igmp_mc_init(); | |
1da177e4 LT |
1596 | #endif |
1597 | } |