Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * The Internet Protocol (IP) output module. | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Donald Becker, <becker@super.org> | |
11 | * Alan Cox, <Alan.Cox@linux.org> | |
12 | * Richard Underwood | |
13 | * Stefan Becker, <stefanb@yello.ping.de> | |
14 | * Jorge Cwik, <jorge@laser.satlink.net> | |
15 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
16 | * Hirokazu Takahashi, <taka@valinux.co.jp> | |
17 | * | |
18 | * See ip_input.c for original log | |
19 | * | |
20 | * Fixes: | |
21 | * Alan Cox : Missing nonblock feature in ip_build_xmit. | |
22 | * Mike Kilburn : htons() missing in ip_build_xmit. | |
e905a9ed | 23 | * Bradford Johnson: Fix faulty handling of some frames when |
1da177e4 LT |
24 | * no route is found. |
25 | * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit | |
26 | * (in case if packet not accepted by | |
27 | * output firewall rules) | |
28 | * Mike McLagan : Routing by source | |
29 | * Alexey Kuznetsov: use new route cache | |
30 | * Andi Kleen: Fix broken PMTU recovery and remove | |
31 | * some redundant tests. | |
32 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. | |
33 | * Andi Kleen : Replace ip_reply with ip_send_reply. | |
e905a9ed YH |
34 | * Andi Kleen : Split fast and slow ip_build_xmit path |
35 | * for decreased register pressure on x86 | |
36 | * and more readibility. | |
1da177e4 LT |
37 | * Marc Boucher : When call_out_firewall returns FW_QUEUE, |
38 | * silently drop skb instead of failing with -EPERM. | |
39 | * Detlev Wengorz : Copy protocol for fragments. | |
40 | * Hirokazu Takahashi: HW checksumming for outgoing UDP | |
41 | * datagrams. | |
42 | * Hirokazu Takahashi: sendfile() on UDP works now. | |
43 | */ | |
44 | ||
45 | #include <asm/uaccess.h> | |
46 | #include <asm/system.h> | |
47 | #include <linux/module.h> | |
48 | #include <linux/types.h> | |
49 | #include <linux/kernel.h> | |
1da177e4 LT |
50 | #include <linux/mm.h> |
51 | #include <linux/string.h> | |
52 | #include <linux/errno.h> | |
a1f8e7f7 | 53 | #include <linux/highmem.h> |
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; |
1da177e4 LT |
84 | |
85 | /* Generate a checksum for an outgoing IP datagram. */ | |
86 | __inline__ void ip_send_check(struct iphdr *iph) | |
87 | { | |
88 | iph->check = 0; | |
89 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
90 | } | |
91 | ||
c439cb2e HX |
92 | int __ip_local_out(struct sk_buff *skb) |
93 | { | |
94 | struct iphdr *iph = ip_hdr(skb); | |
95 | ||
96 | iph->tot_len = htons(skb->len); | |
97 | ip_send_check(iph); | |
adf30907 | 98 | return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev, |
c439cb2e HX |
99 | dst_output); |
100 | } | |
101 | ||
102 | int ip_local_out(struct sk_buff *skb) | |
103 | { | |
104 | int err; | |
105 | ||
106 | err = __ip_local_out(skb); | |
107 | if (likely(err == 1)) | |
108 | err = dst_output(skb); | |
109 | ||
110 | return err; | |
111 | } | |
112 | EXPORT_SYMBOL_GPL(ip_local_out); | |
113 | ||
1da177e4 LT |
114 | /* dev_loopback_xmit for use with netfilter. */ |
115 | static int ip_dev_loopback_xmit(struct sk_buff *newskb) | |
116 | { | |
459a98ed | 117 | skb_reset_mac_header(newskb); |
bbe735e4 | 118 | __skb_pull(newskb, skb_network_offset(newskb)); |
1da177e4 LT |
119 | newskb->pkt_type = PACKET_LOOPBACK; |
120 | newskb->ip_summed = CHECKSUM_UNNECESSARY; | |
adf30907 | 121 | WARN_ON(!skb_dst(newskb)); |
1da177e4 LT |
122 | netif_rx(newskb); |
123 | return 0; | |
124 | } | |
125 | ||
126 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) | |
127 | { | |
128 | int ttl = inet->uc_ttl; | |
129 | ||
130 | if (ttl < 0) | |
131 | ttl = dst_metric(dst, RTAX_HOPLIMIT); | |
132 | return ttl; | |
133 | } | |
134 | ||
e905a9ed | 135 | /* |
1da177e4 LT |
136 | * Add an ip header to a skbuff and send it out. |
137 | * | |
138 | */ | |
139 | int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, | |
13d8eaa0 | 140 | __be32 saddr, __be32 daddr, struct ip_options *opt) |
1da177e4 LT |
141 | { |
142 | struct inet_sock *inet = inet_sk(sk); | |
511c3f92 | 143 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
144 | struct iphdr *iph; |
145 | ||
146 | /* Build the IP header. */ | |
8856dfa3 ACM |
147 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0)); |
148 | skb_reset_network_header(skb); | |
eddc9ec5 | 149 | iph = ip_hdr(skb); |
1da177e4 LT |
150 | iph->version = 4; |
151 | iph->ihl = 5; | |
152 | iph->tos = inet->tos; | |
153 | if (ip_dont_fragment(sk, &rt->u.dst)) | |
154 | iph->frag_off = htons(IP_DF); | |
155 | else | |
156 | iph->frag_off = 0; | |
157 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
158 | iph->daddr = rt->rt_dst; | |
159 | iph->saddr = rt->rt_src; | |
160 | iph->protocol = sk->sk_protocol; | |
1da177e4 | 161 | ip_select_ident(iph, &rt->u.dst, sk); |
1da177e4 LT |
162 | |
163 | if (opt && opt->optlen) { | |
164 | iph->ihl += opt->optlen>>2; | |
165 | ip_options_build(skb, opt, daddr, rt, 0); | |
166 | } | |
1da177e4 LT |
167 | |
168 | skb->priority = sk->sk_priority; | |
4a19ec58 | 169 | skb->mark = sk->sk_mark; |
1da177e4 LT |
170 | |
171 | /* Send it out. */ | |
c439cb2e | 172 | return ip_local_out(skb); |
1da177e4 LT |
173 | } |
174 | ||
d8c97a94 ACM |
175 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
176 | ||
1da177e4 LT |
177 | static inline int ip_finish_output2(struct sk_buff *skb) |
178 | { | |
adf30907 | 179 | struct dst_entry *dst = skb_dst(skb); |
80787ebc | 180 | struct rtable *rt = (struct rtable *)dst; |
1da177e4 | 181 | struct net_device *dev = dst->dev; |
c2636b4d | 182 | unsigned int hh_len = LL_RESERVED_SPACE(dev); |
1da177e4 | 183 | |
edf391ff NH |
184 | if (rt->rt_type == RTN_MULTICAST) { |
185 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len); | |
186 | } else if (rt->rt_type == RTN_BROADCAST) | |
187 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len); | |
80787ebc | 188 | |
1da177e4 | 189 | /* Be paranoid, rather than too clever. */ |
3b04ddde | 190 | if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { |
1da177e4 LT |
191 | struct sk_buff *skb2; |
192 | ||
193 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
194 | if (skb2 == NULL) { | |
195 | kfree_skb(skb); | |
196 | return -ENOMEM; | |
197 | } | |
198 | if (skb->sk) | |
