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 | * | |
8 | * Version: $Id: ip_output.c,v 1.100 2002/02/01 22:01:03 davem Exp $ | |
9 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * Donald Becker, <becker@super.org> | |
13 | * Alan Cox, <Alan.Cox@linux.org> | |
14 | * Richard Underwood | |
15 | * Stefan Becker, <stefanb@yello.ping.de> | |
16 | * Jorge Cwik, <jorge@laser.satlink.net> | |
17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
18 | * Hirokazu Takahashi, <taka@valinux.co.jp> | |
19 | * | |
20 | * See ip_input.c for original log | |
21 | * | |
22 | * Fixes: | |
23 | * Alan Cox : Missing nonblock feature in ip_build_xmit. | |
24 | * Mike Kilburn : htons() missing in ip_build_xmit. | |
e905a9ed | 25 | * Bradford Johnson: Fix faulty handling of some frames when |
1da177e4 LT |
26 | * no route is found. |
27 | * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit | |
28 | * (in case if packet not accepted by | |
29 | * output firewall rules) | |
30 | * Mike McLagan : Routing by source | |
31 | * Alexey Kuznetsov: use new route cache | |
32 | * Andi Kleen: Fix broken PMTU recovery and remove | |
33 | * some redundant tests. | |
34 | * Vitaly E. Lavrov : Transparent proxy revived after year coma. | |
35 | * Andi Kleen : Replace ip_reply with ip_send_reply. | |
e905a9ed YH |
36 | * Andi Kleen : Split fast and slow ip_build_xmit path |
37 | * for decreased register pressure on x86 | |
38 | * and more readibility. | |
1da177e4 LT |
39 | * Marc Boucher : When call_out_firewall returns FW_QUEUE, |
40 | * silently drop skb instead of failing with -EPERM. | |
41 | * Detlev Wengorz : Copy protocol for fragments. | |
42 | * Hirokazu Takahashi: HW checksumming for outgoing UDP | |
43 | * datagrams. | |
44 | * Hirokazu Takahashi: sendfile() on UDP works now. | |
45 | */ | |
46 | ||
47 | #include <asm/uaccess.h> | |
48 | #include <asm/system.h> | |
49 | #include <linux/module.h> | |
50 | #include <linux/types.h> | |
51 | #include <linux/kernel.h> | |
1da177e4 LT |
52 | #include <linux/mm.h> |
53 | #include <linux/string.h> | |
54 | #include <linux/errno.h> | |
a1f8e7f7 | 55 | #include <linux/highmem.h> |
1da177e4 LT |
56 | |
57 | #include <linux/socket.h> | |
58 | #include <linux/sockios.h> | |
59 | #include <linux/in.h> | |
60 | #include <linux/inet.h> | |
61 | #include <linux/netdevice.h> | |
62 | #include <linux/etherdevice.h> | |
63 | #include <linux/proc_fs.h> | |
64 | #include <linux/stat.h> | |
65 | #include <linux/init.h> | |
66 | ||
67 | #include <net/snmp.h> | |
68 | #include <net/ip.h> | |
69 | #include <net/protocol.h> | |
70 | #include <net/route.h> | |
cfacb057 | 71 | #include <net/xfrm.h> |
1da177e4 LT |
72 | #include <linux/skbuff.h> |
73 | #include <net/sock.h> | |
74 | #include <net/arp.h> | |
75 | #include <net/icmp.h> | |
1da177e4 LT |
76 | #include <net/checksum.h> |
77 | #include <net/inetpeer.h> | |
78 | #include <net/checksum.h> | |
79 | #include <linux/igmp.h> | |
80 | #include <linux/netfilter_ipv4.h> | |
81 | #include <linux/netfilter_bridge.h> | |
82 | #include <linux/mroute.h> | |
83 | #include <linux/netlink.h> | |
6cbb0df7 | 84 | #include <linux/tcp.h> |
1da177e4 | 85 | |
ab32ea5d | 86 | int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; |
1da177e4 LT |
87 | |
88 | /* Generate a checksum for an outgoing IP datagram. */ | |
89 | __inline__ void ip_send_check(struct iphdr *iph) | |
90 | { | |
91 | iph->check = 0; | |
92 | iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); | |
93 | } | |
94 | ||
95 | /* dev_loopback_xmit for use with netfilter. */ | |
96 | static int ip_dev_loopback_xmit(struct sk_buff *newskb) | |
97 | { | |
459a98ed | 98 | skb_reset_mac_header(newskb); |
bbe735e4 | 99 | __skb_pull(newskb, skb_network_offset(newskb)); |
1da177e4 LT |
100 | newskb->pkt_type = PACKET_LOOPBACK; |
101 | newskb->ip_summed = CHECKSUM_UNNECESSARY; | |
102 | BUG_TRAP(newskb->dst); | |
1da177e4 LT |
103 | netif_rx(newskb); |
104 | return 0; | |
105 | } | |
106 | ||
107 | static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst) | |
108 | { | |
109 | int ttl = inet->uc_ttl; | |
110 | ||
111 | if (ttl < 0) | |
112 | ttl = dst_metric(dst, RTAX_HOPLIMIT); | |
113 | return ttl; | |
114 | } | |
115 | ||
e905a9ed | 116 | /* |
1da177e4 LT |
117 | * Add an ip header to a skbuff and send it out. |
118 | * | |
119 | */ | |
120 | int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk, | |
13d8eaa0 | 121 | __be32 saddr, __be32 daddr, struct ip_options *opt) |
1da177e4 LT |
122 | { |
123 | struct inet_sock *inet = inet_sk(sk); | |
124 | struct rtable *rt = (struct rtable *)skb->dst; | |
125 | struct iphdr *iph; | |
126 | ||
127 | /* Build the IP header. */ | |
8856dfa3 ACM |
128 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0)); |
129 | skb_reset_network_header(skb); | |
eddc9ec5 | 130 | iph = ip_hdr(skb); |
1da177e4 LT |
131 | iph->version = 4; |
132 | iph->ihl = 5; | |
133 | iph->tos = inet->tos; | |
134 | if (ip_dont_fragment(sk, &rt->u.dst)) | |
135 | iph->frag_off = htons(IP_DF); | |
136 | else | |
137 | iph->frag_off = 0; | |
138 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
139 | iph->daddr = rt->rt_dst; | |
140 | iph->saddr = rt->rt_src; | |
141 | iph->protocol = sk->sk_protocol; | |
142 | iph->tot_len = htons(skb->len); | |
143 | ip_select_ident(iph, &rt->u.dst, sk); | |
1da177e4 LT |
144 | |
145 | if (opt && opt->optlen) { | |
146 | iph->ihl += opt->optlen>>2; | |
