]>
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> | |
14972cbd | 76 | #include <net/lwtunnel.h> |
33b48679 | 77 | #include <linux/bpf-cgroup.h> |
1da177e4 LT |
78 | #include <linux/igmp.h> |
79 | #include <linux/netfilter_ipv4.h> | |
80 | #include <linux/netfilter_bridge.h> | |
1da177e4 | 81 | #include <linux/netlink.h> |
6cbb0df7 | 82 | #include <linux/tcp.h> |
1da177e4 | 83 | |
694869b3 EB |
84 | static int |
85 | ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, | |
86 | unsigned int mtu, | |
87 | int (*output)(struct net *, struct sock *, struct sk_buff *)); | |
49d16b23 | 88 | |
1da177e4 | 89 | /* Generate a checksum for an outgoing IP datagram. */ |
2fbd9679 | 90 | void ip_send_check(struct iphdr *iph) |
1da177e4 LT |
91 | { |
92 | iph->check = 0; | |
93 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
94 | } | |
4bc2f18b | 95 | EXPORT_SYMBOL(ip_send_check); |
1da177e4 | 96 | |
cf91a99d | 97 | int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
98 | { |
99 | struct iphdr *iph = ip_hdr(skb); | |
100 | ||
101 | iph->tot_len = htons(skb->len); | |
102 | ip_send_check(iph); | |
a8e3e1a9 DA |
103 | |
104 | /* if egress device is enslaved to an L3 master device pass the | |
105 | * skb to its handler for processing | |
106 | */ | |
107 | skb = l3mdev_ip_out(sk, skb); | |
108 | if (unlikely(!skb)) | |
109 | return 0; | |
110 | ||
f4180439 EC |
111 | skb->protocol = htons(ETH_P_IP); |
112 | ||
29a26a56 EB |
113 | return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, |
114 | net, sk, skb, NULL, skb_dst(skb)->dev, | |
13206b6b | 115 | dst_output); |
7026b1dd DM |
116 | } |
117 | ||
33224b16 | 118 | int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb) |
c439cb2e HX |
119 | { |
120 | int err; | |
121 | ||
cf91a99d | 122 | err = __ip_local_out(net, sk, skb); |
c439cb2e | 123 | if (likely(err == 1)) |
13206b6b | 124 | err = dst_output(net, sk, skb); |
c439cb2e HX |
125 | |
126 | return err; | |
127 | } | |
e2cb77db | 128 | EXPORT_SYMBOL_GPL(ip_local_out); |
c439cb2e | 129 | |
1da177e4 LT |
130 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) |
131 | { | |
132 | int ttl = inet->uc_ttl; | |
133 | ||
134 | if (ttl < 0) | |
323e126f | 135 | ttl = ip4_dst_hoplimit(dst); |
1da177e4 LT |
136 | return ttl; |
137 | } | |
138 | ||
e905a9ed | 139 | /* |
1da177e4 LT |
140 | * Add an ip header to a skbuff and send it out. |
141 | * | |
142 | */ | |
cfe673b0 | 143 | int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk, |
f6d8bd05 | 144 | __be32 saddr, __be32 daddr, struct ip_options_rcu *opt) |
1da177e4 LT |
145 | { |
146 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 147 | struct rtable *rt = skb_rtable(skb); |
77589ce0 | 148 | struct net *net = sock_net(sk); |
1da177e4 LT |
149 | struct iphdr *iph; |
150 | ||
151 | /* Build the IP header. */ | |
f6d8bd05 | 152 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0)); |
8856dfa3 | 153 | skb_reset_network_header(skb); |
eddc9ec5 | 154 | iph = ip_hdr(skb); |
1da177e4 LT |
155 | iph->version = 4; |
156 | iph->ihl = 5; | |
157 | iph->tos = inet->tos; | |
d8d1f30b | 158 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
dd927a26 DM |
159 | iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr); |
160 | iph->saddr = saddr; | |
1da177e4 | 161 | iph->protocol = sk->sk_protocol; |
cfe673b0 ED |
162 | if (ip_dont_fragment(sk, &rt->dst)) { |
163 | iph->frag_off = htons(IP_DF); | |
164 | iph->id = 0; | |
165 | } else { | |
166 | iph->frag_off = 0; | |
77589ce0 | 167 | __ip_select_ident(net, iph, 1); |
cfe673b0 | 168 | } |
1da177e4 | 169 | |
f6d8bd05 ED |
170 | if (opt && opt->opt.optlen) { |
171 | iph->ihl += opt->opt.optlen>>2; | |
172 | ip_options_build(skb, &opt->opt, daddr, rt, 0); | |
1da177e4 | 173 | } |
1da177e4 LT |
174 | |
175 | skb->priority = sk->sk_priority; | |
4a19ec58 | 176 | skb->mark = sk->sk_mark; |
1da177e4 LT |
177 | |
178 | /* Send it out. */ | |
33224b16 | 179 | return ip_local_out(net, skb->sk, skb); |
1da177e4 | 180 | } |
d8c97a94 ACM |
181 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
182 | ||
694869b3 | 183 | static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 184 | { |
adf30907 | 185 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 186 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 187 | struct net_device *dev = dst->dev; |
c2636b4d | 188 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
f6b72b62 | 189 | struct neighbour *neigh; |
a263b309 | 190 | u32 nexthop; |
1da177e4 | 191 | |
edf391ff | 192 | if (rt->rt_type == RTN_MULTICAST) { |
4ba1bf42 | 193 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len); |
edf391ff | 194 | } else if (rt->rt_type == RTN_BROADCAST) |
4ba1bf42 | 195 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len); |
80787ebc | 196 | |
1da177e4 | 197 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 198 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
199 | struct sk_buff *skb2; |
200 | ||
201 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
51456b29 | 202 | if (!skb2) { |
1da177e4 LT |
203 | kfree_skb(skb); |
204 | return -ENOMEM; | |
205 | } | |
206 | if (skb->sk) | |
207 | skb_set_owner_w(skb2, skb->sk); | |
5d0ba55b | 208 | consume_skb(skb); |
1da177e4 LT |
209 | skb = skb2; |
210 | } | |
211 | ||
14972cbd RP |
212 | if (lwtunnel_xmit_redirect(dst->lwtstate)) { |
213 | int res = lwtunnel_xmit(skb); | |
214 | ||
215 | if (res < 0 || res == LWTUNNEL_XMIT_DONE) | |
216 | return res; | |
217 | } | |
218 | ||
a263b309 | 219 | rcu_read_lock_bh(); |
155e8336 | 220 | nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr); |
a263b309 DM |
221 | neigh = __ipv4_neigh_lookup_noref(dev, nexthop); |
222 | if (unlikely(!neigh)) | |
223 | neigh = __neigh_create(&arp_tbl, &nexthop, dev, false); | |
9871f1ad | 224 | if (!IS_ERR(neigh)) { |
5110effe | 225 | int res = dst_neigh_output(dst, neigh, skb); |
f2c31e32 | 226 | |
a263b309 | 227 | rcu_read_unlock_bh(); |
f2c31e32 ED |
228 | return res; |
229 | } | |
a263b309 | 230 | rcu_read_unlock_bh(); |
05e3aa09 | 231 | |
e87cc472 JP |
232 | net_dbg_ratelimited("%s: No header cache and no neighbour!\n", |
233 | __func__); | |
1da177e4 LT |
234 | kfree_skb(skb); |
235 | return -EINVAL; | |
236 | } | |
237 | ||
694869b3 EB |
238 | static int ip_finish_output_gso(struct net *net, struct sock *sk, |
239 | struct sk_buff *skb, unsigned int mtu) | |
c7ba65d7 FW |
240 | { |
241 | netdev_features_t features; | |
242 | struct sk_buff *segs; | |
243 | int ret = 0; | |
244 | ||
9ee6c5dc | 245 | /* common case: seglen is <= mtu |
359ebda2 | 246 | */ |
9ee6c5dc | 247 | if (skb_gso_validate_mtu(skb, mtu)) |
694869b3 | 248 | return ip_finish_output2(net, sk, skb); |
c7ba65d7 | 249 | |
0ace81ec | 250 | /* Slowpath - GSO segment length exceeds the egress MTU. |
c7ba65d7 | 251 | * |
0ace81ec LR |
252 | * This can happen in several cases: |
253 | * - Forwarding of a TCP GRO skb, when DF flag is not set. | |
254 | * - Forwarding of an skb that arrived on a virtualization interface | |
255 | * (virtio-net/vhost/tap) with TSO/GSO size set by other network | |
256 | * stack. | |
257 | * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an | |
258 | * interface with a smaller MTU. | |
259 | * - Arriving GRO skb (or GSO skb in a virtualized environment) that is | |
260 | * bridged to a NETIF_F_TSO tunnel stacked over an interface with an | |