199 | skb_set_owner_w(skb2, skb->sk); | |
200 | kfree_skb(skb); | |
201 | skb = skb2; | |
202 | } | |
203 | ||
3644f0ce SH |
204 | if (dst->hh) |
205 | return neigh_hh_output(dst->hh, skb); | |
206 | else if (dst->neighbour) | |
1da177e4 LT |
207 | return dst->neighbour->output(skb); |
208 | ||
209 | if (net_ratelimit()) | |
210 | printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n"); | |
211 | kfree_skb(skb); | |
212 | return -EINVAL; | |
213 | } | |
214 | ||
628a5c56 JH |
215 | static inline int ip_skb_dst_mtu(struct sk_buff *skb) |
216 | { | |
217 | struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL; | |
218 | ||
219 | return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ? | |
adf30907 | 220 | skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb)); |
628a5c56 JH |
221 | } |
222 | ||
861d0486 | 223 | static int ip_finish_output(struct sk_buff *skb) |
1da177e4 | 224 | { |
5c901daa PM |
225 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
226 | /* Policy lookup after SNAT yielded a new policy */ | |
adf30907 | 227 | if (skb_dst(skb)->xfrm != NULL) { |
48d5cad8 PM |
228 | IPCB(skb)->flags |= IPSKB_REROUTED; |
229 | return dst_output(skb); | |
230 | } | |
5c901daa | 231 | #endif |
628a5c56 | 232 | if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb)) |
1bd9bef6 PM |
233 | return ip_fragment(skb, ip_finish_output2); |
234 | else | |
235 | return ip_finish_output2(skb); | |
1da177e4 LT |
236 | } |
237 | ||
238 | int ip_mc_output(struct sk_buff *skb) | |
239 | { | |
240 | struct sock *sk = skb->sk; | |
511c3f92 | 241 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
242 | struct net_device *dev = rt->u.dst.dev; |
243 | ||
244 | /* | |
245 | * If the indicated interface is up and running, send the packet. | |
246 | */ | |
edf391ff | 247 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 LT |
248 | |
249 | skb->dev = dev; | |
250 | skb->protocol = htons(ETH_P_IP); | |
251 | ||
252 | /* | |
253 | * Multicasts are looped back for other local users | |
254 | */ | |
255 | ||
256 | if (rt->rt_flags&RTCF_MULTICAST) { | |
257 | if ((!sk || inet_sk(sk)->mc_loop) | |
258 | #ifdef CONFIG_IP_MROUTE | |
259 | /* Small optimization: do not loopback not local frames, | |
260 | which returned after forwarding; they will be dropped | |
261 | by ip_mr_input in any case. | |
262 | Note, that local frames are looped back to be delivered | |
263 | to local recipients. | |
264 | ||
265 | This check is duplicated in ip_mr_input at the moment. | |
266 | */ | |
9d4fb27d JP |
267 | && |
268 | ((rt->rt_flags & RTCF_LOCAL) || | |
269 | !(IPCB(skb)->flags & IPSKB_FORWARDED)) | |
1da177e4 | 270 | #endif |
9d4fb27d | 271 | ) { |
1da177e4 LT |
272 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); |
273 | if (newskb) | |
6e23ae2a PM |
274 | NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, |
275 | NULL, newskb->dev, | |
1da177e4 LT |
276 | ip_dev_loopback_xmit); |
277 | } | |
278 | ||
279 | /* Multicasts with ttl 0 must not go beyond the host */ | |
280 | ||
eddc9ec5 | 281 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
282 | kfree_skb(skb); |
283 | return 0; | |
284 | } | |
285 | } | |
286 | ||
287 | if (rt->rt_flags&RTCF_BROADCAST) { | |
288 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
289 | if (newskb) | |
6e23ae2a | 290 | NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL, |
1da177e4 LT |
291 | newskb->dev, ip_dev_loopback_xmit); |
292 | } | |
293 | ||
6e23ae2a | 294 | return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev, |
48d5cad8 PM |
295 | ip_finish_output, |
296 | !(IPCB(skb)->flags & IPSKB_REROUTED)); | |
1da177e4 LT |
297 | } |
298 | ||
299 | int ip_output(struct sk_buff *skb) | |
300 | { | |
adf30907 | 301 | struct net_device *dev = skb_dst(skb)->dev; |
1bd9bef6 | 302 | |
edf391ff | 303 | IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len); |
1da177e4 | 304 | |
1bd9bef6 PM |
305 | skb->dev = dev; |
306 | skb->protocol = htons(ETH_P_IP); | |
307 | ||
6e23ae2a | 308 | return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev, |
e905a9ed | 309 | ip_finish_output, |
48d5cad8 | 310 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
311 | } |
312 | ||
e89862f4 | 313 | int ip_queue_xmit(struct sk_buff *skb, int ipfragok) |
1da177e4 | 314 | { |
e89862f4 | 315 | struct sock *sk = skb->sk; |
1da177e4 LT |
316 | struct inet_sock *inet = inet_sk(sk); |
317 | struct ip_options *opt = inet->opt; | |
318 | struct rtable *rt; | |
319 | struct iphdr *iph; | |
320 | ||
321 | /* Skip all of this if the packet is already routed, | |
322 | * f.e. by something like SCTP. | |
323 | */ | |
511c3f92 | 324 | rt = skb_rtable(skb); |
1da177e4 LT |
325 | if (rt != NULL) |
326 | goto packet_routed; | |
327 | ||
328 | /* Make sure we can route this packet. */ | |
329 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
330 | if (rt == NULL) { | |
3ca3c68e | 331 | __be32 daddr; |
1da177e4 LT |
332 | |
333 | /* Use correct destination address if we have options. */ | |
c720c7e8 | 334 | daddr = inet->inet_daddr; |
1da177e4 LT |
335 | if(opt && opt->srr) |
336 | daddr = opt->faddr; | |
337 | ||
338 | { | |
339 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
914a9ab3 | 340 | .mark = sk->sk_mark, |
1da177e4 LT |
341 | .nl_u = { .ip4_u = |
342 | { .daddr = daddr, | |
c720c7e8 | 343 | .saddr = inet->inet_saddr, |
1da177e4 LT |
344 | .tos = RT_CONN_FLAGS(sk) } }, |
345 | .proto = sk->sk_protocol, | |
86b08d86 | 346 | .flags = inet_sk_flowi_flags(sk), |
1da177e4 | 347 | .uli_u = { .ports = |
c720c7e8 ED |
348 | { .sport = inet->inet_sport, |
349 | .dport = inet->inet_dport } } }; | |
1da177e4 LT |
350 | |
351 | /* If this fails, retransmit mechanism of transport layer will | |
352 | * keep trying until route appears or the connection times | |
353 | * itself out. | |
354 | */ | |
beb8d13b | 355 | security_sk_classify_flow(sk, &fl); |
3b1e0a65 | 356 | if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0)) |
1da177e4 LT |
357 | goto no_route; |
358 | } | |
6cbb0df7 | 359 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 | 360 | } |
adf30907 | 361 | skb_dst_set(skb, dst_clone(&rt->u.dst)); |
1da177e4 LT |
362 | |
363 | packet_routed: | |
364 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) | |
365 | goto no_route; | |
366 | ||