147 | ip_options_build(skb, opt, daddr, rt, 0); | |
148 | } | |
149 | ip_send_check(iph); | |
150 | ||
151 | skb->priority = sk->sk_priority; | |
152 | ||
153 | /* Send it out. */ | |
154 | return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev, | |
155 | dst_output); | |
156 | } | |
157 | ||
d8c97a94 ACM |
158 | EXPORT_SYMBOL_GPL(ip_build_and_send_pkt); |
159 | ||
1da177e4 LT |
160 | static inline int ip_finish_output2(struct sk_buff *skb) |
161 | { | |
162 | struct dst_entry *dst = skb->dst; | |
1da177e4 LT |
163 | struct net_device *dev = dst->dev; |
164 | int hh_len = LL_RESERVED_SPACE(dev); | |
165 | ||
166 | /* Be paranoid, rather than too clever. */ | |
167 | if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) { | |
168 | struct sk_buff *skb2; | |
169 | ||
170 | skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev)); | |
171 | if (skb2 == NULL) { | |
172 | kfree_skb(skb); | |
173 | return -ENOMEM; | |
174 | } | |
175 | if (skb->sk) | |
176 | skb_set_owner_w(skb2, skb->sk); | |
177 | kfree_skb(skb); | |
178 | skb = skb2; | |
179 | } | |
180 | ||
3644f0ce SH |
181 | if (dst->hh) |
182 | return neigh_hh_output(dst->hh, skb); | |
183 | else if (dst->neighbour) | |
1da177e4 LT |
184 | return dst->neighbour->output(skb); |
185 | ||
186 | if (net_ratelimit()) | |
187 | printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n"); | |
188 | kfree_skb(skb); | |
189 | return -EINVAL; | |
190 | } | |
191 | ||
33d043d6 | 192 | static inline int ip_finish_output(struct sk_buff *skb) |
1da177e4 | 193 | { |
5c901daa PM |
194 | #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM) |
195 | /* Policy lookup after SNAT yielded a new policy */ | |
48d5cad8 PM |
196 | if (skb->dst->xfrm != NULL) { |
197 | IPCB(skb)->flags |= IPSKB_REROUTED; | |
198 | return dst_output(skb); | |
199 | } | |
5c901daa | 200 | #endif |
89114afd | 201 | if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb)) |
1bd9bef6 PM |
202 | return ip_fragment(skb, ip_finish_output2); |
203 | else | |
204 | return ip_finish_output2(skb); | |
1da177e4 LT |
205 | } |
206 | ||
207 | int ip_mc_output(struct sk_buff *skb) | |
208 | { | |
209 | struct sock *sk = skb->sk; | |
210 | struct rtable *rt = (struct rtable*)skb->dst; | |
211 | struct net_device *dev = rt->u.dst.dev; | |
212 | ||
213 | /* | |
214 | * If the indicated interface is up and running, send the packet. | |
215 | */ | |
216 | IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS); | |
217 | ||
218 | skb->dev = dev; | |
219 | skb->protocol = htons(ETH_P_IP); | |
220 | ||
221 | /* | |
222 | * Multicasts are looped back for other local users | |
223 | */ | |
224 | ||
225 | if (rt->rt_flags&RTCF_MULTICAST) { | |
226 | if ((!sk || inet_sk(sk)->mc_loop) | |
227 | #ifdef CONFIG_IP_MROUTE | |
228 | /* Small optimization: do not loopback not local frames, | |
229 | which returned after forwarding; they will be dropped | |
230 | by ip_mr_input in any case. | |
231 | Note, that local frames are looped back to be delivered | |
232 | to local recipients. | |
233 | ||
234 | This check is duplicated in ip_mr_input at the moment. | |
235 | */ | |
236 | && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED)) | |
237 | #endif | |
238 | ) { | |
239 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
240 | if (newskb) | |
241 | NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL, | |
e905a9ed | 242 | newskb->dev, |
1da177e4 LT |
243 | ip_dev_loopback_xmit); |
244 | } | |
245 | ||
246 | /* Multicasts with ttl 0 must not go beyond the host */ | |
247 | ||
eddc9ec5 | 248 | if (ip_hdr(skb)->ttl == 0) { |
1da177e4 LT |
249 | kfree_skb(skb); |
250 | return 0; | |
251 | } | |
252 | } | |
253 | ||
254 | if (rt->rt_flags&RTCF_BROADCAST) { | |
255 | struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC); | |
256 | if (newskb) | |
257 | NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL, | |
258 | newskb->dev, ip_dev_loopback_xmit); | |
259 | } | |
260 | ||
48d5cad8 PM |
261 | return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev, |
262 | ip_finish_output, | |
263 | !(IPCB(skb)->flags & IPSKB_REROUTED)); | |
1da177e4 LT |
264 | } |
265 | ||
266 | int ip_output(struct sk_buff *skb) | |
267 | { | |
1bd9bef6 PM |
268 | struct net_device *dev = skb->dst->dev; |
269 | ||
1da177e4 LT |
270 | IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS); |
271 | ||
1bd9bef6 PM |
272 | skb->dev = dev; |
273 | skb->protocol = htons(ETH_P_IP); | |
274 | ||
48d5cad8 | 275 | return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev, |
e905a9ed | 276 | ip_finish_output, |
48d5cad8 | 277 | !(IPCB(skb)->flags & IPSKB_REROUTED)); |
1da177e4 LT |
278 | } |
279 | ||
e89862f4 | 280 | int ip_queue_xmit(struct sk_buff *skb, int ipfragok) |
1da177e4 | 281 | { |
e89862f4 | 282 | struct sock *sk = skb->sk; |
1da177e4 LT |
283 | struct inet_sock *inet = inet_sk(sk); |
284 | struct ip_options *opt = inet->opt; | |
285 | struct rtable *rt; | |
286 | struct iphdr *iph; | |
287 | ||
288 | /* Skip all of this if the packet is already routed, | |
289 | * f.e. by something like SCTP. | |
290 | */ | |
291 | rt = (struct rtable *) skb->dst; | |
292 | if (rt != NULL) | |
293 | goto packet_routed; | |
294 | ||
295 | /* Make sure we can route this packet. */ | |
296 | rt = (struct rtable *)__sk_dst_check(sk, 0); | |
297 | if (rt == NULL) { | |
3ca3c68e | 298 | __be32 daddr; |
1da177e4 LT |
299 | |