261 | * insufficent MTU. | |
c7ba65d7 FW |
262 | */ |
263 | features = netif_skb_features(skb); | |
9207f9d4 | 264 | BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET); |
c7ba65d7 | 265 | segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); |
330966e5 | 266 | if (IS_ERR_OR_NULL(segs)) { |
c7ba65d7 FW |
267 | kfree_skb(skb); |
268 | return -ENOMEM; | |
269 | } | |
270 | ||
271 | consume_skb(skb); | |
272 | ||
273 | do { | |
274 | struct sk_buff *nskb = segs->next; | |
275 | int err; | |
276 | ||
277 | segs->next = NULL; | |
694869b3 | 278 | err = ip_fragment(net, sk, segs, mtu, ip_finish_output2); |
c7ba65d7 FW |
279 | |
280 | if (err && ret == 0) | |
281 | ret = err; | |
282 | segs = nskb; | |
283 | } while (segs); | |
284 | ||
285 | return ret; | |
286 | } | |
287 | ||
0c4b51f0 | 288 | static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 289 | { |
c5501eb3 | 290 | unsigned int mtu; |
33b48679 DM |
291 | int ret; |
292 | ||
293 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
294 | if (ret) { | |
295 | kfree_skb(skb); | |
296 | return ret; | |
297 | } | |
c5501eb3 | 298 | |
5c901daa PM |
299 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
300 | /* Policy lookup after SNAT yielded a new policy */ | |
00db4124 | 301 | if (skb_dst(skb)->xfrm) { |
48d5cad8 | 302 | IPCB(skb)->flags |= IPSKB_REROUTED; |
13206b6b | 303 | return dst_output(net, sk, skb); |
48d5cad8 | 304 | } |
5c901daa | 305 | #endif |
fedbb6b4 | 306 | mtu = ip_skb_dst_mtu(sk, skb); |
c7ba65d7 | 307 | if (skb_is_gso(skb)) |
694869b3 | 308 | return ip_finish_output_gso(net, sk, skb, mtu); |
c7ba65d7 | 309 | |
d6b915e2 | 310 | if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU)) |
694869b3 | 311 | return ip_fragment(net, sk, skb, mtu, ip_finish_output2); |
c7ba65d7 | 312 | |
694869b3 | 313 | return ip_finish_output2(net, sk, skb); |
1da177e4 LT |
314 | } |
315 | ||
33b48679 DM |
316 | static int ip_mc_finish_output(struct net *net, struct sock *sk, |
317 | struct sk_buff *skb) | |
318 | { | |
319 | int ret; | |
320 | ||
321 | ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb); | |
322 | if (ret) { | |
323 | kfree_skb(skb); | |
324 | return ret; | |
325 | } | |
326 | ||
327 | return dev_loopback_xmit(net, sk, skb); | |
328 | } | |
329 | ||
ede2059d | 330 | int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 331 | { |
511c3f92 | 332 | struct rtable *rt = skb_rtable(skb); |
d8d1f30b | 333 | struct net_device *dev = rt->dst.dev; |
1da177e4 LT |
334 | |
335 | /* | |
336 | * If the indicated interface is up and running, send the packet. | |
337 | */ | |
88f5cc24 | 338 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
339 | |
340 | skb->dev = dev; | |
341 | skb->protocol = htons(ETH_P_IP); | |
342 | ||
343 | /* | |
344 | * Multicasts are looped back for other local users | |
345 | */ | |
346 | ||
347 | if (rt->rt_flags&RTCF_MULTICAST) { | |
7ad6848c | 348 | if (sk_mc_loop(sk) |
1da177e4 LT |
349 | #ifdef CONFIG_IP_MROUTE |
350 | /* Small optimization: do not loopback not local frames, | |
351 | which returned after forwarding; they will be dropped | |
352 | by ip_mr_input in any case. | |
353 | Note, that local frames are looped back to be delivered | |
354 | to local recipients. | |
355 | ||
356 | This check is duplicated in ip_mr_input at the moment. | |
357 | */ | |
9d4fb27d JP |
358 | && |
359 | ((rt->rt_flags & RTCF_LOCAL) || | |
360 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 361 | #endif |
9d4fb27d | 362 | ) { |
1da177e4 LT |
363 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
364 | if (newskb) | |
9bbc768a | 365 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
29a26a56 | 366 | net, sk, newskb, NULL, newskb->dev, |
33b48679 | 367 | ip_mc_finish_output); |
1da177e4 LT |
368 | } |
369 | ||
370 | /* Multicasts with ttl 0 must not go beyond the host */ | |
371 | ||
eddc9ec5 | 372 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
373 | kfree_skb(skb); |
374 | return 0; | |
375 | } | |
376 | } | |
377 | ||
378 | if (rt->rt_flags&RTCF_BROADCAST) { | |
379 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
380 | if (newskb) | |
29a26a56 EB |
381 | NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
382 | net, sk, newskb, NULL, newskb->dev, | |
33b48679 | 383 | ip_mc_finish_output); |
1da177e4 LT |
384 | } |
385 | ||
29a26a56 EB |
386 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
387 | net, sk, skb, NULL, skb->dev, | |
388 | ip_finish_output, | |
48d5cad8 | 389 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
390 | } |
391 | ||
ede2059d | 392 | int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) |
1da177e4 | 393 | { |
adf30907 | 394 | struct net_device *dev = skb_dst(skb)->dev; |
1bd9bef6 | 395 | |
88f5cc24 | 396 | IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 397 | |
1bd9bef6 PM |
398 | skb->dev = dev; |
399 | skb->protocol = htons(ETH_P_IP); | |
400 | ||
29a26a56 EB |
401 | return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, |
402 | net, sk, skb, NULL, dev, | |
e905a9ed | 403 | ip_finish_output, |
48d5cad8 | 404 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
405 | } |
406 | ||
84f9307c ED |
407 | /* |
408 | * copy saddr and daddr, possibly using 64bit load/stores | |
409 | * Equivalent to : | |
410 | * iph->saddr = fl4->saddr; | |
411 | * iph->daddr = fl4->daddr; | |
412 | */ | |
413 | static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4) | |
414 | { | |
415 | BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) != | |
416 | offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr)); | |
417 | memcpy(&iph->saddr, &fl4->saddr, | |
418 | sizeof(fl4->saddr) + sizeof(fl4->daddr)); | |
419 | } | |
420 | ||
b0270e91 ED |
421 | /* Note: skb->sk can be different from sk, in case of tunnels */ |
422 | int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl) | |
1da177e4 | 423 | { |
1da177e4 | 424 | struct inet_sock *inet = inet_sk(sk); |
77589ce0 | 425 | struct net *net = sock_net(sk); |
f6d8bd05 | 426 | struct ip_options_rcu *inet_opt; |
b57ae01a | 427 | struct flowi4 *fl4; |
1da177e4 LT |
428 | struct rtable *rt; |
429 | struct iphdr *iph; | |
ab6e3feb | 430 | int res; |
1da177e4 LT |
431 | |
432 | /* Skip all of this if the packet is already routed, | |
433 | * f.e. by something like SCTP. | |
434 | */ | |
ab6e3feb | 435 | rcu_read_lock(); |
f6d8bd05 | 436 | inet_opt = rcu_dereference(inet->inet_opt); |
ea4fc0d6 | 437 | fl4 = &fl->u.ip4; |
511c3f92 | 438 | rt = skb_rtable(skb); |
00db4124 | 439 | if (rt) |
1da177e4 LT |
440 | goto packet_routed; |
441 | ||
442 | /* Make sure we can route this packet. */ | |
443 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
51456b29 | 444 | if (!rt) { |
3ca3c68e | 445 | __be32 daddr; |
1da177e4 LT |
446 | |
447 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 448 | daddr = inet->inet_daddr; |
f6d8bd05 ED |
449 | if (inet_opt && inet_opt->opt.srr) |
450 | daddr = inet_opt->opt.faddr; | |
1da177e4 | 451 | |
78fbfd8a DM |
452 | /* If this fails, retransmit mechanism of transport layer will |
453 | * keep trying until route appears or the connection times | |
454 | * itself out. | |
455 | */ | |
77589ce0 | 456 | rt = ip_route_output_ports(net, fl4, sk, |
78fbfd8a DM |
457 | daddr, inet->inet_saddr, |
458 | inet->inet_dport, | |
459 | inet->inet_sport, | |