367 | /* OK, we know where to send it, allocate and build IP header. */ | |
8856dfa3 ACM |
368 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0)); |
369 | skb_reset_network_header(skb); | |
eddc9ec5 | 370 | iph = ip_hdr(skb); |
714e85be | 371 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
1da177e4 LT |
372 | if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok) |
373 | iph->frag_off = htons(IP_DF); | |
374 | else | |
375 | iph->frag_off = 0; | |
376 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
377 | iph->protocol = sk->sk_protocol; | |
378 | iph->saddr = rt->rt_src; | |
379 | iph->daddr = rt->rt_dst; | |
1da177e4 LT |
380 | /* Transport layer set skb->h.foo itself. */ |
381 | ||
382 | if (opt && opt->optlen) { | |
383 | iph->ihl += opt->optlen >> 2; | |
c720c7e8 | 384 | ip_options_build(skb, opt, inet->inet_daddr, rt, 0); |
1da177e4 LT |
385 | } |
386 | ||
89f5f0ae | 387 | ip_select_ident_more(iph, &rt->u.dst, sk, |
7967168c | 388 | (skb_shinfo(skb)->gso_segs ?: 1) - 1); |
1da177e4 | 389 | |
1da177e4 | 390 | skb->priority = sk->sk_priority; |
4a19ec58 | 391 | skb->mark = sk->sk_mark; |
1da177e4 | 392 | |
c439cb2e | 393 | return ip_local_out(skb); |
1da177e4 LT |
394 | |
395 | no_route: | |
5e38e270 | 396 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
397 | kfree_skb(skb); |
398 | return -EHOSTUNREACH; | |
399 | } | |
400 | ||
401 | ||
402 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
403 | { | |
404 | to->pkt_type = from->pkt_type; | |
405 | to->priority = from->priority; | |
406 | to->protocol = from->protocol; | |
adf30907 ED |
407 | skb_dst_drop(to); |
408 | skb_dst_set(to, dst_clone(skb_dst(from))); | |
1da177e4 | 409 | to->dev = from->dev; |
82e91ffe | 410 | to->mark = from->mark; |
1da177e4 LT |
411 | |
412 | /* Copy the flags to each fragment. */ | |
413 | IPCB(to)->flags = IPCB(from)->flags; | |
414 | ||
415 | #ifdef CONFIG_NET_SCHED | |
416 | to->tc_index = from->tc_index; | |
417 | #endif | |
e7ac05f3 | 418 | nf_copy(to, from); |
ba9dda3a JK |
419 | #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ |
420 | defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) | |
421 | to->nf_trace = from->nf_trace; | |
422 | #endif | |
c98d80ed JA |
423 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
424 | to->ipvs_property = from->ipvs_property; | |
1da177e4 | 425 | #endif |
984bc16c | 426 | skb_copy_secmark(to, from); |
1da177e4 LT |
427 | } |
428 | ||
429 | /* | |
430 | * This IP datagram is too large to be sent in one piece. Break it up into | |
431 | * smaller pieces (each of size equal to IP header plus | |
432 | * a block of the data of the original IP data part) that will yet fit in a | |
433 | * single device frame, and queue such a frame for sending. | |
434 | */ | |
435 | ||
d9319100 | 436 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)) |
1da177e4 LT |
437 | { |
438 | struct iphdr *iph; | |
439 | int raw = 0; | |
440 | int ptr; | |
441 | struct net_device *dev; | |
442 | struct sk_buff *skb2; | |
9bcfcaf5 | 443 | unsigned int mtu, hlen, left, len, ll_rs, pad; |
1da177e4 | 444 | int offset; |
76ab608d | 445 | __be16 not_last_frag; |
511c3f92 | 446 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
447 | int err = 0; |
448 | ||
449 | dev = rt->u.dst.dev; | |
450 | ||
451 | /* | |
452 | * Point into the IP datagram header. | |
453 | */ | |
454 | ||
eddc9ec5 | 455 | iph = ip_hdr(skb); |
1da177e4 LT |
456 | |
457 | if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { | |
5e38e270 | 458 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 | 459 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
628a5c56 | 460 | htonl(ip_skb_dst_mtu(skb))); |
1da177e4 LT |
461 | kfree_skb(skb); |
462 | return -EMSGSIZE; | |
463 | } | |
464 | ||
465 | /* | |
466 | * Setup starting values. | |
467 | */ | |
468 | ||
469 | hlen = iph->ihl * 4; | |
470 | mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */ | |
89cee8b1 | 471 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
472 | |
473 | /* When frag_list is given, use it. First, check its validity: | |
474 | * some transformers could create wrong frag_list or break existing | |
475 | * one, it is not prohibited. In this case fall back to copying. | |
476 | * | |
477 | * LATER: this step can be merged to real generation of fragments, | |
478 | * we can switch to copy when see the first bad fragment. | |
479 | */ | |
d7fcf1a5 | 480 | if (skb_has_frags(skb)) { |
1da177e4 LT |
481 | struct sk_buff *frag; |
482 | int first_len = skb_pagelen(skb); | |
29ffe1a5 | 483 | int truesizes = 0; |
1da177e4 LT |
484 | |
485 | if (first_len - hlen > mtu || | |
486 | ((first_len - hlen) & 7) || | |
487 | (iph->frag_off & htons(IP_MF|IP_OFFSET)) || | |
488 | skb_cloned(skb)) | |
489 | goto slow_path; | |
490 | ||
d7fcf1a5 | 491 | skb_walk_frags(skb, frag) { |
1da177e4 LT |
492 | /* Correct geometry. */ |
493 | if (frag->len > mtu || | |
494 | ((frag->len & 7) && frag->next) || | |
495 | skb_headroom(frag) < hlen) | |
496 | goto slow_path; | |
497 | ||
498 | /* Partially cloned skb? */ | |
499 | if (skb_shared(frag)) | |
500 | goto slow_path; | |
2fdba6b0 HX |
501 | |
502 | BUG_ON(frag->sk); | |
503 | if (skb->sk) { | |
2fdba6b0 HX |
504 | frag->sk = skb->sk; |
505 | frag->destructor = sock_wfree; | |
29ffe1a5 | 506 | truesizes += frag->truesize; |
2fdba6b0 | 507 | } |
1da177e4 LT |
508 | } |
509 | ||
510 | /* Everything is OK. Generate! */ | |
511 | ||
512 | err = 0; | |
513 | offset = 0; | |
514 | frag = skb_shinfo(skb)->frag_list; | |
d7fcf1a5 | 515 | skb_frag_list_init(skb); |
1da177e4 | 516 | skb->data_len = first_len - skb_headlen(skb); |
29ffe1a5 | 517 | skb->truesize -= truesizes; |
1da177e4 LT |
518 | skb->len = first_len; |
519 | iph->tot_len = htons(first_len); | |
520 | iph->frag_off = htons(IP_MF); | |
521 | ip_send_check(iph); | |
522 | ||
523 | for (;;) { | |
524 | /* Prepare header of the next frame, | |
525 | * before previous one went down. */ | |
526 | if (frag) { | |
527 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 528 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
529 | __skb_push(frag, hlen); |
530 | skb_reset_network_header(frag); | |
d56f90a7 | 531 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 532 | iph = ip_hdr(frag); |
1da177e4 LT |