300 | /* Use correct destination address if we have options. */ | |
301 | daddr = inet->daddr; | |
302 | if(opt && opt->srr) | |
303 | daddr = opt->faddr; | |
304 | ||
305 | { | |
306 | struct flowi fl = { .oif = sk->sk_bound_dev_if, | |
307 | .nl_u = { .ip4_u = | |
308 | { .daddr = daddr, | |
309 | .saddr = inet->saddr, | |
310 | .tos = RT_CONN_FLAGS(sk) } }, | |
311 | .proto = sk->sk_protocol, | |
312 | .uli_u = { .ports = | |
313 | { .sport = inet->sport, | |
314 | .dport = inet->dport } } }; | |
315 | ||
316 | /* If this fails, retransmit mechanism of transport layer will | |
317 | * keep trying until route appears or the connection times | |
318 | * itself out. | |
319 | */ | |
beb8d13b | 320 | security_sk_classify_flow(sk, &fl); |
1da177e4 LT |
321 | if (ip_route_output_flow(&rt, &fl, sk, 0)) |
322 | goto no_route; | |
323 | } | |
6cbb0df7 | 324 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
325 | } |
326 | skb->dst = dst_clone(&rt->u.dst); | |
327 | ||
328 | packet_routed: | |
329 | if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) | |
330 | goto no_route; | |
331 | ||
332 | /* OK, we know where to send it, allocate and build IP header. */ | |
8856dfa3 ACM |
333 | skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0)); |
334 | skb_reset_network_header(skb); | |
eddc9ec5 | 335 | iph = ip_hdr(skb); |
714e85be | 336 | *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff)); |
1da177e4 LT |
337 | iph->tot_len = htons(skb->len); |
338 | if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok) | |
339 | iph->frag_off = htons(IP_DF); | |
340 | else | |
341 | iph->frag_off = 0; | |
342 | iph->ttl = ip_select_ttl(inet, &rt->u.dst); | |
343 | iph->protocol = sk->sk_protocol; | |
344 | iph->saddr = rt->rt_src; | |
345 | iph->daddr = rt->rt_dst; | |
1da177e4 LT |
346 | /* Transport layer set skb->h.foo itself. */ |
347 | ||
348 | if (opt && opt->optlen) { | |
349 | iph->ihl += opt->optlen >> 2; | |
350 | ip_options_build(skb, opt, inet->daddr, rt, 0); | |
351 | } | |
352 | ||
89f5f0ae | 353 | ip_select_ident_more(iph, &rt->u.dst, sk, |
7967168c | 354 | (skb_shinfo(skb)->gso_segs ?: 1) - 1); |
1da177e4 LT |
355 | |
356 | /* Add an IP checksum. */ | |
357 | ip_send_check(iph); | |
358 | ||
359 | skb->priority = sk->sk_priority; | |
360 | ||
361 | return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev, | |
362 | dst_output); | |
363 | ||
364 | no_route: | |
365 | IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES); | |
366 | kfree_skb(skb); | |
367 | return -EHOSTUNREACH; | |
368 | } | |
369 | ||
370 | ||
371 | static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from) | |
372 | { | |
373 | to->pkt_type = from->pkt_type; | |
374 | to->priority = from->priority; | |
375 | to->protocol = from->protocol; | |
1da177e4 LT |
376 | dst_release(to->dst); |
377 | to->dst = dst_clone(from->dst); | |
378 | to->dev = from->dev; | |
82e91ffe | 379 | to->mark = from->mark; |
1da177e4 LT |
380 | |
381 | /* Copy the flags to each fragment. */ | |
382 | IPCB(to)->flags = IPCB(from)->flags; | |
383 | ||
384 | #ifdef CONFIG_NET_SCHED | |
385 | to->tc_index = from->tc_index; | |
386 | #endif | |
387 | #ifdef CONFIG_NETFILTER | |
1da177e4 LT |
388 | /* Connection association is same as pre-frag packet */ |
389 | nf_conntrack_put(to->nfct); | |
390 | to->nfct = from->nfct; | |
391 | nf_conntrack_get(to->nfct); | |
392 | to->nfctinfo = from->nfctinfo; | |
c98d80ed JA |
393 | #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) |
394 | to->ipvs_property = from->ipvs_property; | |
395 | #endif | |
1da177e4 LT |
396 | #ifdef CONFIG_BRIDGE_NETFILTER |
397 | nf_bridge_put(to->nf_bridge); | |
398 | to->nf_bridge = from->nf_bridge; | |
399 | nf_bridge_get(to->nf_bridge); | |
400 | #endif | |
1da177e4 | 401 | #endif |
984bc16c | 402 | skb_copy_secmark(to, from); |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * This IP datagram is too large to be sent in one piece. Break it up into | |
407 | * smaller pieces (each of size equal to IP header plus | |
408 | * a block of the data of the original IP data part) that will yet fit in a | |
409 | * single device frame, and queue such a frame for sending. | |
410 | */ | |
411 | ||
2e2f7aef | 412 | int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*)) |
1da177e4 LT |
413 | { |
414 | struct iphdr *iph; | |
415 | int raw = 0; | |
416 | int ptr; | |
417 | struct net_device *dev; | |
418 | struct sk_buff *skb2; | |
9bcfcaf5 | 419 | unsigned int mtu, hlen, left, len, ll_rs, pad; |
1da177e4 | 420 | int offset; |
76ab608d | 421 | __be16 not_last_frag; |
1da177e4 LT |
422 | struct rtable *rt = (struct rtable*)skb->dst; |
423 | int err = 0; | |
424 | ||
425 | dev = rt->u.dst.dev; | |
426 | ||
427 | /* | |
428 | * Point into the IP datagram header. | |
429 | */ | |
430 | ||
eddc9ec5 | 431 | iph = ip_hdr(skb); |
1da177e4 LT |
432 | |
433 | if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) { | |
0668b472 | 434 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); |
1da177e4 LT |
435 | icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, |
436 | htonl(dst_mtu(&rt->u.dst))); | |
437 | kfree_skb(skb); | |
438 | return -EMSGSIZE; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Setup starting values. | |
443 | */ | |
444 | ||
445 | hlen = iph->ihl * 4; | |
446 | mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */ | |
89cee8b1 | 447 | IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE; |