460 | sk->sk_protocol, | |
461 | RT_CONN_FLAGS(sk), | |
462 | sk->sk_bound_dev_if); | |
463 | if (IS_ERR(rt)) | |
464 | goto no_route; | |
d8d1f30b | 465 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 466 | } |
d8d1f30b | 467 | skb_dst_set_noref(skb, &rt->dst); |
1da177e4 LT |
468 | |
469 | packet_routed: | |
155e8336 | 470 | if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway) |
1da177e4 LT |
471 | goto no_route; |
472 | ||
473 | /* OK, we know where to send it, allocate and build IP header. */ | |
f6d8bd05 | 474 | skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0)); |
8856dfa3 | 475 | skb_reset_network_header(skb); |
eddc9ec5 | 476 | iph = ip_hdr(skb); |
714e85be | 477 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
60ff7467 | 478 | if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df) |
1da177e4 LT |
479 | iph->frag_off = htons(IP_DF); |
480 | else | |
481 | iph->frag_off = 0; | |
d8d1f30b | 482 | iph->ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 483 | iph->protocol = sk->sk_protocol; |
84f9307c ED |
484 | ip_copy_addrs(iph, fl4); |
485 | ||
1da177e4 LT |
486 | /* Transport layer set skb->h.foo itself. */ |
487 | ||
f6d8bd05 ED |
488 | if (inet_opt && inet_opt->opt.optlen) { |
489 | iph->ihl += inet_opt->opt.optlen >> 2; | |
490 | ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0); | |
1da177e4 LT |
491 | } |
492 | ||
77589ce0 | 493 | ip_select_ident_segs(net, skb, sk, |
b6a7719a | 494 | skb_shinfo(skb)->gso_segs ?: 1); |
1da177e4 | 495 | |
b0270e91 | 496 | /* TODO : should we use skb->sk here instead of sk ? */ |
1da177e4 | 497 | skb->priority = sk->sk_priority; |
4a19ec58 | 498 | skb->mark = sk->sk_mark; |
1da177e4 | 499 | |
33224b16 | 500 | res = ip_local_out(net, sk, skb); |
ab6e3feb ED |
501 | rcu_read_unlock(); |
502 | return res; | |
1da177e4 LT |
503 | |
504 | no_route: | |
ab6e3feb | 505 | rcu_read_unlock(); |
77589ce0 | 506 | IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
507 | kfree_skb(skb); |
508 | return -EHOSTUNREACH; | |
509 | } | |
4bc2f18b | 510 | EXPORT_SYMBOL(ip_queue_xmit); |
1da177e4 | 511 | |
1da177e4 LT |
512 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) |
513 | { | |
514 | to->pkt_type = from->pkt_type; | |
515 | to->priority = from->priority; | |
516 | to->protocol = from->protocol; | |
adf30907 | 517 | skb_dst_drop(to); |
fe76cda3 | 518 | skb_dst_copy(to, from); |
1da177e4 | 519 | to->dev = from->dev; |
82e91ffe | 520 | to->mark = from->mark; |
1da177e4 LT |
521 | |
522 | /* Copy the flags to each fragment. */ | |
523 | IPCB(to)->flags = IPCB(from)->flags; | |
524 | ||
525 | #ifdef CONFIG_NET_SCHED | |
526 | to->tc_index = from->tc_index; | |
527 | #endif | |
e7ac05f3 | 528 | nf_copy(to, from); |
6ca40d4e | 529 | #if IS_ENABLED(CONFIG_IP_VS) |
c98d80ed | 530 | to->ipvs_property = from->ipvs_property; |
1da177e4 | 531 | #endif |
984bc16c | 532 | skb_copy_secmark(to, from); |
1da177e4 LT |
533 | } |
534 | ||
694869b3 | 535 | static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
c5501eb3 | 536 | unsigned int mtu, |
694869b3 | 537 | int (*output)(struct net *, struct sock *, struct sk_buff *)) |
49d16b23 AZ |
538 | { |
539 | struct iphdr *iph = ip_hdr(skb); | |
49d16b23 | 540 | |
d6b915e2 | 541 | if ((iph->frag_off & htons(IP_DF)) == 0) |
694869b3 | 542 | return ip_do_fragment(net, sk, skb, output); |
d6b915e2 FW |
543 | |
544 | if (unlikely(!skb->ignore_df || | |
49d16b23 AZ |
545 | (IPCB(skb)->frag_max_size && |
546 | IPCB(skb)->frag_max_size > mtu))) { | |
9479b0af | 547 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
49d16b23 AZ |
548 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
549 | htonl(mtu)); | |
550 | kfree_skb(skb); | |
551 | return -EMSGSIZE; | |
552 | } | |
553 | ||
694869b3 | 554 | return ip_do_fragment(net, sk, skb, output); |
49d16b23 AZ |
555 | } |
556 | ||
1da177e4 LT |
557 | /* |
558 | * This IP datagram is too large to be sent in one piece. Break it up into | |
559 | * smaller pieces (each of size equal to IP header plus | |
560 | * a block of the data of the original IP data part) that will yet fit in a | |
561 | * single device frame, and queue such a frame for sending. | |
562 | */ | |
563 | ||
694869b3 EB |
564 | int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb, |
565 | int (*output)(struct net *, struct sock *, struct sk_buff *)) | |
1da177e4 LT |
566 | { |
567 | struct iphdr *iph; | |
1da177e4 | 568 | int ptr; |
1da177e4 | 569 | struct sk_buff *skb2; |
c893b806 | 570 | unsigned int mtu, hlen, left, len, ll_rs; |
1da177e4 | 571 | int offset; |
76ab608d | 572 | __be16 not_last_frag; |
511c3f92 | 573 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
574 | int err = 0; |
575 | ||
dbd3393c HFS |
576 | /* for offloaded checksums cleanup checksum before fragmentation */ |
577 | if (skb->ip_summed == CHECKSUM_PARTIAL && | |
578 | (err = skb_checksum_help(skb))) | |
579 | goto fail; | |
580 | ||
1da177e4 LT |
581 | /* |
582 | * Point into the IP datagram header. | |
583 | */ | |
584 | ||
eddc9ec5 | 585 | iph = ip_hdr(skb); |
1da177e4 | 586 | |
fedbb6b4 | 587 | mtu = ip_skb_dst_mtu(sk, skb); |
d6b915e2 FW |
588 | if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu) |
589 | mtu = IPCB(skb)->frag_max_size; | |
1da177e4 LT |
590 | |
591 | /* | |
592 | * Setup starting values. | |
593 | */ | |
594 | ||
595 | hlen = iph->ihl * 4; | |
f87c10a8 | 596 | mtu = mtu - hlen; /* Size of data space */ |
89cee8b1 | 597 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
598 | |
599 | /* When frag_list is given, use it. First, check its validity: | |
600 | * some transformers could create wrong frag_list or break existing | |
601 | * one, it is not prohibited. In this case fall back to copying. | |
602 | * | |
603 | * LATER: this step can be merged to real generation of fragments, | |
604 | * we can switch to copy when see the first bad fragment. | |
605 | */ | |
21dc3301 | 606 | if (skb_has_frag_list(skb)) { |
3d13008e | 607 | struct sk_buff *frag, *frag2; |
c72d8cda | 608 | unsigned int first_len = skb_pagelen(skb); |
1da177e4 LT |
609 | |
610 | if (first_len - hlen > mtu || | |
611 | ((first_len - hlen) & 7) || | |
56f8a75c | 612 | ip_is_fragment(iph) || |
1da177e4 LT |
613 | skb_cloned(skb)) |
614 | goto slow_path; | |
615 | ||
d7fcf1a5 | 616 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
617 | /* Correct geometry. */ |
618 | if (frag->len > mtu || | |
619 | ((frag->len & 7) && frag->next) || | |
620 | skb_headroom(frag) < hlen) | |
3d13008e | 621 | goto slow_path_clean; |
1da177e4 LT |
622 | |
623 | /* Partially cloned skb? */ | |
624 | if (skb_shared(frag)) | |
3d13008e | 625 | goto slow_path_clean; |
2fdba6b0 HX |
626 | |
627 | BUG_ON(frag->sk); | |
628 | if (skb->sk) { | |
2fdba6b0 HX |
629 | frag->sk = skb->sk; |
630 | frag->destructor = sock_wfree; | |
2fdba6b0 | 631 | } |
3d13008e | 632 | skb->truesize -= frag->truesize; |
1da177e4 LT |
633 | } |
634 | ||
635 | /* Everything is OK. Generate! */ | |
636 | ||
637 | err = 0; | |
638 | offset = 0; | |
639 | frag = skb_shinfo(skb)->frag_list; | |
d7fcf1a5 | 640 | skb_frag_list_init(skb); |
1da177e4 LT |
641 | skb->data_len = first_len - skb_headlen(skb); |
642 | skb->len = first_len; | |
643 | iph->tot_len = htons(first_len); | |
644 | iph->frag_off = htons(IP_MF); | |
645 | ip_send_check(iph); | |
646 | ||