533 | iph->tot_len = htons(frag->len); |
534 | ip_copy_metadata(frag, skb); | |
535 | if (offset == 0) | |
536 | ip_options_fragment(frag); | |
537 | offset += skb->len - hlen; | |
538 | iph->frag_off = htons(offset>>3); | |
539 | if (frag->next != NULL) | |
540 | iph->frag_off |= htons(IP_MF); | |
541 | /* Ready, complete checksum */ | |
542 | ip_send_check(iph); | |
543 | } | |
544 | ||
545 | err = output(skb); | |
546 | ||
dafee490 | 547 | if (!err) |
5e38e270 | 548 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
549 | if (err || !frag) |
550 | break; | |
551 | ||
552 | skb = frag; | |
553 | frag = skb->next; | |
554 | skb->next = NULL; | |
555 | } | |
556 | ||
557 | if (err == 0) { | |
5e38e270 | 558 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
559 | return 0; |
560 | } | |
561 | ||
562 | while (frag) { | |
563 | skb = frag->next; | |
564 | kfree_skb(frag); | |
565 | frag = skb; | |
566 | } | |
5e38e270 | 567 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
568 | return err; |
569 | } | |
570 | ||
571 | slow_path: | |
572 | left = skb->len - hlen; /* Space per frame */ | |
573 | ptr = raw + hlen; /* Where to start from */ | |
574 | ||
1da177e4 | 575 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, |
9bcfcaf5 SH |
576 | * we need to make room for the encapsulating header |
577 | */ | |
578 | pad = nf_bridge_pad(skb); | |
579 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad); | |
580 | mtu -= pad; | |
581 | ||
1da177e4 LT |
582 | /* |
583 | * Fragment the datagram. | |
584 | */ | |
585 | ||
586 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
587 | not_last_frag = iph->frag_off & htons(IP_MF); | |
588 | ||
589 | /* | |
590 | * Keep copying data until we run out. | |
591 | */ | |
592 | ||
132adf54 | 593 | while (left > 0) { |
1da177e4 LT |
594 | len = left; |
595 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
596 | if (len > mtu) | |
597 | len = mtu; | |
598 | /* IF: we are not sending upto and including the packet end | |
599 | then align the next start on an eight byte boundary */ | |
600 | if (len < left) { | |
601 | len &= ~7; | |
602 | } | |
603 | /* | |
604 | * Allocate buffer. | |
605 | */ | |
606 | ||
607 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 608 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
609 | err = -ENOMEM; |
610 | goto fail; | |
611 | } | |
612 | ||
613 | /* | |
614 | * Set up data on packet | |
615 | */ | |
616 | ||
617 | ip_copy_metadata(skb2, skb); | |
618 | skb_reserve(skb2, ll_rs); | |
619 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 620 | skb_reset_network_header(skb2); |
b0e380b1 | 621 | skb2->transport_header = skb2->network_header + hlen; |
1da177e4 LT |
622 | |
623 | /* | |
624 | * Charge the memory for the fragment to any owner | |
625 | * it might possess | |
626 | */ | |
627 | ||
628 | if (skb->sk) | |
629 | skb_set_owner_w(skb2, skb->sk); | |
630 | ||
631 | /* | |
632 | * Copy the packet header into the new buffer. | |
633 | */ | |
634 | ||
d626f62b | 635 | skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen); |
1da177e4 LT |
636 | |
637 | /* | |
638 | * Copy a block of the IP datagram. | |
639 | */ | |
bff9b61c | 640 | if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len)) |
1da177e4 LT |
641 | BUG(); |
642 | left -= len; | |
643 | ||
644 | /* | |
645 | * Fill in the new header fields. | |
646 | */ | |
eddc9ec5 | 647 | iph = ip_hdr(skb2); |
1da177e4 LT |
648 | iph->frag_off = htons((offset >> 3)); |
649 | ||
650 | /* ANK: dirty, but effective trick. Upgrade options only if | |
651 | * the segment to be fragmented was THE FIRST (otherwise, | |
652 | * options are already fixed) and make it ONCE | |
653 | * on the initial skb, so that all the following fragments | |
654 | * will inherit fixed options. | |
655 | */ | |
656 | if (offset == 0) | |
657 | ip_options_fragment(skb); | |
658 | ||
659 | /* | |
660 | * Added AC : If we are fragmenting a fragment that's not the | |
661 | * last fragment then keep MF on each bit | |
662 | */ | |
663 | if (left > 0 || not_last_frag) | |
664 | iph->frag_off |= htons(IP_MF); | |
665 | ptr += len; | |
666 | offset += len; | |
667 | ||
668 | /* | |
669 | * Put this fragment into the sending queue. | |
670 | */ | |
1da177e4 LT |
671 | iph->tot_len = htons(len + hlen); |
672 | ||
673 | ip_send_check(iph); | |
674 | ||
675 | err = output(skb2); | |
676 | if (err) | |
677 | goto fail; | |
dafee490 | 678 | |
5e38e270 | 679 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES); |
1da177e4 LT |
680 | } |
681 | kfree_skb(skb); | |
5e38e270 | 682 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS); |
1da177e4 LT |
683 | return err; |
684 | ||
685 | fail: | |
e905a9ed | 686 | kfree_skb(skb); |
5e38e270 | 687 | IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
688 | return err; |
689 | } | |
690 | ||
2e2f7aef PM |
691 | EXPORT_SYMBOL(ip_fragment); |
692 | ||
1da177e4 LT |
693 | int |
694 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
695 | { | |
696 | struct iovec *iov = from; | |
697 | ||
84fa7933 | 698 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1da177e4 LT |
699 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) |
700 | return -EFAULT; | |
701 | } else { | |
44bb9363 | 702 | __wsum csum = 0; |
1da177e4 LT |
703 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) |
704 | return -EFAULT; | |
705 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
706 | } | |
707 | return 0; | |
708 | } | |
709 | ||
44bb9363 | 710 | static inline __wsum |
1da177e4 LT |
711 | csum_page(struct page *page, int offset, int copy) |
712 | { | |
713 | char *kaddr; | |
44bb9363 | 714 | __wsum csum; |
1da177e4 LT |
715 | kaddr = kmap(page); |
716 | csum = csum_partial(kaddr + offset, copy, 0); | |
717 | kunmap(page); | |
718 | return csum; | |
719 | } | |
720 | ||
4b30b1c6 | 721 | static inline int ip_ufo_append_data(struct sock *sk, |
e89e9cf5 AR |
722 | int getfrag(void *from, char *to, int offset, int len, |
723 | int odd, struct sk_buff *skb), | |
724 | void *from, int length, int hh_len, int fragheaderlen, | |
d9319100 | 725 | int transhdrlen, int mtu, unsigned int flags) |
e89e9cf5 AR |
726 | { |
727 | struct sk_buff *skb; | |
728 | int err; | |
729 | ||
730 | /* There is support for UDP fragmentation offload by network | |