1da177e4 LT |
448 | |
449 | /* When frag_list is given, use it. First, check its validity: | |
450 | * some transformers could create wrong frag_list or break existing | |
451 | * one, it is not prohibited. In this case fall back to copying. | |
452 | * | |
453 | * LATER: this step can be merged to real generation of fragments, | |
454 | * we can switch to copy when see the first bad fragment. | |
455 | */ | |
456 | if (skb_shinfo(skb)->frag_list) { | |
457 | struct sk_buff *frag; | |
458 | int first_len = skb_pagelen(skb); | |
459 | ||
460 | if (first_len - hlen > mtu || | |
461 | ((first_len - hlen) & 7) || | |
462 | (iph->frag_off & htons(IP_MF|IP_OFFSET)) || | |
463 | skb_cloned(skb)) | |
464 | goto slow_path; | |
465 | ||
466 | for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) { | |
467 | /* Correct geometry. */ | |
468 | if (frag->len > mtu || | |
469 | ((frag->len & 7) && frag->next) || | |
470 | skb_headroom(frag) < hlen) | |
471 | goto slow_path; | |
472 | ||
473 | /* Partially cloned skb? */ | |
474 | if (skb_shared(frag)) | |
475 | goto slow_path; | |
2fdba6b0 HX |
476 | |
477 | BUG_ON(frag->sk); | |
478 | if (skb->sk) { | |
479 | sock_hold(skb->sk); | |
480 | frag->sk = skb->sk; | |
481 | frag->destructor = sock_wfree; | |
482 | skb->truesize -= frag->truesize; | |
483 | } | |
1da177e4 LT |
484 | } |
485 | ||
486 | /* Everything is OK. Generate! */ | |
487 | ||
488 | err = 0; | |
489 | offset = 0; | |
490 | frag = skb_shinfo(skb)->frag_list; | |
491 | skb_shinfo(skb)->frag_list = NULL; | |
492 | skb->data_len = first_len - skb_headlen(skb); | |
493 | skb->len = first_len; | |
494 | iph->tot_len = htons(first_len); | |
495 | iph->frag_off = htons(IP_MF); | |
496 | ip_send_check(iph); | |
497 | ||
498 | for (;;) { | |
499 | /* Prepare header of the next frame, | |
500 | * before previous one went down. */ | |
501 | if (frag) { | |
502 | frag->ip_summed = CHECKSUM_NONE; | |
badff6d0 | 503 | skb_reset_transport_header(frag); |
e2d1bca7 ACM |
504 | __skb_push(frag, hlen); |
505 | skb_reset_network_header(frag); | |
d56f90a7 | 506 | memcpy(skb_network_header(frag), iph, hlen); |
eddc9ec5 | 507 | iph = ip_hdr(frag); |
1da177e4 LT |
508 | iph->tot_len = htons(frag->len); |
509 | ip_copy_metadata(frag, skb); | |
510 | if (offset == 0) | |
511 | ip_options_fragment(frag); | |
512 | offset += skb->len - hlen; | |
513 | iph->frag_off = htons(offset>>3); | |
514 | if (frag->next != NULL) | |
515 | iph->frag_off |= htons(IP_MF); | |
516 | /* Ready, complete checksum */ | |
517 | ip_send_check(iph); | |
518 | } | |
519 | ||
520 | err = output(skb); | |
521 | ||
dafee490 WD |
522 | if (!err) |
523 | IP_INC_STATS(IPSTATS_MIB_FRAGCREATES); | |
1da177e4 LT |
524 | if (err || !frag) |
525 | break; | |
526 | ||
527 | skb = frag; | |
528 | frag = skb->next; | |
529 | skb->next = NULL; | |
530 | } | |
531 | ||
532 | if (err == 0) { | |
533 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
534 | return 0; | |
535 | } | |
536 | ||
537 | while (frag) { | |
538 | skb = frag->next; | |
539 | kfree_skb(frag); | |
540 | frag = skb; | |
541 | } | |
542 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); | |
543 | return err; | |
544 | } | |
545 | ||
546 | slow_path: | |
547 | left = skb->len - hlen; /* Space per frame */ | |
548 | ptr = raw + hlen; /* Where to start from */ | |
549 | ||
1da177e4 | 550 | /* for bridged IP traffic encapsulated inside f.e. a vlan header, |
9bcfcaf5 SH |
551 | * we need to make room for the encapsulating header |
552 | */ | |
553 | pad = nf_bridge_pad(skb); | |
554 | ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad); | |
555 | mtu -= pad; | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * Fragment the datagram. | |
559 | */ | |
560 | ||
561 | offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3; | |
562 | not_last_frag = iph->frag_off & htons(IP_MF); | |
563 | ||
564 | /* | |
565 | * Keep copying data until we run out. | |
566 | */ | |
567 | ||
132adf54 | 568 | while (left > 0) { |
1da177e4 LT |
569 | len = left; |
570 | /* IF: it doesn't fit, use 'mtu' - the data space left */ | |
571 | if (len > mtu) | |
572 | len = mtu; | |
573 | /* IF: we are not sending upto and including the packet end | |
574 | then align the next start on an eight byte boundary */ | |
575 | if (len < left) { | |
576 | len &= ~7; | |
577 | } | |
578 | /* | |
579 | * Allocate buffer. | |
580 | */ | |
581 | ||
582 | if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) { | |
64ce2073 | 583 | NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n"); |
1da177e4 LT |
584 | err = -ENOMEM; |
585 | goto fail; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Set up data on packet | |
590 | */ | |
591 | ||
592 | ip_copy_metadata(skb2, skb); | |
593 | skb_reserve(skb2, ll_rs); | |
594 | skb_put(skb2, len + hlen); | |
c1d2bbe1 | 595 | skb_reset_network_header(skb2); |
967b05f6 | 596 | skb2->h.raw = skb2->nh.raw + hlen; |
1da177e4 LT |
597 | |
598 | /* | |
599 | * Charge the memory for the fragment to any owner | |
600 | * it might possess | |
601 | */ | |
602 | ||
603 | if (skb->sk) | |
604 | skb_set_owner_w(skb2, skb->sk); | |
605 | ||
606 | /* | |
607 | * Copy the packet header into the new buffer. | |
608 | */ | |
609 | ||
d56f90a7 | 610 | memcpy(skb_network_header(skb2), skb->data, hlen); |
1da177e4 LT |
611 | |
612 | /* | |
613 | * Copy a block of the IP datagram. | |