647 | for (;;) { | |
648 | /* Prepare header of the next frame, | |
649 | * before previous one went down. */ | |
650 | if (frag) { | |
651 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 652 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
653 | __skb_push(frag, hlen); |
654 | skb_reset_network_header(frag); | |
d56f90a7 | 655 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 656 | iph = ip_hdr(frag); |
1da177e4 LT |
657 | iph->tot_len = htons(frag->len); |
658 | ip_copy_metadata(frag, skb); | |
659 | if (offset == 0) | |
660 | ip_options_fragment(frag); | |
661 | offset += skb->len - hlen; | |
662 | iph->frag_off = htons(offset>>3); | |
00db4124 | 663 | if (frag->next) |
1da177e4 LT |
664 | iph->frag_off |= htons(IP_MF); |
665 | /* Ready, complete checksum */ | |
666 | ip_send_check(iph); | |
667 | } | |
668 | ||
694869b3 | 669 | err = output(net, sk, skb); |
1da177e4 | 670 | |
dafee490 | 671 | if (!err) |
26a949db | 672 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
673 | if (err || !frag) |
674 | break; | |
675 | ||
676 | skb = frag; | |
677 | frag = skb->next; | |
678 | skb->next = NULL; | |
679 | } | |
680 | ||
681 | if (err == 0) { | |
26a949db | 682 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
683 | return 0; |
684 | } | |
685 | ||
686 | while (frag) { | |
687 | skb = frag->next; | |
688 | kfree_skb(frag); | |
689 | frag = skb; | |
690 | } | |
26a949db | 691 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 692 | return err; |
3d13008e ED |
693 | |
694 | slow_path_clean: | |
695 | skb_walk_frags(skb, frag2) { | |
696 | if (frag2 == frag) | |
697 | break; | |
698 | frag2->sk = NULL; | |
699 | frag2->destructor = NULL; | |
700 | skb->truesize += frag2->truesize; | |
701 | } | |
1da177e4 LT |
702 | } |
703 | ||
704 | slow_path: | |
c9af6db4 | 705 | iph = ip_hdr(skb); |
fc70fb64 | 706 | |
1da177e4 | 707 | left = skb->len - hlen; /* Space per frame */ |
49085bd7 | 708 | ptr = hlen; /* Where to start from */ |
1da177e4 | 709 | |
8d045163 | 710 | ll_rs = LL_RESERVED_SPACE(rt->dst.dev); |
9bcfcaf5 | 711 | |
1da177e4 LT |
712 | /* |
713 | * Fragment the datagram. | |
714 | */ | |
715 | ||
716 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
717 | not_last_frag = iph->frag_off & htons(IP_MF); | |
718 | ||
719 | /* | |
720 | * Keep copying data until we run out. | |
721 | */ | |
722 | ||
132adf54 | 723 | while (left > 0) { |
1da177e4 LT |
724 | len = left; |
725 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
726 | if (len > mtu) | |
727 | len = mtu; | |
25985edc | 728 | /* IF: we are not sending up to and including the packet end |
1da177e4 LT |
729 | then align the next start on an eight byte boundary */ |
730 | if (len < left) { | |
731 | len &= ~7; | |
732 | } | |
1da177e4 | 733 | |
cbffccc9 JP |
734 | /* Allocate buffer */ |
735 | skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC); | |
736 | if (!skb2) { | |
1da177e4 LT |
737 | err = -ENOMEM; |
738 | goto fail; | |
739 | } | |
740 | ||
741 | /* | |
742 | * Set up data on packet | |
743 | */ | |
744 | ||
745 | ip_copy_metadata(skb2, skb); | |
746 | skb_reserve(skb2, ll_rs); | |
747 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 748 | skb_reset_network_header(skb2); |
b0e380b1 | 749 | skb2->transport_header = skb2->network_header + hlen; |
1da177e4 LT |
750 | |
751 | /* | |
752 | * Charge the memory for the fragment to any owner | |
753 | * it might possess | |
754 | */ | |
755 | ||
756 | if (skb->sk) | |
757 | skb_set_owner_w(skb2, skb->sk); | |
758 | ||
759 | /* | |
760 | * Copy the packet header into the new buffer. | |
761 | */ | |
762 | ||
d626f62b | 763 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); |
1da177e4 LT |
764 | |
765 | /* | |
766 | * Copy a block of the IP datagram. | |
767 | */ | |
bff9b61c | 768 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) |
1da177e4 LT |
769 | BUG(); |
770 | left -= len; | |
771 | ||
772 | /* | |
773 | * Fill in the new header fields. | |
774 | */ | |
eddc9ec5 | 775 | iph = ip_hdr(skb2); |
1da177e4 LT |
776 | iph->frag_off = htons((offset >> 3)); |
777 | ||
d6b915e2 FW |
778 | if (IPCB(skb)->flags & IPSKB_FRAG_PMTU) |
779 | iph->frag_off |= htons(IP_DF); | |
780 | ||
1da177e4 LT |
781 | /* ANK: dirty, but effective trick. Upgrade options only if |
782 | * the segment to be fragmented was THE FIRST (otherwise, | |
783 | * options are already fixed) and make it ONCE | |
784 | * on the initial skb, so that all the following fragments | |
785 | * will inherit fixed options. | |
786 | */ | |
787 | if (offset == 0) | |
788 | ip_options_fragment(skb); | |
789 | ||
790 | /* | |
791 | * Added AC : If we are fragmenting a fragment that's not the | |
792 | * last fragment then keep MF on each bit | |
793 | */ | |
794 | if (left > 0 || not_last_frag) | |
795 | iph->frag_off |= htons(IP_MF); | |
796 | ptr += len; | |
797 | offset += len; | |
798 | ||
799 | /* | |
800 | * Put this fragment into the sending queue. | |
801 | */ | |
1da177e4 LT |
802 | iph->tot_len = htons(len + hlen); |
803 | ||
804 | ip_send_check(iph); | |
805 | ||
694869b3 | 806 | err = output(net, sk, skb2); |
1da177e4 LT |
807 | if (err) |
808 | goto fail; | |
dafee490 | 809 | |
26a949db | 810 | IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES); |
1da177e4 | 811 | } |
5d0ba55b | 812 | consume_skb(skb); |
26a949db | 813 | IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
814 | return err; |
815 | ||
816 | fail: | |
e905a9ed | 817 | kfree_skb(skb); |
26a949db | 818 | IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
819 | return err; |
820 | } | |
49d16b23 | 821 | EXPORT_SYMBOL(ip_do_fragment); |
2e2f7aef | 822 | |
1da177e4 LT |
823 | int |
824 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
825 | { | |
f69e6d13 | 826 | struct msghdr *msg = from; |
1da177e4 | 827 | |
84fa7933 | 828 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
0b62fca2 | 829 | if (!copy_from_iter_full(to, len, &msg->msg_iter)) |
1da177e4 LT |
830 | return -EFAULT; |
831 | } else { | |
44bb9363 | 832 | __wsum csum = 0; |
0b62fca2 | 833 | if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter)) |
1da177e4 LT |
834 | return -EFAULT; |
835 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
836 | } | |
837 | return 0; | |
838 | } | |
4bc2f18b | 839 | EXPORT_SYMBOL(ip_generic_getfrag); |
1da177e4 | 840 | |
44bb9363 | 841 | static inline __wsum |
1da177e4 LT |
842 | csum_page(struct page *page, int offset, int copy) |
843 | { | |
844 | char *kaddr; | |
44bb9363 | 845 | __wsum csum; |
1da177e4 LT |
846 | kaddr = kmap(page); |
847 | csum = csum_partial(kaddr + offset, copy, 0); | |
848 | kunmap(page); | |
849 | return csum; | |
850 | } | |
851 | ||
4b30b1c6 | 852 | static inline int ip_ufo_append_data(struct sock *sk, |
1470ddf7 | 853 | struct sk_buff_head *queue, |
e89e9cf5 AR |
854 | int getfrag(void *from, char *to, int offset, int len, |
855 | int odd, struct sk_buff *skb), | |
856 | void *from, int length, int hh_len, int fragheaderlen, | |
d9be4f7a | 857 | int transhdrlen, int maxfraglen, unsigned int flags) |
e89e9cf5 AR |
858 | { |
859 | struct sk_buff *skb; | |
860 | int err; | |
861 | ||
862 | /* There is support for UDP fragmentation offload by network | |
863 | * device, so create one single skb packet containing complete | |
864 | * udp datagram | |
865 | */ | |
51456b29 IM |