731 | * device, so create one single skb packet containing complete | |
732 | * udp datagram | |
733 | */ | |
734 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) { | |
735 | skb = sock_alloc_send_skb(sk, | |
736 | hh_len + fragheaderlen + transhdrlen + 20, | |
737 | (flags & MSG_DONTWAIT), &err); | |
738 | ||
739 | if (skb == NULL) | |
740 | return err; | |
741 | ||
742 | /* reserve space for Hardware header */ | |
743 | skb_reserve(skb, hh_len); | |
744 | ||
745 | /* create space for UDP/IP header */ | |
d9319100 | 746 | skb_put(skb, fragheaderlen + transhdrlen); |
e89e9cf5 AR |
747 | |
748 | /* initialize network header pointer */ | |
c1d2bbe1 | 749 | skb_reset_network_header(skb); |
e89e9cf5 AR |
750 | |
751 | /* initialize protocol header pointer */ | |
b0e380b1 | 752 | skb->transport_header = skb->network_header + fragheaderlen; |
e89e9cf5 | 753 | |
84fa7933 | 754 | skb->ip_summed = CHECKSUM_PARTIAL; |
e89e9cf5 AR |
755 | skb->csum = 0; |
756 | sk->sk_sndmsg_off = 0; | |
e89e9cf5 | 757 | |
be9164e7 | 758 | /* specify the length of each IP datagram fragment */ |
7967168c | 759 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 760 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
e89e9cf5 | 761 | __skb_queue_tail(&sk->sk_write_queue, skb); |
e89e9cf5 | 762 | } |
be9164e7 K |
763 | |
764 | return skb_append_datato_frags(sk, skb, getfrag, from, | |
765 | (length - transhdrlen)); | |
e89e9cf5 AR |
766 | } |
767 | ||
1da177e4 LT |
768 | /* |
769 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
770 | * from many pieces of data. Each pieces will be holded on the socket | |
771 | * until ip_push_pending_frames() is called. Each piece can be a page | |
772 | * or non-page data. | |
e905a9ed | 773 | * |
1da177e4 LT |
774 | * Not only UDP, other transport protocols - e.g. raw sockets - can use |
775 | * this interface potentially. | |
776 | * | |
777 | * LATER: length must be adjusted by pad at tail, when it is required. | |
778 | */ | |
779 | int ip_append_data(struct sock *sk, | |
780 | int getfrag(void *from, char *to, int offset, int len, | |
781 | int odd, struct sk_buff *skb), | |
782 | void *from, int length, int transhdrlen, | |
2e77d89b | 783 | struct ipcm_cookie *ipc, struct rtable **rtp, |
1da177e4 LT |
784 | unsigned int flags) |
785 | { | |
786 | struct inet_sock *inet = inet_sk(sk); | |
787 | struct sk_buff *skb; | |
788 | ||
789 | struct ip_options *opt = NULL; | |
790 | int hh_len; | |
791 | int exthdrlen; | |
792 | int mtu; | |
793 | int copy; | |
794 | int err; | |
795 | int offset = 0; | |
796 | unsigned int maxfraglen, fragheaderlen; | |
797 | int csummode = CHECKSUM_NONE; | |
2e77d89b | 798 | struct rtable *rt; |
1da177e4 LT |
799 | |
800 | if (flags&MSG_PROBE) | |
801 | return 0; | |
802 | ||
803 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
804 | /* | |
805 | * setup for corking. | |
806 | */ | |
807 | opt = ipc->opt; | |
808 | if (opt) { | |
809 | if (inet->cork.opt == NULL) { | |
810 | inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation); | |
811 | if (unlikely(inet->cork.opt == NULL)) | |
812 | return -ENOBUFS; | |
813 | } | |
814 | memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen); | |
815 | inet->cork.flags |= IPCORK_OPT; | |
816 | inet->cork.addr = ipc->addr; | |
817 | } | |
2e77d89b | 818 | rt = *rtp; |
788d908f JT |
819 | if (unlikely(!rt)) |
820 | return -EFAULT; | |
2e77d89b ED |
821 | /* |
822 | * We steal reference to this route, caller should not release it | |
823 | */ | |
824 | *rtp = NULL; | |
628a5c56 JH |
825 | inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ? |
826 | rt->u.dst.dev->mtu : | |
827 | dst_mtu(rt->u.dst.path); | |
c8cdaf99 | 828 | inet->cork.dst = &rt->u.dst; |
1da177e4 LT |
829 | inet->cork.length = 0; |
830 | sk->sk_sndmsg_page = NULL; | |
831 | sk->sk_sndmsg_off = 0; | |
832 | if ((exthdrlen = rt->u.dst.header_len) != 0) { | |
833 | length += exthdrlen; | |
834 | transhdrlen += exthdrlen; | |
835 | } | |
836 | } else { | |
c8cdaf99 | 837 | rt = (struct rtable *)inet->cork.dst; |
1da177e4 LT |
838 | if (inet->cork.flags & IPCORK_OPT) |
839 | opt = inet->cork.opt; | |
840 | ||
841 | transhdrlen = 0; | |
842 | exthdrlen = 0; | |
843 | mtu = inet->cork.fragsize; | |
844 | } | |
845 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
846 | ||
847 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
848 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
849 | ||
850 | if (inet->cork.length + length > 0xFFFF - fragheaderlen) { | |
c720c7e8 ED |
851 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->inet_dport, |
852 | mtu-exthdrlen); | |
1da177e4 LT |
853 | return -EMSGSIZE; |
854 | } | |
855 | ||
856 | /* | |
857 | * transhdrlen > 0 means that this is the first fragment and we wish | |
858 | * it won't be fragmented in the future. | |
859 | */ | |
860 | if (transhdrlen && | |
861 | length + fragheaderlen <= mtu && | |
d212f87b | 862 | rt->u.dst.dev->features & NETIF_F_V4_CSUM && |
1da177e4 | 863 | !exthdrlen) |
84fa7933 | 864 | csummode = CHECKSUM_PARTIAL; |
1da177e4 LT |
865 | |
866 | inet->cork.length += length; | |
be9164e7 K |
867 | if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) && |
868 | (sk->sk_protocol == IPPROTO_UDP) && | |
869 | (rt->u.dst.dev->features & NETIF_F_UFO)) { | |
baa829d8 PM |
870 | err = ip_ufo_append_data(sk, getfrag, from, length, hh_len, |
871 | fragheaderlen, transhdrlen, mtu, | |
872 | flags); | |
873 | if (err) | |
e89e9cf5 | 874 | goto error; |
e89e9cf5 AR |
875 | return 0; |
876 | } | |
1da177e4 LT |
877 | |
878 | /* So, what's going on in the loop below? | |
879 | * | |
880 | * We use calculated fragment length to generate chained skb, | |
881 | * each of segments is IP fragment ready for sending to network after | |
882 | * adding appropriate IP header. | |
883 | */ | |
884 | ||
885 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
886 | goto alloc_new_skb; | |
887 | ||
888 | while (length > 0) { | |
889 | /* Check if the remaining data fits into current packet. */ | |
890 | copy = mtu - skb->len; | |
891 | if (copy < length) | |
892 | copy = maxfraglen - skb->len; | |
893 | if (copy <= 0) { | |
894 | char *data; | |
895 | unsigned int datalen; | |
896 | unsigned int fraglen; | |
897 | unsigned int fraggap; | |
898 | unsigned int alloclen; | |