614 | */ | |
615 | if (skb_copy_bits(skb, ptr, skb2->h.raw, len)) | |
616 | BUG(); | |
617 | left -= len; | |
618 | ||
619 | /* | |
620 | * Fill in the new header fields. | |
621 | */ | |
eddc9ec5 | 622 | iph = ip_hdr(skb2); |
1da177e4 LT |
623 | iph->frag_off = htons((offset >> 3)); |
624 | ||
625 | /* ANK: dirty, but effective trick. Upgrade options only if | |
626 | * the segment to be fragmented was THE FIRST (otherwise, | |
627 | * options are already fixed) and make it ONCE | |
628 | * on the initial skb, so that all the following fragments | |
629 | * will inherit fixed options. | |
630 | */ | |
631 | if (offset == 0) | |
632 | ip_options_fragment(skb); | |
633 | ||
634 | /* | |
635 | * Added AC : If we are fragmenting a fragment that's not the | |
636 | * last fragment then keep MF on each bit | |
637 | */ | |
638 | if (left > 0 || not_last_frag) | |
639 | iph->frag_off |= htons(IP_MF); | |
640 | ptr += len; | |
641 | offset += len; | |
642 | ||
643 | /* | |
644 | * Put this fragment into the sending queue. | |
645 | */ | |
1da177e4 LT |
646 | iph->tot_len = htons(len + hlen); |
647 | ||
648 | ip_send_check(iph); | |
649 | ||
650 | err = output(skb2); | |
651 | if (err) | |
652 | goto fail; | |
dafee490 WD |
653 | |
654 | IP_INC_STATS(IPSTATS_MIB_FRAGCREATES); | |
1da177e4 LT |
655 | } |
656 | kfree_skb(skb); | |
657 | IP_INC_STATS(IPSTATS_MIB_FRAGOKS); | |
658 | return err; | |
659 | ||
660 | fail: | |
e905a9ed | 661 | kfree_skb(skb); |
1da177e4 LT |
662 | IP_INC_STATS(IPSTATS_MIB_FRAGFAILS); |
663 | return err; | |
664 | } | |
665 | ||
2e2f7aef PM |
666 | EXPORT_SYMBOL(ip_fragment); |
667 | ||
1da177e4 LT |
668 | int |
669 | ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb) | |
670 | { | |
671 | struct iovec *iov = from; | |
672 | ||
84fa7933 | 673 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
1da177e4 LT |
674 | if (memcpy_fromiovecend(to, iov, offset, len) < 0) |
675 | return -EFAULT; | |
676 | } else { | |
44bb9363 | 677 | __wsum csum = 0; |
1da177e4 LT |
678 | if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0) |
679 | return -EFAULT; | |
680 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
681 | } | |
682 | return 0; | |
683 | } | |
684 | ||
44bb9363 | 685 | static inline __wsum |
1da177e4 LT |
686 | csum_page(struct page *page, int offset, int copy) |
687 | { | |
688 | char *kaddr; | |
44bb9363 | 689 | __wsum csum; |
1da177e4 LT |
690 | kaddr = kmap(page); |
691 | csum = csum_partial(kaddr + offset, copy, 0); | |
692 | kunmap(page); | |
693 | return csum; | |
694 | } | |
695 | ||
4b30b1c6 | 696 | static inline int ip_ufo_append_data(struct sock *sk, |
e89e9cf5 AR |
697 | int getfrag(void *from, char *to, int offset, int len, |
698 | int odd, struct sk_buff *skb), | |
699 | void *from, int length, int hh_len, int fragheaderlen, | |
700 | int transhdrlen, int mtu,unsigned int flags) | |
701 | { | |
702 | struct sk_buff *skb; | |
703 | int err; | |
704 | ||
705 | /* There is support for UDP fragmentation offload by network | |
706 | * device, so create one single skb packet containing complete | |
707 | * udp datagram | |
708 | */ | |
709 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) { | |
710 | skb = sock_alloc_send_skb(sk, | |
711 | hh_len + fragheaderlen + transhdrlen + 20, | |
712 | (flags & MSG_DONTWAIT), &err); | |
713 | ||
714 | if (skb == NULL) | |
715 | return err; | |
716 | ||
717 | /* reserve space for Hardware header */ | |
718 | skb_reserve(skb, hh_len); | |
719 | ||
720 | /* create space for UDP/IP header */ | |
721 | skb_put(skb,fragheaderlen + transhdrlen); | |
722 | ||
723 | /* initialize network header pointer */ | |
c1d2bbe1 | 724 | skb_reset_network_header(skb); |
e89e9cf5 AR |
725 | |
726 | /* initialize protocol header pointer */ | |
967b05f6 | 727 | skb->h.raw = skb->nh.raw + fragheaderlen; |
e89e9cf5 | 728 | |
84fa7933 | 729 | skb->ip_summed = CHECKSUM_PARTIAL; |
e89e9cf5 AR |
730 | skb->csum = 0; |
731 | sk->sk_sndmsg_off = 0; | |
732 | } | |
733 | ||
734 | err = skb_append_datato_frags(sk,skb, getfrag, from, | |
735 | (length - transhdrlen)); | |
736 | if (!err) { | |
737 | /* specify the length of each IP datagram fragment*/ | |
7967168c | 738 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; |
f83ef8c0 | 739 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
e89e9cf5 AR |
740 | __skb_queue_tail(&sk->sk_write_queue, skb); |
741 | ||
742 | return 0; | |
743 | } | |
744 | /* There is not enough support do UFO , | |
745 | * so follow normal path | |
746 | */ | |
747 | kfree_skb(skb); | |
748 | return err; | |
749 | } | |
750 | ||
1da177e4 LT |
751 | /* |
752 | * ip_append_data() and ip_append_page() can make one large IP datagram | |
753 | * from many pieces of data. Each pieces will be holded on the socket | |
754 | * until ip_push_pending_frames() is called. Each piece can be a page | |
755 | * or non-page data. | |
e905a9ed | 756 | * |
1da177e4 LT |
757 | * Not only UDP, other transport protocols - e.g. raw sockets - can use |
758 | * this interface potentially. | |
759 | * | |
760 | * LATER: length must be adjusted by pad at tail, when it is required. | |
761 | */ | |
762 | int ip_append_data(struct sock *sk, | |
763 | int getfrag(void *from, char *to, int offset, int len, | |
764 | int odd, struct sk_buff *skb), | |
765 | void *from, int length, int transhdrlen, | |
766 | struct ipcm_cookie *ipc, struct rtable *rt, | |