866 | skb = skb_peek_tail(queue); |
867 | if (!skb) { | |
e89e9cf5 AR |
868 | skb = sock_alloc_send_skb(sk, |
869 | hh_len + fragheaderlen + transhdrlen + 20, | |
870 | (flags & MSG_DONTWAIT), &err); | |
871 | ||
51456b29 | 872 | if (!skb) |
e89e9cf5 AR |
873 | return err; |
874 | ||
875 | /* reserve space for Hardware header */ | |
876 | skb_reserve(skb, hh_len); | |
877 | ||
878 | /* create space for UDP/IP header */ | |
d9319100 | 879 | skb_put(skb, fragheaderlen + transhdrlen); |
e89e9cf5 AR |
880 | |
881 | /* initialize network header pointer */ | |
c1d2bbe1 | 882 | skb_reset_network_header(skb); |
e89e9cf5 AR |
883 | |
884 | /* initialize protocol header pointer */ | |
b0e380b1 | 885 | skb->transport_header = skb->network_header + fragheaderlen; |
e89e9cf5 | 886 | |
e89e9cf5 | 887 | skb->csum = 0; |
e89e9cf5 | 888 | |
1470ddf7 | 889 | __skb_queue_tail(queue, skb); |
e93b7d74 JP |
890 | } else if (skb_is_gso(skb)) { |
891 | goto append; | |
e89e9cf5 | 892 | } |
be9164e7 | 893 | |
e93b7d74 JP |
894 | skb->ip_summed = CHECKSUM_PARTIAL; |
895 | /* specify the length of each IP datagram fragment */ | |
896 | skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen; | |
897 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
898 | ||
899 | append: | |
be9164e7 K |
900 | return skb_append_datato_frags(sk, skb, getfrag, from, |
901 | (length - transhdrlen)); | |
e89e9cf5 AR |
902 | } |
903 | ||
f5fca608 DM |
904 | static int __ip_append_data(struct sock *sk, |
905 | struct flowi4 *fl4, | |
906 | struct sk_buff_head *queue, | |
1470ddf7 | 907 | struct inet_cork *cork, |
5640f768 | 908 | struct page_frag *pfrag, |
1470ddf7 HX |
909 | int getfrag(void *from, char *to, int offset, |
910 | int len, int odd, struct sk_buff *skb), | |
911 | void *from, int length, int transhdrlen, | |
912 | unsigned int flags) | |
1da177e4 LT |
913 | { |
914 | struct inet_sock *inet = inet_sk(sk); | |
915 | struct sk_buff *skb; | |
916 | ||
07df5294 | 917 | struct ip_options *opt = cork->opt; |
1da177e4 LT |
918 | int hh_len; |
919 | int exthdrlen; | |
920 | int mtu; | |
921 | int copy; | |
922 | int err; | |
923 | int offset = 0; | |
daba287b | 924 | unsigned int maxfraglen, fragheaderlen, maxnonfragsize; |
1da177e4 | 925 | int csummode = CHECKSUM_NONE; |
1470ddf7 | 926 | struct rtable *rt = (struct rtable *)cork->dst; |
09c2d251 | 927 | u32 tskey = 0; |
1da177e4 | 928 | |
96d7303e SK |
929 | skb = skb_peek_tail(queue); |
930 | ||
931 | exthdrlen = !skb ? rt->dst.header_len : 0; | |
07df5294 | 932 | mtu = cork->fragsize; |
09c2d251 WB |
933 | if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP && |
934 | sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) | |
935 | tskey = sk->sk_tskey++; | |
1da177e4 | 936 | |
d8d1f30b | 937 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
1da177e4 LT |
938 | |
939 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
940 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 941 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 942 | |
daba287b | 943 | if (cork->length + length > maxnonfragsize - fragheaderlen) { |
f5fca608 | 944 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
61e7f09d | 945 | mtu - (opt ? opt->optlen : 0)); |
1da177e4 LT |
946 | return -EMSGSIZE; |
947 | } | |
948 | ||
949 | /* | |
950 | * transhdrlen > 0 means that this is the first fragment and we wish | |
951 | * it won't be fragmented in the future. | |
952 | */ | |
953 | if (transhdrlen && | |
954 | length + fragheaderlen <= mtu && | |
c8cd0989 | 955 | rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) && |
d749c9cb | 956 | !(flags & MSG_MORE) && |
1da177e4 | 957 | !exthdrlen) |
84fa7933 | 958 | csummode = CHECKSUM_PARTIAL; |
1da177e4 | 959 | |
1470ddf7 | 960 | cork->length += length; |
0a28cfd5 | 961 | if ((((length + fragheaderlen) > mtu) || (skb && skb_is_gso(skb))) && |
be9164e7 | 962 | (sk->sk_protocol == IPPROTO_UDP) && |
acf8dd0a | 963 | (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len && |
40ba3302 | 964 | (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx) { |
1470ddf7 HX |
965 | err = ip_ufo_append_data(sk, queue, getfrag, from, length, |
966 | hh_len, fragheaderlen, transhdrlen, | |
d9be4f7a | 967 | maxfraglen, flags); |
baa829d8 | 968 | if (err) |
e89e9cf5 | 969 | goto error; |
e89e9cf5 AR |
970 | return 0; |
971 | } | |
1da177e4 LT |
972 | |
973 | /* So, what's going on in the loop below? | |
974 | * | |
975 | * We use calculated fragment length to generate chained skb, | |
976 | * each of segments is IP fragment ready for sending to network after | |
977 | * adding appropriate IP header. | |
978 | */ | |
979 | ||
26cde9f7 | 980 | if (!skb) |
1da177e4 LT |
981 | goto alloc_new_skb; |
982 | ||
983 | while (length > 0) { | |
984 | /* Check if the remaining data fits into current packet. */ | |
985 | copy = mtu - skb->len; | |
986 | if (copy < length) | |
987 | copy = maxfraglen - skb->len; | |
988 | if (copy <= 0) { | |
989 | char *data; | |
990 | unsigned int datalen; | |
991 | unsigned int fraglen; | |
992 | unsigned int fraggap; | |
993 | unsigned int alloclen; | |
994 | struct sk_buff *skb_prev; | |
995 | alloc_new_skb: | |
996 | skb_prev = skb; | |
997 | if (skb_prev) | |
998 | fraggap = skb_prev->len - maxfraglen; | |
999 | else | |
1000 | fraggap = 0; | |
1001 | ||
1002 | /* | |
1003 | * If remaining data exceeds the mtu, | |
1004 | * we know we need more fragment(s). | |
1005 | */ | |
1006 | datalen = length + fraggap; | |
1007 | if (datalen > mtu - fragheaderlen) | |
1008 | datalen = maxfraglen - fragheaderlen; | |
1009 | fraglen = datalen + fragheaderlen; | |
1010 | ||
e905a9ed | 1011 | if ((flags & MSG_MORE) && |
d8d1f30b | 1012 | !(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
1013 | alloclen = mtu; |
1014 | else | |
59104f06 | 1015 | alloclen = fraglen; |
1da177e4 | 1016 | |
353e5c9a SK |
1017 | alloclen += exthdrlen; |
1018 | ||
1da177e4 LT |
1019 | /* The last fragment gets additional space at tail. |
1020 | * Note, with MSG_MORE we overallocate on fragments, | |
1021 | * because we have no idea what fragment will be | |
1022 | * the last. | |
1023 | */ | |
33f99dc7 | 1024 | if (datalen == length + fraggap) |
d8d1f30b | 1025 | alloclen += rt->dst.trailer_len; |
33f99dc7 | 1026 | |
1da177e4 | 1027 | if (transhdrlen) { |
e905a9ed | 1028 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
1029 | alloclen + hh_len + 15, |
1030 | (flags & MSG_DONTWAIT), &err); | |
1031 | } else { | |
1032 | skb = NULL; | |
1033 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
1034 | 2 * sk->sk_sndbuf) | |
e905a9ed | 1035 | skb = sock_wmalloc(sk, |
1da177e4 LT |
1036 | alloclen + hh_len + 15, 1, |
1037 | sk->sk_allocation); | |
51456b29 | 1038 | if (unlikely(!skb)) |
1da177e4 LT |
1039 | err = -ENOBUFS; |
1040 | } | |
51456b29 | 1041 | if (!skb) |
1da177e4 LT |
1042 | goto error; |
1043 | ||
1044 | /* | |
1045 | * Fill in the control structures | |
1046 | */ | |
1047 | skb->ip_summed = csummode; | |
1048 | skb->csum = 0; | |
1049 | skb_reserve(skb, hh_len); | |
11878b40 WB |
1050 | |
1051 | /* only the initial fragment is time stamped */ | |
1470ddf7 | 1052 | skb_shinfo(skb)->tx_flags = cork->tx_flags; |
11878b40 | 1053 | cork->tx_flags = 0; |
09c2d251 WB |
1054 | skb_shinfo(skb)->tskey = tskey; |
1055 | tskey = 0; | |
1da177e4 LT |
1056 | |
1057 | /* | |