899 | struct sk_buff *skb_prev; | |
900 | alloc_new_skb: | |
901 | skb_prev = skb; | |
902 | if (skb_prev) | |
903 | fraggap = skb_prev->len - maxfraglen; | |
904 | else | |
905 | fraggap = 0; | |
906 | ||
907 | /* | |
908 | * If remaining data exceeds the mtu, | |
909 | * we know we need more fragment(s). | |
910 | */ | |
911 | datalen = length + fraggap; | |
912 | if (datalen > mtu - fragheaderlen) | |
913 | datalen = maxfraglen - fragheaderlen; | |
914 | fraglen = datalen + fragheaderlen; | |
915 | ||
e905a9ed | 916 | if ((flags & MSG_MORE) && |
1da177e4 LT |
917 | !(rt->u.dst.dev->features&NETIF_F_SG)) |
918 | alloclen = mtu; | |
919 | else | |
920 | alloclen = datalen + fragheaderlen; | |
921 | ||
922 | /* The last fragment gets additional space at tail. | |
923 | * Note, with MSG_MORE we overallocate on fragments, | |
924 | * because we have no idea what fragment will be | |
925 | * the last. | |
926 | */ | |
3d9dd756 | 927 | if (datalen == length + fraggap) |
1da177e4 LT |
928 | alloclen += rt->u.dst.trailer_len; |
929 | ||
930 | if (transhdrlen) { | |
e905a9ed | 931 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
932 | alloclen + hh_len + 15, |
933 | (flags & MSG_DONTWAIT), &err); | |
934 | } else { | |
935 | skb = NULL; | |
936 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
937 | 2 * sk->sk_sndbuf) | |
e905a9ed | 938 | skb = sock_wmalloc(sk, |
1da177e4 LT |
939 | alloclen + hh_len + 15, 1, |
940 | sk->sk_allocation); | |
941 | if (unlikely(skb == NULL)) | |
942 | err = -ENOBUFS; | |
51f31cab PO |
943 | else |
944 | /* only the initial fragment is | |
945 | time stamped */ | |
946 | ipc->shtx.flags = 0; | |
1da177e4 LT |
947 | } |
948 | if (skb == NULL) | |
949 | goto error; | |
950 | ||
951 | /* | |
952 | * Fill in the control structures | |
953 | */ | |
954 | skb->ip_summed = csummode; | |
955 | skb->csum = 0; | |
956 | skb_reserve(skb, hh_len); | |
51f31cab | 957 | *skb_tx(skb) = ipc->shtx; |
1da177e4 LT |
958 | |
959 | /* | |
960 | * Find where to start putting bytes. | |
961 | */ | |
962 | data = skb_put(skb, fraglen); | |
c14d2450 | 963 | skb_set_network_header(skb, exthdrlen); |
b0e380b1 ACM |
964 | skb->transport_header = (skb->network_header + |
965 | fragheaderlen); | |
1da177e4 | 966 | data += fragheaderlen; |
1da177e4 LT |
967 | |
968 | if (fraggap) { | |
969 | skb->csum = skb_copy_and_csum_bits( | |
970 | skb_prev, maxfraglen, | |
971 | data + transhdrlen, fraggap, 0); | |
972 | skb_prev->csum = csum_sub(skb_prev->csum, | |
973 | skb->csum); | |
974 | data += fraggap; | |
e9fa4f7b | 975 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
976 | } |
977 | ||
978 | copy = datalen - transhdrlen - fraggap; | |
979 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
980 | err = -EFAULT; | |
981 | kfree_skb(skb); | |
982 | goto error; | |
983 | } | |
984 | ||
985 | offset += copy; | |
986 | length -= datalen - fraggap; | |
987 | transhdrlen = 0; | |
988 | exthdrlen = 0; | |
989 | csummode = CHECKSUM_NONE; | |
990 | ||
991 | /* | |
992 | * Put the packet on the pending queue. | |
993 | */ | |
994 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
995 | continue; | |
996 | } | |
997 | ||
998 | if (copy > length) | |
999 | copy = length; | |
1000 | ||
1001 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) { | |
1002 | unsigned int off; | |
1003 | ||
1004 | off = skb->len; | |
e905a9ed | 1005 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
1006 | offset, copy, off, skb) < 0) { |
1007 | __skb_trim(skb, off); | |
1008 | err = -EFAULT; | |
1009 | goto error; | |
1010 | } | |
1011 | } else { | |
1012 | int i = skb_shinfo(skb)->nr_frags; | |
1013 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; | |
1014 | struct page *page = sk->sk_sndmsg_page; | |
1015 | int off = sk->sk_sndmsg_off; | |
1016 | unsigned int left; | |
1017 | ||
1018 | if (page && (left = PAGE_SIZE - off) > 0) { | |
1019 | if (copy >= left) | |
1020 | copy = left; | |
1021 | if (page != frag->page) { | |
1022 | if (i == MAX_SKB_FRAGS) { | |
1023 | err = -EMSGSIZE; | |
1024 | goto error; | |
1025 | } | |
1026 | get_page(page); | |
e905a9ed | 1027 | skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0); |
1da177e4 LT |
1028 | frag = &skb_shinfo(skb)->frags[i]; |
1029 | } | |
1030 | } else if (i < MAX_SKB_FRAGS) { | |
1031 | if (copy > PAGE_SIZE) | |
1032 | copy = PAGE_SIZE; | |
1033 | page = alloc_pages(sk->sk_allocation, 0); | |
1034 | if (page == NULL) { | |
1035 | err = -ENOMEM; | |
1036 | goto error; | |
1037 | } | |
1038 | sk->sk_sndmsg_page = page; | |
1039 | sk->sk_sndmsg_off = 0; | |
1040 | ||
1041 | skb_fill_page_desc(skb, i, page, 0, 0); | |
1042 | frag = &skb_shinfo(skb)->frags[i]; | |
1da177e4 LT |
1043 | } else { |
1044 | err = -EMSGSIZE; | |
1045 | goto error; | |
1046 | } | |
1047 | if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { | |
1048 | err = -EFAULT; | |
1049 | goto error; | |
1050 | } | |
1051 | sk->sk_sndmsg_off += copy; | |
1052 | frag->size += copy; | |
1053 | skb->len += copy; | |
1054 | skb->data_len += copy; | |
f945fa7a HX |
1055 | skb->truesize += copy; |
1056 | atomic_add(copy, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1057 | } |
1058 | offset += copy; | |
1059 | length -= copy; | |
1060 | } | |
1061 | ||
1062 | return 0; | |
1063 | ||
1064 | error: | |
1065 | inet->cork.length -= length; | |
5e38e270 | 1066 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
e905a9ed | 1067 | return err; |
1da177e4 LT |
1068 | } |
1069 | ||
1070 | ssize_t ip_append_page(struct sock *sk, struct page *page, | |
1071 | int offset, size_t size, int flags) | |
1072 | { | |
1073 | struct inet_sock *inet = inet_sk(sk); | |
1074 | struct sk_buff *skb; | |
1075 | struct rtable *rt; | |
1076 | struct ip_options *opt = NULL; | |
1077 | int hh_len; | |
1078 | int mtu; | |
1079 | int len; | |
1080 | int err; | |
1081 | unsigned int maxfraglen, fragheaderlen, fraggap; | |
1082 | ||
1083 | if (inet->hdrincl) | |
1084 | return -EPERM; | |
1085 | ||
1086 | if (flags&MSG_PROBE) | |
1087 | return 0; | |
1088 | ||
1089 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1090 | return -EINVAL; | |
1091 | ||
c8cdaf99 | 1092 | rt = (struct rtable *)inet->cork.dst; |
1da177e4 LT |
1093 | if (inet->cork.flags & IPCORK_OPT) |
1094 | opt = inet->cork.opt; | |