767 | unsigned int flags) | |
768 | { | |
769 | struct inet_sock *inet = inet_sk(sk); | |
770 | struct sk_buff *skb; | |
771 | ||
772 | struct ip_options *opt = NULL; | |
773 | int hh_len; | |
774 | int exthdrlen; | |
775 | int mtu; | |
776 | int copy; | |
777 | int err; | |
778 | int offset = 0; | |
779 | unsigned int maxfraglen, fragheaderlen; | |
780 | int csummode = CHECKSUM_NONE; | |
781 | ||
782 | if (flags&MSG_PROBE) | |
783 | return 0; | |
784 | ||
785 | if (skb_queue_empty(&sk->sk_write_queue)) { | |
786 | /* | |
787 | * setup for corking. | |
788 | */ | |
789 | opt = ipc->opt; | |
790 | if (opt) { | |
791 | if (inet->cork.opt == NULL) { | |
792 | inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation); | |
793 | if (unlikely(inet->cork.opt == NULL)) | |
794 | return -ENOBUFS; | |
795 | } | |
796 | memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen); | |
797 | inet->cork.flags |= IPCORK_OPT; | |
798 | inet->cork.addr = ipc->addr; | |
799 | } | |
800 | dst_hold(&rt->u.dst); | |
801 | inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path); | |
802 | inet->cork.rt = rt; | |
803 | inet->cork.length = 0; | |
804 | sk->sk_sndmsg_page = NULL; | |
805 | sk->sk_sndmsg_off = 0; | |
806 | if ((exthdrlen = rt->u.dst.header_len) != 0) { | |
807 | length += exthdrlen; | |
808 | transhdrlen += exthdrlen; | |
809 | } | |
810 | } else { | |
811 | rt = inet->cork.rt; | |
812 | if (inet->cork.flags & IPCORK_OPT) | |
813 | opt = inet->cork.opt; | |
814 | ||
815 | transhdrlen = 0; | |
816 | exthdrlen = 0; | |
817 | mtu = inet->cork.fragsize; | |
818 | } | |
819 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
820 | ||
821 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
822 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
823 | ||
824 | if (inet->cork.length + length > 0xFFFF - fragheaderlen) { | |
825 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen); | |
826 | return -EMSGSIZE; | |
827 | } | |
828 | ||
829 | /* | |
830 | * transhdrlen > 0 means that this is the first fragment and we wish | |
831 | * it won't be fragmented in the future. | |
832 | */ | |
833 | if (transhdrlen && | |
834 | length + fragheaderlen <= mtu && | |
8648b305 | 835 | rt->u.dst.dev->features & NETIF_F_ALL_CSUM && |
1da177e4 | 836 | !exthdrlen) |
84fa7933 | 837 | csummode = CHECKSUM_PARTIAL; |
1da177e4 LT |
838 | |
839 | inet->cork.length += length; | |
e89e9cf5 AR |
840 | if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) && |
841 | (rt->u.dst.dev->features & NETIF_F_UFO)) { | |
842 | ||
baa829d8 PM |
843 | err = ip_ufo_append_data(sk, getfrag, from, length, hh_len, |
844 | fragheaderlen, transhdrlen, mtu, | |
845 | flags); | |
846 | if (err) | |
e89e9cf5 | 847 | goto error; |
e89e9cf5 AR |
848 | return 0; |
849 | } | |
1da177e4 LT |
850 | |
851 | /* So, what's going on in the loop below? | |
852 | * | |
853 | * We use calculated fragment length to generate chained skb, | |
854 | * each of segments is IP fragment ready for sending to network after | |
855 | * adding appropriate IP header. | |
856 | */ | |
857 | ||
858 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
859 | goto alloc_new_skb; | |
860 | ||
861 | while (length > 0) { | |
862 | /* Check if the remaining data fits into current packet. */ | |
863 | copy = mtu - skb->len; | |
864 | if (copy < length) | |
865 | copy = maxfraglen - skb->len; | |
866 | if (copy <= 0) { | |
867 | char *data; | |
868 | unsigned int datalen; | |
869 | unsigned int fraglen; | |
870 | unsigned int fraggap; | |
871 | unsigned int alloclen; | |
872 | struct sk_buff *skb_prev; | |
873 | alloc_new_skb: | |
874 | skb_prev = skb; | |
875 | if (skb_prev) | |
876 | fraggap = skb_prev->len - maxfraglen; | |
877 | else | |
878 | fraggap = 0; | |
879 | ||
880 | /* | |
881 | * If remaining data exceeds the mtu, | |
882 | * we know we need more fragment(s). | |
883 | */ | |
884 | datalen = length + fraggap; | |
885 | if (datalen > mtu - fragheaderlen) | |
886 | datalen = maxfraglen - fragheaderlen; | |
887 | fraglen = datalen + fragheaderlen; | |
888 | ||
e905a9ed | 889 | if ((flags & MSG_MORE) && |
1da177e4 LT |
890 | !(rt->u.dst.dev->features&NETIF_F_SG)) |
891 | alloclen = mtu; | |
892 | else | |
893 | alloclen = datalen + fragheaderlen; | |
894 | ||
895 | /* The last fragment gets additional space at tail. | |
896 | * Note, with MSG_MORE we overallocate on fragments, | |
897 | * because we have no idea what fragment will be | |
898 | * the last. | |
899 | */ | |
3d9dd756 | 900 | if (datalen == length + fraggap) |
1da177e4 LT |
901 | alloclen += rt->u.dst.trailer_len; |
902 | ||
903 | if (transhdrlen) { | |
e905a9ed | 904 | skb = sock_alloc_send_skb(sk, |
1da177e4 LT |
905 | alloclen + hh_len + 15, |
906 | (flags & MSG_DONTWAIT), &err); | |
907 | } else { | |
908 | skb = NULL; | |
909 | if (atomic_read(&sk->sk_wmem_alloc) <= | |
910 | 2 * sk->sk_sndbuf) | |
e905a9ed | 911 | skb = sock_wmalloc(sk, |
1da177e4 LT |
912 | alloclen + hh_len + 15, 1, |
913 | sk->sk_allocation); | |
914 | if (unlikely(skb == NULL)) | |
915 | err = -ENOBUFS; | |
916 | } | |
917 | if (skb == NULL) | |
918 | goto error; | |
919 | ||
920 | /* | |
921 | * Fill in the control structures | |
922 | */ | |
923 | skb->ip_summed = csummode; | |
924 | skb->csum = 0; | |
925 | skb_reserve(skb, hh_len); | |
926 | ||
927 | /* | |
928 | * Find where to start putting bytes. | |
929 | */ | |
930 | data = skb_put(skb, fraglen); | |