1058 | * Find where to start putting bytes. | |
1059 | */ | |
353e5c9a | 1060 | data = skb_put(skb, fraglen + exthdrlen); |
c14d2450 | 1061 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
1062 | skb->transport_header = (skb->network_header + |
1063 | fragheaderlen); | |
353e5c9a | 1064 | data += fragheaderlen + exthdrlen; |
1da177e4 LT |
1065 | |
1066 | if (fraggap) { | |
1067 | skb->csum = skb_copy_and_csum_bits( | |
1068 | skb_prev, maxfraglen, | |
1069 | data + transhdrlen, fraggap, 0); | |
1070 | skb_prev->csum = csum_sub(skb_prev->csum, | |
1071 | skb->csum); | |
1072 | data += fraggap; | |
e9fa4f7b | 1073 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | copy = datalen - transhdrlen - fraggap; | |
1077 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
1078 | err = -EFAULT; | |
1079 | kfree_skb(skb); | |
1080 | goto error; | |
1081 | } | |
1082 | ||
1083 | offset += copy; | |
1084 | length -= datalen - fraggap; | |
1085 | transhdrlen = 0; | |
1086 | exthdrlen = 0; | |
1087 | csummode = CHECKSUM_NONE; | |
1088 | ||
1089 | /* | |
1090 | * Put the packet on the pending queue. | |
1091 | */ | |
1470ddf7 | 1092 | __skb_queue_tail(queue, skb); |
1da177e4 LT |
1093 | continue; |
1094 | } | |
1095 | ||
1096 | if (copy > length) | |
1097 | copy = length; | |
1098 | ||
d8d1f30b | 1099 | if (!(rt->dst.dev->features&NETIF_F_SG)) { |
1da177e4 LT |
1100 | unsigned int off; |
1101 | ||
1102 | off = skb->len; | |
e905a9ed | 1103 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
1104 | offset, copy, off, skb) < 0) { |
1105 | __skb_trim(skb, off); | |
1106 | err = -EFAULT; | |
1107 | goto error; | |
1108 | } | |
1109 | } else { | |
1110 | int i = skb_shinfo(skb)->nr_frags; | |
1da177e4 | 1111 | |
5640f768 ED |
1112 | err = -ENOMEM; |
1113 | if (!sk_page_frag_refill(sk, pfrag)) | |
1da177e4 | 1114 | goto error; |
5640f768 ED |
1115 | |
1116 | if (!skb_can_coalesce(skb, i, pfrag->page, | |
1117 | pfrag->offset)) { | |
1118 | err = -EMSGSIZE; | |
1119 | if (i == MAX_SKB_FRAGS) | |
1120 | goto error; | |
1121 | ||
1122 | __skb_fill_page_desc(skb, i, pfrag->page, | |
1123 | pfrag->offset, 0); | |
1124 | skb_shinfo(skb)->nr_frags = ++i; | |
1125 | get_page(pfrag->page); | |
1da177e4 | 1126 | } |
5640f768 ED |
1127 | copy = min_t(int, copy, pfrag->size - pfrag->offset); |
1128 | if (getfrag(from, | |
1129 | page_address(pfrag->page) + pfrag->offset, | |
1130 | offset, copy, skb->len, skb) < 0) | |
1131 | goto error_efault; | |
1132 | ||
1133 | pfrag->offset += copy; | |
1134 | skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy); | |
1da177e4 LT |
1135 | skb->len += copy; |
1136 | skb->data_len += copy; | |
f945fa7a HX |
1137 | skb->truesize += copy; |
1138 | atomic_add(copy, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1139 | } |
1140 | offset += copy; | |
1141 | length -= copy; | |
1142 | } | |
1143 | ||
1144 | return 0; | |
1145 | ||
5640f768 ED |
1146 | error_efault: |
1147 | err = -EFAULT; | |
1da177e4 | 1148 | error: |
1470ddf7 | 1149 | cork->length -= length; |
5e38e270 | 1150 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
e905a9ed | 1151 | return err; |
1da177e4 LT |
1152 | } |
1153 | ||
1470ddf7 HX |
1154 | static int ip_setup_cork(struct sock *sk, struct inet_cork *cork, |
1155 | struct ipcm_cookie *ipc, struct rtable **rtp) | |
1156 | { | |
f6d8bd05 | 1157 | struct ip_options_rcu *opt; |
1470ddf7 HX |
1158 | struct rtable *rt; |
1159 | ||
1160 | /* | |
1161 | * setup for corking. | |
1162 | */ | |
1163 | opt = ipc->opt; | |
1164 | if (opt) { | |
51456b29 | 1165 | if (!cork->opt) { |
1470ddf7 HX |
1166 | cork->opt = kmalloc(sizeof(struct ip_options) + 40, |
1167 | sk->sk_allocation); | |
51456b29 | 1168 | if (unlikely(!cork->opt)) |
1470ddf7 HX |
1169 | return -ENOBUFS; |
1170 | } | |
f6d8bd05 | 1171 | memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen); |
1470ddf7 HX |
1172 | cork->flags |= IPCORK_OPT; |
1173 | cork->addr = ipc->addr; | |
1174 | } | |
1175 | rt = *rtp; | |
1176 | if (unlikely(!rt)) | |
1177 | return -EFAULT; | |
1178 | /* | |
1179 | * We steal reference to this route, caller should not release it | |
1180 | */ | |
1181 | *rtp = NULL; | |
482fc609 HFS |
1182 | cork->fragsize = ip_sk_use_pmtu(sk) ? |
1183 | dst_mtu(&rt->dst) : rt->dst.dev->mtu; | |
1470ddf7 HX |
1184 | cork->dst = &rt->dst; |
1185 | cork->length = 0; | |
aa661581 FF |
1186 | cork->ttl = ipc->ttl; |
1187 | cork->tos = ipc->tos; | |
1188 | cork->priority = ipc->priority; | |
1470ddf7 | 1189 | cork->tx_flags = ipc->tx_flags; |
1470ddf7 HX |
1190 | |
1191 | return 0; | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
1196 | * from many pieces of data. Each pieces will be holded on the socket | |
1197 | * until ip_push_pending_frames() is called. Each piece can be a page | |
1198 | * or non-page data. | |
1199 | * | |
1200 | * Not only UDP, other transport protocols - e.g. raw sockets - can use | |
1201 | * this interface potentially. | |
1202 | * | |
1203 | * LATER: length must be adjusted by pad at tail, when it is required. | |
1204 | */ | |
f5fca608 | 1205 | int ip_append_data(struct sock *sk, struct flowi4 *fl4, |
1470ddf7 HX |
1206 | int getfrag(void *from, char *to, int offset, int len, |
1207 | int odd, struct sk_buff *skb), | |
1208 | void *from, int length, int transhdrlen, | |
1209 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1210 | unsigned int flags) | |
1211 | { | |
1212 | struct inet_sock *inet = inet_sk(sk); | |
1213 | int err; | |
1214 | ||
1215 | if (flags&MSG_PROBE) | |
1216 | return 0; | |
1217 | ||
1218 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
bdc712b4 | 1219 | err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp); |
1470ddf7 HX |
1220 | if (err) |
1221 | return err; | |
1222 | } else { | |
1223 | transhdrlen = 0; | |
1224 | } | |
1225 | ||
5640f768 ED |
1226 | return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base, |
1227 | sk_page_frag(sk), getfrag, | |
1470ddf7 HX |
1228 | from, length, transhdrlen, flags); |
1229 | } | |
1230 | ||
f5fca608 | 1231 | ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page, |
1da177e4 LT |
1232 | int offset, size_t size, int flags) |
1233 | { | |
1234 | struct inet_sock *inet = inet_sk(sk); | |
1235 | struct sk_buff *skb; | |
1236 | struct rtable *rt; | |
1237 | struct ip_options *opt = NULL; | |
bdc712b4 | 1238 | struct inet_cork *cork; |
1da177e4 LT |
1239 | int hh_len; |
1240 | int mtu; | |
1241 | int len; | |
1242 | int err; | |
daba287b | 1243 | unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize; |
1da177e4 LT |
1244 | |
1245 | if (inet->hdrincl) | |
1246 | return -EPERM; | |
1247 | ||
1248 | if (flags&MSG_PROBE) | |
1249 | return 0; | |
1250 | ||
1251 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1252 | return -EINVAL; | |
1253 | ||
bdc712b4 DM |
1254 | cork = &inet->cork.base; |
1255 | rt = (struct rtable *)cork->dst; | |
1256 | if (cork->flags & IPCORK_OPT) | |
1257 | opt = cork->opt; | |
1da177e4 | 1258 | |
d8d1f30b | 1259 | if (!(rt->dst.dev->features&NETIF_F_SG)) |
1da177e4 LT |
1260 | return -EOPNOTSUPP; |
1261 | ||
d8d1f30b | 1262 | hh_len = LL_RESERVED_SPACE(rt->dst.dev); |
bdc712b4 | 1263 | mtu = cork->fragsize; |
1da177e4 LT |
1264 | |
1265 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1266 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