1095 | ||
1096 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) | |
1097 | return -EOPNOTSUPP; | |
1098 | ||
1099 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
1100 | mtu = inet->cork.fragsize; | |
1101 | ||
1102 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1103 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1104 | ||
1105 | if (inet->cork.length + size > 0xFFFF - fragheaderlen) { | |
c720c7e8 | 1106 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->inet_dport, mtu); |
1da177e4 LT |
1107 | return -EMSGSIZE; |
1108 | } | |
1109 | ||
1110 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1111 | return -EINVAL; | |
1112 | ||
1113 | inet->cork.length += size; | |
e89e9cf5 | 1114 | if ((sk->sk_protocol == IPPROTO_UDP) && |
7967168c HX |
1115 | (rt->u.dst.dev->features & NETIF_F_UFO)) { |
1116 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; | |
f83ef8c0 | 1117 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1118 | } |
e89e9cf5 | 1119 | |
1da177e4 LT |
1120 | |
1121 | while (size > 0) { | |
1122 | int i; | |
1123 | ||
89114afd | 1124 | if (skb_is_gso(skb)) |
e89e9cf5 AR |
1125 | len = size; |
1126 | else { | |
1127 | ||
1128 | /* Check if the remaining data fits into current packet. */ | |
1129 | len = mtu - skb->len; | |
1130 | if (len < size) | |
1131 | len = maxfraglen - skb->len; | |
1132 | } | |
1da177e4 LT |
1133 | if (len <= 0) { |
1134 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1135 | int alloclen; |
1136 | ||
1137 | skb_prev = skb; | |
0d0d2bba | 1138 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1139 | |
1140 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1141 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1142 | if (unlikely(!skb)) { | |
1143 | err = -ENOBUFS; | |
1144 | goto error; | |
1145 | } | |
1146 | ||
1147 | /* | |
1148 | * Fill in the control structures | |
1149 | */ | |
1150 | skb->ip_summed = CHECKSUM_NONE; | |
1151 | skb->csum = 0; | |
1152 | skb_reserve(skb, hh_len); | |
1153 | ||
1154 | /* | |
1155 | * Find where to start putting bytes. | |
1156 | */ | |
967b05f6 | 1157 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1158 | skb_reset_network_header(skb); |
b0e380b1 ACM |
1159 | skb->transport_header = (skb->network_header + |
1160 | fragheaderlen); | |
1da177e4 | 1161 | if (fraggap) { |
967b05f6 ACM |
1162 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1163 | maxfraglen, | |
9c70220b | 1164 | skb_transport_header(skb), |
967b05f6 | 1165 | fraggap, 0); |
1da177e4 LT |
1166 | skb_prev->csum = csum_sub(skb_prev->csum, |
1167 | skb->csum); | |
e9fa4f7b | 1168 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1169 | } |
1170 | ||
1171 | /* | |
1172 | * Put the packet on the pending queue. | |
1173 | */ | |
1174 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1175 | continue; | |
1176 | } | |
1177 | ||
1178 | i = skb_shinfo(skb)->nr_frags; | |
1179 | if (len > size) | |
1180 | len = size; | |
1181 | if (skb_can_coalesce(skb, i, page, offset)) { | |
1182 | skb_shinfo(skb)->frags[i-1].size += len; | |
1183 | } else if (i < MAX_SKB_FRAGS) { | |
1184 | get_page(page); | |
1185 | skb_fill_page_desc(skb, i, page, offset, len); | |
1186 | } else { | |
1187 | err = -EMSGSIZE; | |
1188 | goto error; | |
1189 | } | |
1190 | ||
1191 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1192 | __wsum csum; |
1da177e4 LT |
1193 | csum = csum_page(page, offset, len); |
1194 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1195 | } | |
1196 | ||
1197 | skb->len += len; | |
1198 | skb->data_len += len; | |
1e34a11d DM |
1199 | skb->truesize += len; |
1200 | atomic_add(len, &sk->sk_wmem_alloc); | |
1da177e4 LT |
1201 | offset += len; |
1202 | size -= len; | |
1203 | } | |
1204 | return 0; | |
1205 | ||
1206 | error: | |
1207 | inet->cork.length -= size; | |
5e38e270 | 1208 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1209 | return err; |
1210 | } | |
1211 | ||
429f08e9 PE |
1212 | static void ip_cork_release(struct inet_sock *inet) |
1213 | { | |
1214 | inet->cork.flags &= ~IPCORK_OPT; | |
1215 | kfree(inet->cork.opt); | |
1216 | inet->cork.opt = NULL; | |
c8cdaf99 YH |
1217 | dst_release(inet->cork.dst); |
1218 | inet->cork.dst = NULL; | |
429f08e9 PE |
1219 | } |
1220 | ||
1da177e4 LT |
1221 | /* |
1222 | * Combined all pending IP fragments on the socket as one IP datagram | |
1223 | * and push them out. | |
1224 | */ | |
1225 | int ip_push_pending_frames(struct sock *sk) | |
1226 | { | |
1227 | struct sk_buff *skb, *tmp_skb; | |
1228 | struct sk_buff **tail_skb; | |
1229 | struct inet_sock *inet = inet_sk(sk); | |
0388b004 | 1230 | struct net *net = sock_net(sk); |
1da177e4 | 1231 | struct ip_options *opt = NULL; |
c8cdaf99 | 1232 | struct rtable *rt = (struct rtable *)inet->cork.dst; |
1da177e4 | 1233 | struct iphdr *iph; |
76ab608d | 1234 | __be16 df = 0; |
1da177e4 LT |
1235 | __u8 ttl; |
1236 | int err = 0; | |
1237 | ||
1238 | if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL) | |
1239 | goto out; | |
1240 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1241 | ||
1242 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1243 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1244 | __skb_pull(skb, skb_network_offset(skb)); |
1da177e4 | 1245 | while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { |
cfe1fc77 | 1246 | __skb_pull(tmp_skb, skb_network_header_len(skb)); |
1da177e4 LT |
1247 | *tail_skb = tmp_skb; |
1248 | tail_skb = &(tmp_skb->next); | |
1249 | skb->len += tmp_skb->len; | |
1250 | skb->data_len += tmp_skb->len; | |
1251 | skb->truesize += tmp_skb->truesize; | |
1da177e4 LT |
1252 | tmp_skb->destructor = NULL; |
1253 | tmp_skb->sk = NULL; | |
1254 | } | |
1255 | ||
1256 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1257 | * to fragment the frame generated here. No matter, what transforms | |
1258 | * how transforms change size of the packet, it will come out. | |
1259 | */ | |
628a5c56 | 1260 | if (inet->pmtudisc < IP_PMTUDISC_DO) |
1da177e4 LT |
1261 | skb->local_df = 1; |
1262 | ||
1263 | /* DF bit is set when we want to see DF on outgoing frames. | |
1264 | * If local_df is set too, we still allow to fragment this frame | |
1265 | * locally. */ | |
628a5c56 | 1266 | if (inet->pmtudisc >= IP_PMTUDISC_DO || |
1da177e4 LT |