c14d2450 ACM |
931 | skb_set_network_header(skb, exthdrlen); |
932 | skb->h.raw = skb->nh.raw + fragheaderlen; | |
1da177e4 | 933 | data += fragheaderlen; |
1da177e4 LT |
934 | |
935 | if (fraggap) { | |
936 | skb->csum = skb_copy_and_csum_bits( | |
937 | skb_prev, maxfraglen, | |
938 | data + transhdrlen, fraggap, 0); | |
939 | skb_prev->csum = csum_sub(skb_prev->csum, | |
940 | skb->csum); | |
941 | data += fraggap; | |
e9fa4f7b | 942 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
943 | } |
944 | ||
945 | copy = datalen - transhdrlen - fraggap; | |
946 | if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) { | |
947 | err = -EFAULT; | |
948 | kfree_skb(skb); | |
949 | goto error; | |
950 | } | |
951 | ||
952 | offset += copy; | |
953 | length -= datalen - fraggap; | |
954 | transhdrlen = 0; | |
955 | exthdrlen = 0; | |
956 | csummode = CHECKSUM_NONE; | |
957 | ||
958 | /* | |
959 | * Put the packet on the pending queue. | |
960 | */ | |
961 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
962 | continue; | |
963 | } | |
964 | ||
965 | if (copy > length) | |
966 | copy = length; | |
967 | ||
968 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) { | |
969 | unsigned int off; | |
970 | ||
971 | off = skb->len; | |
e905a9ed | 972 | if (getfrag(from, skb_put(skb, copy), |
1da177e4 LT |
973 | offset, copy, off, skb) < 0) { |
974 | __skb_trim(skb, off); | |
975 | err = -EFAULT; | |
976 | goto error; | |
977 | } | |
978 | } else { | |
979 | int i = skb_shinfo(skb)->nr_frags; | |
980 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1]; | |
981 | struct page *page = sk->sk_sndmsg_page; | |
982 | int off = sk->sk_sndmsg_off; | |
983 | unsigned int left; | |
984 | ||
985 | if (page && (left = PAGE_SIZE - off) > 0) { | |
986 | if (copy >= left) | |
987 | copy = left; | |
988 | if (page != frag->page) { | |
989 | if (i == MAX_SKB_FRAGS) { | |
990 | err = -EMSGSIZE; | |
991 | goto error; | |
992 | } | |
993 | get_page(page); | |
e905a9ed | 994 | skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0); |
1da177e4 LT |
995 | frag = &skb_shinfo(skb)->frags[i]; |
996 | } | |
997 | } else if (i < MAX_SKB_FRAGS) { | |
998 | if (copy > PAGE_SIZE) | |
999 | copy = PAGE_SIZE; | |
1000 | page = alloc_pages(sk->sk_allocation, 0); | |
1001 | if (page == NULL) { | |
1002 | err = -ENOMEM; | |
1003 | goto error; | |
1004 | } | |
1005 | sk->sk_sndmsg_page = page; | |
1006 | sk->sk_sndmsg_off = 0; | |
1007 | ||
1008 | skb_fill_page_desc(skb, i, page, 0, 0); | |
1009 | frag = &skb_shinfo(skb)->frags[i]; | |
1010 | skb->truesize += PAGE_SIZE; | |
1011 | atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); | |
1012 | } else { | |
1013 | err = -EMSGSIZE; | |
1014 | goto error; | |
1015 | } | |
1016 | if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) { | |
1017 | err = -EFAULT; | |
1018 | goto error; | |
1019 | } | |
1020 | sk->sk_sndmsg_off += copy; | |
1021 | frag->size += copy; | |
1022 | skb->len += copy; | |
1023 | skb->data_len += copy; | |
1024 | } | |
1025 | offset += copy; | |
1026 | length -= copy; | |
1027 | } | |
1028 | ||
1029 | return 0; | |
1030 | ||
1031 | error: | |
1032 | inet->cork.length -= length; | |
1033 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
e905a9ed | 1034 | return err; |
1da177e4 LT |
1035 | } |
1036 | ||
1037 | ssize_t ip_append_page(struct sock *sk, struct page *page, | |
1038 | int offset, size_t size, int flags) | |
1039 | { | |
1040 | struct inet_sock *inet = inet_sk(sk); | |
1041 | struct sk_buff *skb; | |
1042 | struct rtable *rt; | |
1043 | struct ip_options *opt = NULL; | |
1044 | int hh_len; | |
1045 | int mtu; | |
1046 | int len; | |
1047 | int err; | |
1048 | unsigned int maxfraglen, fragheaderlen, fraggap; | |
1049 | ||
1050 | if (inet->hdrincl) | |
1051 | return -EPERM; | |
1052 | ||
1053 | if (flags&MSG_PROBE) | |
1054 | return 0; | |
1055 | ||
1056 | if (skb_queue_empty(&sk->sk_write_queue)) | |
1057 | return -EINVAL; | |
1058 | ||
1059 | rt = inet->cork.rt; | |
1060 | if (inet->cork.flags & IPCORK_OPT) | |
1061 | opt = inet->cork.opt; | |
1062 | ||
1063 | if (!(rt->u.dst.dev->features&NETIF_F_SG)) | |
1064 | return -EOPNOTSUPP; | |
1065 | ||
1066 | hh_len = LL_RESERVED_SPACE(rt->u.dst.dev); | |
1067 | mtu = inet->cork.fragsize; | |
1068 | ||
1069 | fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0); | |
1070 | maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen; | |
1071 | ||
1072 | if (inet->cork.length + size > 0xFFFF - fragheaderlen) { | |
1073 | ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu); | |
1074 | return -EMSGSIZE; | |
1075 | } | |
1076 | ||
1077 | if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) | |
1078 | return -EINVAL; | |
1079 | ||
1080 | inet->cork.length += size; | |
e89e9cf5 | 1081 | if ((sk->sk_protocol == IPPROTO_UDP) && |
7967168c HX |
1082 | (rt->u.dst.dev->features & NETIF_F_UFO)) { |
1083 | skb_shinfo(skb)->gso_size = mtu - fragheaderlen; | |
f83ef8c0 | 1084 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
7967168c | 1085 | } |
e89e9cf5 | 1086 | |
1da177e4 LT |
1087 | |
1088 | while (size > 0) { | |
1089 | int i; | |
1090 | ||
89114afd | 1091 | if (skb_is_gso(skb)) |
e89e9cf5 AR |
1092 | len = size; |
1093 | else { | |
1094 | ||
1095 | /* Check if the remaining data fits into current packet. */ | |
1096 | len = mtu - skb->len; | |
1097 | if (len < size) | |
1098 | len = maxfraglen - skb->len; | |