60ff7467 | 1267 | maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu; |
1da177e4 | 1268 | |
daba287b | 1269 | if (cork->length + size > maxnonfragsize - fragheaderlen) { |
61e7f09d HFS |
1270 | ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport, |
1271 | mtu - (opt ? opt->optlen : 0)); | |
1da177e4 LT |
1272 | return -EMSGSIZE; |
1273 | } | |
1274 | ||
51456b29 IM |
1275 | skb = skb_peek_tail(&sk->sk_write_queue); |
1276 | if (!skb) | |
1da177e4 LT |
1277 | return -EINVAL; |
1278 | ||
26cde9f7 HX |
1279 | if ((size + skb->len > mtu) && |
1280 | (sk->sk_protocol == IPPROTO_UDP) && | |
d8d1f30b | 1281 | (rt->dst.dev->features & NETIF_F_UFO)) { |
a8c4a252 HFS |
1282 | if (skb->ip_summed != CHECKSUM_PARTIAL) |
1283 | return -EOPNOTSUPP; | |
1284 | ||
7967168c | 1285 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 1286 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1287 | } |
a8c4a252 | 1288 | cork->length += size; |
e89e9cf5 | 1289 | |
1da177e4 | 1290 | while (size > 0) { |
be12a1fe | 1291 | if (skb_is_gso(skb)) { |
e89e9cf5 | 1292 | len = size; |
be12a1fe | 1293 | } else { |
e89e9cf5 AR |
1294 | |
1295 | /* Check if the remaining data fits into current packet. */ | |
1296 | len = mtu - skb->len; | |
1297 | if (len < size) | |
1298 | len = maxfraglen - skb->len; | |
1299 | } | |
1da177e4 LT |
1300 | if (len <= 0) { |
1301 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1302 | int alloclen; |
1303 | ||
1304 | skb_prev = skb; | |
0d0d2bba | 1305 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1306 | |
1307 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1308 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1309 | if (unlikely(!skb)) { | |
1310 | err = -ENOBUFS; | |
1311 | goto error; | |
1312 | } | |
1313 | ||
1314 | /* | |
1315 | * Fill in the control structures | |
1316 | */ | |
1317 | skb->ip_summed = CHECKSUM_NONE; | |
1318 | skb->csum = 0; | |
1319 | skb_reserve(skb, hh_len); | |
1320 | ||
1321 | /* | |
1322 | * Find where to start putting bytes. | |
1323 | */ | |
967b05f6 | 1324 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1325 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1326 | skb->transport_header = (skb->network_header + |
1327 | fragheaderlen); | |
1da177e4 | 1328 | if (fraggap) { |
967b05f6 ACM |
1329 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1330 | maxfraglen, | |
9c70220b | 1331 | skb_transport_header(skb), |
967b05f6 | 1332 | fraggap, 0); |
1da177e4 LT |
1333 | skb_prev->csum = csum_sub(skb_prev->csum, |
1334 | skb->csum); | |
e9fa4f7b | 1335 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | /* | |
1339 | * Put the packet on the pending queue. | |
1340 | */ | |
1341 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1342 | continue; | |
1343 | } | |
1344 | ||
1da177e4 LT |
1345 | if (len > size) |
1346 | len = size; | |
be12a1fe HFS |
1347 | |
1348 | if (skb_append_pagefrags(skb, page, offset, len)) { | |
1da177e4 LT |
1349 | err = -EMSGSIZE; |
1350 | goto error; | |
1351 | } | |
1352 | ||
1353 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1354 | __wsum csum; |
1da177e4 LT |
1355 | csum = csum_page(page, offset, len); |
1356 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1357 | } | |
1358 | ||
1359 | skb->len += len; | |
1360 | skb->data_len += len; | |
1e34a11d DM |
1361 | skb->truesize += len; |
1362 | atomic_add(len, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1363 | offset += len; |
1364 | size -= len; | |
1365 | } | |
1366 | return 0; | |
1367 | ||
1368 | error: | |
bdc712b4 | 1369 | cork->length -= size; |
5e38e270 | 1370 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1371 | return err; |
1372 | } | |
1373 | ||
1470ddf7 | 1374 | static void ip_cork_release(struct inet_cork *cork) |
429f08e9 | 1375 | { |
1470ddf7 HX |
1376 | cork->flags &= ~IPCORK_OPT; |
1377 | kfree(cork->opt); | |
1378 | cork->opt = NULL; | |
1379 | dst_release(cork->dst); | |
1380 | cork->dst = NULL; | |
429f08e9 PE |
1381 | } |
1382 | ||
1da177e4 LT |
1383 | /* |
1384 | * Combined all pending IP fragments on the socket as one IP datagram | |
1385 | * and push them out. | |
1386 | */ | |
1c32c5ad | 1387 | struct sk_buff *__ip_make_skb(struct sock *sk, |
77968b78 | 1388 | struct flowi4 *fl4, |
1c32c5ad HX |
1389 | struct sk_buff_head *queue, |
1390 | struct inet_cork *cork) | |
1da177e4 LT |
1391 | { |
1392 | struct sk_buff *skb, *tmp_skb; | |
1393 | struct sk_buff **tail_skb; | |
1394 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1395 | struct net *net = sock_net(sk); |
1da177e4 | 1396 | struct ip_options *opt = NULL; |
1470ddf7 | 1397 | struct rtable *rt = (struct rtable *)cork->dst; |
1da177e4 | 1398 | struct iphdr *iph; |
76ab608d | 1399 | __be16 df = 0; |
1da177e4 | 1400 | __u8 ttl; |
1da177e4 | 1401 | |
51456b29 IM |
1402 | skb = __skb_dequeue(queue); |
1403 | if (!skb) | |
1da177e4 LT |
1404 | goto out; |
1405 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1406 | ||
1407 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1408 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1409 | __skb_pull(skb, skb_network_offset(skb)); |
1470ddf7 | 1410 | while ((tmp_skb = __skb_dequeue(queue)) != NULL) { |
cfe1fc77 | 1411 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1412 | *tail_skb = tmp_skb; |
1413 | tail_skb = &(tmp_skb->next); | |
1414 | skb->len += tmp_skb->len; | |
1415 | skb->data_len += tmp_skb->len; | |
1416 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1417 | tmp_skb->destructor = NULL; |
1418 | tmp_skb->sk = NULL; | |
1419 | } | |
1420 | ||
1421 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1422 | * to fragment the frame generated here. No matter, what transforms | |
1423 | * how transforms change size of the packet, it will come out. | |
1424 | */ | |
60ff7467 | 1425 | skb->ignore_df = ip_sk_ignore_df(sk); |
1da177e4 LT |
1426 | |
1427 | /* DF bit is set when we want to see DF on outgoing frames. | |
60ff7467 | 1428 | * If ignore_df is set too, we still allow to fragment this frame |
1da177e4 | 1429 | * locally. */ |
482fc609 HFS |
1430 | if (inet->pmtudisc == IP_PMTUDISC_DO || |
1431 | inet->pmtudisc == IP_PMTUDISC_PROBE || | |
d8d1f30b CG |
1432 | (skb->len <= dst_mtu(&rt->dst) && |
1433 | ip_dont_fragment(sk, &rt->dst))) | |
1da177e4 LT |
1434 | df = htons(IP_DF); |
1435 | ||
1470ddf7 HX |
1436 | if (cork->flags & IPCORK_OPT) |
1437 | opt = cork->opt; | |
1da177e4 | 1438 | |
aa661581 FF |
1439 | if (cork->ttl != 0) |
1440 | ttl = cork->ttl; | |
1441 | else if (rt->rt_type == RTN_MULTICAST) | |
1da177e4 LT |
1442 | ttl = inet->mc_ttl; |
1443 | else | |
d8d1f30b | 1444 | ttl = ip_select_ttl(inet, &rt->dst); |
1da177e4 | 1445 | |
749154aa | 1446 | iph = ip_hdr(skb); |
1da177e4 LT |
1447 | iph->version = 4; |
1448 | iph->ihl = 5; | |
aa661581 | 1449 | iph->tos = (cork->tos != -1) ? cork->tos : inet->tos; |
1da177e4 | 1450 | iph->frag_off = df; |
1da177e4 LT |
1451 | iph->ttl = ttl; |
1452 | iph->protocol = sk->sk_protocol; | |
84f9307c | 1453 | ip_copy_addrs(iph, fl4); |
b6a7719a | 1454 | ip_select_ident(net, skb, sk); |
1da177e4 | 1455 | |
22f728f8 DM |
1456 | if (opt) { |
1457 | iph->ihl += opt->optlen>>2; | |
1458 | ip_options_build(skb, opt, cork->addr, rt, 0); | |
1459 | } | |
1460 | ||
aa661581 | 1461 | skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority; |
4a19ec58 | 1462 | skb->mark = sk->sk_mark; |