1267 | (skb->len <= dst_mtu(&rt->u.dst) && |
1268 | ip_dont_fragment(sk, &rt->u.dst))) | |
1269 | df = htons(IP_DF); | |
1270 | ||
1271 | if (inet->cork.flags & IPCORK_OPT) | |
1272 | opt = inet->cork.opt; | |
1273 | ||
1274 | if (rt->rt_type == RTN_MULTICAST) | |
1275 | ttl = inet->mc_ttl; | |
1276 | else | |
1277 | ttl = ip_select_ttl(inet, &rt->u.dst); | |
1278 | ||
1279 | iph = (struct iphdr *)skb->data; | |
1280 | iph->version = 4; | |
1281 | iph->ihl = 5; | |
1282 | if (opt) { | |
1283 | iph->ihl += opt->optlen>>2; | |
1284 | ip_options_build(skb, opt, inet->cork.addr, rt, 0); | |
1285 | } | |
1286 | iph->tos = inet->tos; | |
1da177e4 | 1287 | iph->frag_off = df; |
1a55d57b | 1288 | ip_select_ident(iph, &rt->u.dst, sk); |
1da177e4 LT |
1289 | iph->ttl = ttl; |
1290 | iph->protocol = sk->sk_protocol; | |
1291 | iph->saddr = rt->rt_src; | |
1292 | iph->daddr = rt->rt_dst; | |
1da177e4 LT |
1293 | |
1294 | skb->priority = sk->sk_priority; | |
4a19ec58 | 1295 | skb->mark = sk->sk_mark; |
a21bba94 ED |
1296 | /* |
1297 | * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec | |
1298 | * on dst refcount | |
1299 | */ | |
1300 | inet->cork.dst = NULL; | |
adf30907 | 1301 | skb_dst_set(skb, &rt->u.dst); |
1da177e4 | 1302 | |
96793b48 | 1303 | if (iph->protocol == IPPROTO_ICMP) |
0388b004 | 1304 | icmp_out_count(net, ((struct icmphdr *) |
96793b48 DS |
1305 | skb_transport_header(skb))->type); |
1306 | ||
1da177e4 | 1307 | /* Netfilter gets whole the not fragmented skb. */ |
c439cb2e | 1308 | err = ip_local_out(skb); |
1da177e4 LT |
1309 | if (err) { |
1310 | if (err > 0) | |
6ce9e7b5 | 1311 | err = net_xmit_errno(err); |
1da177e4 LT |
1312 | if (err) |
1313 | goto error; | |
1314 | } | |
1315 | ||
1316 | out: | |
429f08e9 | 1317 | ip_cork_release(inet); |
1da177e4 LT |
1318 | return err; |
1319 | ||
1320 | error: | |
5e38e270 | 1321 | IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS); |
1da177e4 LT |
1322 | goto out; |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * Throw away all pending data on the socket. | |
1327 | */ | |
1328 | void ip_flush_pending_frames(struct sock *sk) | |
1329 | { | |
1da177e4 LT |
1330 | struct sk_buff *skb; |
1331 | ||
1332 | while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) | |
1333 | kfree_skb(skb); | |
1334 | ||
429f08e9 | 1335 | ip_cork_release(inet_sk(sk)); |
1da177e4 LT |
1336 | } |
1337 | ||
1338 | ||
1339 | /* | |
1340 | * Fetch data from kernel space and fill in checksum if needed. | |
1341 | */ | |
e905a9ed | 1342 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1343 | int len, int odd, struct sk_buff *skb) |
1344 | { | |
5084205f | 1345 | __wsum csum; |
1da177e4 LT |
1346 | |
1347 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1348 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1349 | return 0; |
1da177e4 LT |
1350 | } |
1351 | ||
e905a9ed | 1352 | /* |
1da177e4 LT |
1353 | * Generic function to send a packet as reply to another packet. |
1354 | * Used to send TCP resets so far. ICMP should use this function too. | |
1355 | * | |
e905a9ed | 1356 | * Should run single threaded per socket because it uses the sock |
1da177e4 | 1357 | * structure to pass arguments. |
1da177e4 LT |
1358 | */ |
1359 | void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg, | |
1360 | unsigned int len) | |
1361 | { | |
1362 | struct inet_sock *inet = inet_sk(sk); | |
1363 | struct { | |
1364 | struct ip_options opt; | |
1365 | char data[40]; | |
1366 | } replyopts; | |
1367 | struct ipcm_cookie ipc; | |
3ca3c68e | 1368 | __be32 daddr; |
511c3f92 | 1369 | struct rtable *rt = skb_rtable(skb); |
1da177e4 LT |
1370 | |
1371 | if (ip_options_echo(&replyopts.opt, skb)) | |
1372 | return; | |
1373 | ||
1374 | daddr = ipc.addr = rt->rt_src; | |
1375 | ipc.opt = NULL; | |
51f31cab | 1376 | ipc.shtx.flags = 0; |
1da177e4 LT |
1377 | |
1378 | if (replyopts.opt.optlen) { | |
1379 | ipc.opt = &replyopts.opt; | |
1380 | ||
1381 | if (ipc.opt->srr) | |
1382 | daddr = replyopts.opt.faddr; | |
1383 | } | |
1384 | ||
1385 | { | |
f0e48dbf PM |
1386 | struct flowi fl = { .oif = arg->bound_dev_if, |
1387 | .nl_u = { .ip4_u = | |
1da177e4 LT |
1388 | { .daddr = daddr, |
1389 | .saddr = rt->rt_spec_dst, | |
eddc9ec5 | 1390 | .tos = RT_TOS(ip_hdr(skb)->tos) } }, |
1da177e4 LT |
1391 | /* Not quite clean, but right. */ |
1392 | .uli_u = { .ports = | |
aa8223c7 ACM |
1393 | { .sport = tcp_hdr(skb)->dest, |
1394 | .dport = tcp_hdr(skb)->source } }, | |
86b08d86 KK |
1395 | .proto = sk->sk_protocol, |
1396 | .flags = ip_reply_arg_flowi_flags(arg) }; | |
beb8d13b | 1397 | security_skb_classify_flow(skb, &fl); |
3b1e0a65 | 1398 | if (ip_route_output_key(sock_net(sk), &rt, &fl)) |
1da177e4 LT |
1399 | return; |
1400 | } | |
1401 | ||
1402 | /* And let IP do all the hard work. | |
1403 | ||
1404 | This chunk is not reenterable, hence spinlock. | |
1405 | Note that it uses the fact, that this function is called | |
1406 | with locally disabled BH and that sk cannot be already spinlocked. | |
1407 | */ | |
1408 | bh_lock_sock(sk); | |
eddc9ec5 | 1409 | inet->tos = ip_hdr(skb)->tos; |
1da177e4 | 1410 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1411 | sk->sk_protocol = ip_hdr(skb)->protocol; |
f0e48dbf | 1412 | sk->sk_bound_dev_if = arg->bound_dev_if; |
1da177e4 | 1413 | ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0, |
2e77d89b | 1414 | &ipc, &rt, MSG_DONTWAIT); |
1da177e4 LT |
1415 | if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { |
1416 | if (arg->csumoffset >= 0) | |
9c70220b ACM |
1417 | *((__sum16 *)skb_transport_header(skb) + |
1418 | arg->csumoffset) = csum_fold(csum_add(skb->csum, | |
1419 | arg->csum)); | |
1da177e4 LT |
1420 | skb->ip_summed = CHECKSUM_NONE; |
1421 | ip_push_pending_frames(sk); | |
1422 | } | |
1423 | ||
1424 | bh_unlock_sock(sk); | |
1425 | ||
1426 | ip_rt_put(rt); | |
1427 | } | |
1428 | ||
1da177e4 LT |
1429 | void __init ip_init(void) |
1430 | { | |
1da177e4 LT |
1431 | ip_rt_init(); |
1432 | inet_initpeers(); | |
1433 | ||
1434 | #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) | |
1435 | igmp_mc_proc_init(); | |
1436 | #endif | |
1437 | } | |
1438 | ||
1da177e4 LT |
1439 | EXPORT_SYMBOL(ip_generic_getfrag); |
1440 | EXPORT_SYMBOL(ip_queue_xmit); | |
1441 | EXPORT_SYMBOL(ip_send_check); |