a21bba94 ED |
1463 | /* |
1464 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1465 | * on dst refcount | |
1466 | */ | |
1470ddf7 | 1467 | cork->dst = NULL; |
d8d1f30b | 1468 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 1469 | |
96793b48 | 1470 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1471 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1472 | skb_transport_header(skb))->type); |
1473 | ||
1c32c5ad HX |
1474 | ip_cork_release(cork); |
1475 | out: | |
1476 | return skb; | |
1477 | } | |
1478 | ||
b5ec8eea | 1479 | int ip_send_skb(struct net *net, struct sk_buff *skb) |
1c32c5ad | 1480 | { |
1c32c5ad HX |
1481 | int err; |
1482 | ||
33224b16 | 1483 | err = ip_local_out(net, skb->sk, skb); |
1da177e4 LT |
1484 | if (err) { |
1485 | if (err > 0) | |
6ce9e7b5 | 1486 | err = net_xmit_errno(err); |
1da177e4 | 1487 | if (err) |
1c32c5ad | 1488 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1489 | } |
1490 | ||
1da177e4 | 1491 | return err; |
1da177e4 LT |
1492 | } |
1493 | ||
77968b78 | 1494 | int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4) |
1470ddf7 | 1495 | { |
1c32c5ad HX |
1496 | struct sk_buff *skb; |
1497 | ||
77968b78 | 1498 | skb = ip_finish_skb(sk, fl4); |
1c32c5ad HX |
1499 | if (!skb) |
1500 | return 0; | |
1501 | ||
1502 | /* Netfilter gets whole the not fragmented skb. */ | |
b5ec8eea | 1503 | return ip_send_skb(sock_net(sk), skb); |
1470ddf7 HX |
1504 | } |
1505 | ||
1da177e4 LT |
1506 | /* |
1507 | * Throw away all pending data on the socket. | |
1508 | */ | |
1470ddf7 HX |
1509 | static void __ip_flush_pending_frames(struct sock *sk, |
1510 | struct sk_buff_head *queue, | |
1511 | struct inet_cork *cork) | |
1da177e4 | 1512 | { |
1da177e4 LT |
1513 | struct sk_buff *skb; |
1514 | ||
1470ddf7 | 1515 | while ((skb = __skb_dequeue_tail(queue)) != NULL) |
1da177e4 LT |
1516 | kfree_skb(skb); |
1517 | ||
1470ddf7 HX |
1518 | ip_cork_release(cork); |
1519 | } | |
1520 | ||
1521 | void ip_flush_pending_frames(struct sock *sk) | |
1522 | { | |
bdc712b4 | 1523 | __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base); |
1da177e4 LT |
1524 | } |
1525 | ||
1c32c5ad | 1526 | struct sk_buff *ip_make_skb(struct sock *sk, |
77968b78 | 1527 | struct flowi4 *fl4, |
1c32c5ad HX |
1528 | int getfrag(void *from, char *to, int offset, |
1529 | int len, int odd, struct sk_buff *skb), | |
1530 | void *from, int length, int transhdrlen, | |
1531 | struct ipcm_cookie *ipc, struct rtable **rtp, | |
1532 | unsigned int flags) | |
1533 | { | |
b80d7226 | 1534 | struct inet_cork cork; |
1c32c5ad HX |
1535 | struct sk_buff_head queue; |
1536 | int err; | |
1537 | ||
1538 | if (flags & MSG_PROBE) | |
1539 | return NULL; | |
1540 | ||
1541 | __skb_queue_head_init(&queue); | |
1542 | ||
b80d7226 DM |
1543 | cork.flags = 0; |
1544 | cork.addr = 0; | |
70652728 | 1545 | cork.opt = NULL; |
1c32c5ad HX |
1546 | err = ip_setup_cork(sk, &cork, ipc, rtp); |
1547 | if (err) | |
1548 | return ERR_PTR(err); | |
1549 | ||
5640f768 ED |
1550 | err = __ip_append_data(sk, fl4, &queue, &cork, |
1551 | ¤t->task_frag, getfrag, | |
1c32c5ad HX |
1552 | from, length, transhdrlen, flags); |
1553 | if (err) { | |
1554 | __ip_flush_pending_frames(sk, &queue, &cork); | |
1555 | return ERR_PTR(err); | |
1556 | } | |
1557 | ||
77968b78 | 1558 | return __ip_make_skb(sk, fl4, &queue, &cork); |
1c32c5ad | 1559 | } |
1da177e4 LT |
1560 | |
1561 | /* | |
1562 | * Fetch data from kernel space and fill in checksum if needed. | |
1563 | */ | |
e905a9ed | 1564 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1565 | int len, int odd, struct sk_buff *skb) |
1566 | { | |
5084205f | 1567 | __wsum csum; |
1da177e4 LT |
1568 | |
1569 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1570 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1571 | return 0; |
1da177e4 LT |
1572 | } |
1573 | ||
e905a9ed | 1574 | /* |
1da177e4 | 1575 | * Generic function to send a packet as reply to another packet. |
be9f4a44 | 1576 | * Used to send some TCP resets/acks so far. |
1da177e4 | 1577 | */ |
bdbbb852 | 1578 | void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, |
24a2d43d ED |
1579 | const struct ip_options *sopt, |
1580 | __be32 daddr, __be32 saddr, | |
1581 | const struct ip_reply_arg *arg, | |
70e73416 | 1582 | unsigned int len) |
1da177e4 | 1583 | { |
f6d8bd05 | 1584 | struct ip_options_data replyopts; |
1da177e4 | 1585 | struct ipcm_cookie ipc; |
77968b78 | 1586 | struct flowi4 fl4; |
511c3f92 | 1587 | struct rtable *rt = skb_rtable(skb); |
bdbbb852 | 1588 | struct net *net = sock_net(sk); |
be9f4a44 | 1589 | struct sk_buff *nskb; |
4062090e | 1590 | int err; |
f7ba868b | 1591 | int oif; |
1da177e4 | 1592 | |
24a2d43d | 1593 | if (__ip_options_echo(&replyopts.opt.opt, skb, sopt)) |
1da177e4 LT |
1594 | return; |
1595 | ||
0a5ebb80 | 1596 | ipc.addr = daddr; |
1da177e4 | 1597 | ipc.opt = NULL; |
2244d07b | 1598 | ipc.tx_flags = 0; |
aa661581 FF |
1599 | ipc.ttl = 0; |
1600 | ipc.tos = -1; | |
1da177e4 | 1601 | |
f6d8bd05 | 1602 | if (replyopts.opt.opt.optlen) { |
1da177e4 LT |
1603 | ipc.opt = &replyopts.opt; |
1604 | ||
f6d8bd05 ED |
1605 | if (replyopts.opt.opt.srr) |
1606 | daddr = replyopts.opt.opt.faddr; | |
1da177e4 LT |
1607 | } |
1608 | ||
f7ba868b | 1609 | oif = arg->bound_dev_if; |
9b6c14d5 DA |
1610 | if (!oif && netif_index_is_l3_master(net, skb->skb_iif)) |
1611 | oif = skb->skb_iif; | |
f7ba868b DA |
1612 | |
1613 | flowi4_init_output(&fl4, oif, | |
e110861f | 1614 | IP4_REPLY_MARK(net, skb->mark), |
66b13d99 | 1615 | RT_TOS(arg->tos), |
be9f4a44 | 1616 | RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol, |
77968b78 | 1617 | ip_reply_arg_flowi_flags(arg), |
70e73416 | 1618 | daddr, saddr, |
e2d118a1 LC |
1619 | tcp_hdr(skb)->source, tcp_hdr(skb)->dest, |
1620 | arg->uid); | |
77968b78 | 1621 | security_skb_classify_flow(skb, flowi4_to_flowi(&fl4)); |
be9f4a44 | 1622 | rt = ip_route_output_key(net, &fl4); |
77968b78 DM |
1623 | if (IS_ERR(rt)) |
1624 | return; | |
1da177e4 | 1625 | |
bdbbb852 | 1626 | inet_sk(sk)->tos = arg->tos; |
1da177e4 | 1627 | |
1da177e4 | 1628 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1629 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1630 | sk->sk_bound_dev_if = arg->bound_dev_if; |
be9f4a44 | 1631 | sk->sk_sndbuf = sysctl_wmem_default; |
4062090e VA |
1632 | err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, |
1633 | len, 0, &ipc, &rt, MSG_DONTWAIT); | |
1634 | if (unlikely(err)) { | |
1635 | ip_flush_pending_frames(sk); | |
1636 | goto out; | |
1637 | } | |
1638 | ||
be9f4a44 ED |
1639 | nskb = skb_peek(&sk->sk_write_queue); |
1640 | if (nskb) { | |
1da177e4 | 1641 | if (arg->csumoffset >= 0) |
be9f4a44 ED |
1642 | *((__sum16 *)skb_transport_header(nskb) + |
1643 | arg->csumoffset) = csum_fold(csum_add(nskb->csum, | |
9c70220b | 1644 | arg->csum)); |
be9f4a44 | 1645 | nskb->ip_summed = CHECKSUM_NONE; |
77968b78 | 1646 | ip_push_pending_frames(sk, &fl4); |
1da177e4 | 1647 | } |
4062090e | 1648 | out: |
1da177e4 LT |
1649 | ip_rt_put(rt); |
1650 | } | |
1651 | ||
1da177e4 LT |
1652 | void __init ip_init(void) |
1653 | { | |
1da177e4 LT |
1654 | ip_rt_init(); |
1655 | inet_initpeers(); | |
1656 | ||
72c1d3bd WC |
1657 | #if defined(CONFIG_IP_MULTICAST) |
1658 | igmp_mc_init(); | |
1da177e4 LT |
1659 | #endif |
1660 | } |