1099 | } | |
1da177e4 LT |
1100 | if (len <= 0) { |
1101 | struct sk_buff *skb_prev; | |
1da177e4 LT |
1102 | int alloclen; |
1103 | ||
1104 | skb_prev = skb; | |
0d0d2bba | 1105 | fraggap = skb_prev->len - maxfraglen; |
1da177e4 LT |
1106 | |
1107 | alloclen = fragheaderlen + hh_len + fraggap + 15; | |
1108 | skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation); | |
1109 | if (unlikely(!skb)) { | |
1110 | err = -ENOBUFS; | |
1111 | goto error; | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Fill in the control structures | |
1116 | */ | |
1117 | skb->ip_summed = CHECKSUM_NONE; | |
1118 | skb->csum = 0; | |
1119 | skb_reserve(skb, hh_len); | |
1120 | ||
1121 | /* | |
1122 | * Find where to start putting bytes. | |
1123 | */ | |
967b05f6 | 1124 | skb_put(skb, fragheaderlen + fraggap); |
2ca9e6f2 | 1125 | skb_reset_network_header(skb); |
967b05f6 | 1126 | skb->h.raw = skb->nh.raw + fragheaderlen; |
1da177e4 LT |
1127 | |
1128 | if (fraggap) { | |
967b05f6 ACM |
1129 | skb->csum = skb_copy_and_csum_bits(skb_prev, |
1130 | maxfraglen, | |
1131 | skb->h.raw, | |
1132 | fraggap, 0); | |
1da177e4 LT |
1133 | skb_prev->csum = csum_sub(skb_prev->csum, |
1134 | skb->csum); | |
e9fa4f7b | 1135 | pskb_trim_unique(skb_prev, maxfraglen); |
1da177e4 LT |
1136 | } |
1137 | ||
1138 | /* | |
1139 | * Put the packet on the pending queue. | |
1140 | */ | |
1141 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1142 | continue; | |
1143 | } | |
1144 | ||
1145 | i = skb_shinfo(skb)->nr_frags; | |
1146 | if (len > size) | |
1147 | len = size; | |
1148 | if (skb_can_coalesce(skb, i, page, offset)) { | |
1149 | skb_shinfo(skb)->frags[i-1].size += len; | |
1150 | } else if (i < MAX_SKB_FRAGS) { | |
1151 | get_page(page); | |
1152 | skb_fill_page_desc(skb, i, page, offset, len); | |
1153 | } else { | |
1154 | err = -EMSGSIZE; | |
1155 | goto error; | |
1156 | } | |
1157 | ||
1158 | if (skb->ip_summed == CHECKSUM_NONE) { | |
44bb9363 | 1159 | __wsum csum; |
1da177e4 LT |
1160 | csum = csum_page(page, offset, len); |
1161 | skb->csum = csum_block_add(skb->csum, csum, skb->len); | |
1162 | } | |
1163 | ||
1164 | skb->len += len; | |
1165 | skb->data_len += len; | |
1166 | offset += len; | |
1167 | size -= len; | |
1168 | } | |
1169 | return 0; | |
1170 | ||
1171 | error: | |
1172 | inet->cork.length -= size; | |
1173 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1174 | return err; | |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * Combined all pending IP fragments on the socket as one IP datagram | |
1179 | * and push them out. | |
1180 | */ | |
1181 | int ip_push_pending_frames(struct sock *sk) | |
1182 | { | |
1183 | struct sk_buff *skb, *tmp_skb; | |
1184 | struct sk_buff **tail_skb; | |
1185 | struct inet_sock *inet = inet_sk(sk); | |
1186 | struct ip_options *opt = NULL; | |
1187 | struct rtable *rt = inet->cork.rt; | |
1188 | struct iphdr *iph; | |
76ab608d | 1189 | __be16 df = 0; |
1da177e4 LT |
1190 | __u8 ttl; |
1191 | int err = 0; | |
1192 | ||
1193 | if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL) | |
1194 | goto out; | |
1195 | tail_skb = &(skb_shinfo(skb)->frag_list); | |
1196 | ||
1197 | /* move skb->data to ip header from ext header */ | |
d56f90a7 | 1198 | if (skb->data < skb_network_header(skb)) |
bbe735e4 | 1199 | __skb_pull(skb, skb_network_offset(skb)); |
1da177e4 LT |
1200 | while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) { |
1201 | __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw); | |
1202 | *tail_skb = tmp_skb; | |
1203 | tail_skb = &(tmp_skb->next); | |
1204 | skb->len += tmp_skb->len; | |
1205 | skb->data_len += tmp_skb->len; | |
1206 | skb->truesize += tmp_skb->truesize; | |
1207 | __sock_put(tmp_skb->sk); | |
1208 | tmp_skb->destructor = NULL; | |
1209 | tmp_skb->sk = NULL; | |
1210 | } | |
1211 | ||
1212 | /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow | |
1213 | * to fragment the frame generated here. No matter, what transforms | |
1214 | * how transforms change size of the packet, it will come out. | |
1215 | */ | |
1216 | if (inet->pmtudisc != IP_PMTUDISC_DO) | |
1217 | skb->local_df = 1; | |
1218 | ||
1219 | /* DF bit is set when we want to see DF on outgoing frames. | |
1220 | * If local_df is set too, we still allow to fragment this frame | |
1221 | * locally. */ | |
1222 | if (inet->pmtudisc == IP_PMTUDISC_DO || | |
1223 | (skb->len <= dst_mtu(&rt->u.dst) && | |
1224 | ip_dont_fragment(sk, &rt->u.dst))) | |
1225 | df = htons(IP_DF); | |
1226 | ||
1227 | if (inet->cork.flags & IPCORK_OPT) | |
1228 | opt = inet->cork.opt; | |
1229 | ||
1230 | if (rt->rt_type == RTN_MULTICAST) | |
1231 | ttl = inet->mc_ttl; | |
1232 | else | |
1233 | ttl = ip_select_ttl(inet, &rt->u.dst); | |
1234 | ||
1235 | iph = (struct iphdr *)skb->data; | |
1236 | iph->version = 4; | |
1237 | iph->ihl = 5; | |
1238 | if (opt) { | |
1239 | iph->ihl += opt->optlen>>2; | |
1240 | ip_options_build(skb, opt, inet->cork.addr, rt, 0); | |
1241 | } | |
1242 | iph->tos = inet->tos; | |
1243 | iph->tot_len = htons(skb->len); | |
1244 | iph->frag_off = df; | |
1a55d57b | 1245 | ip_select_ident(iph, &rt->u.dst, sk); |
1da177e4 LT |
1246 | iph->ttl = ttl; |
1247 | iph->protocol = sk->sk_protocol; | |
1248 | iph->saddr = rt->rt_src; | |
1249 | iph->daddr = rt->rt_dst; | |
1250 | ip_send_check(iph); | |
1251 | ||
1252 | skb->priority = sk->sk_priority; | |
1253 | skb->dst = dst_clone(&rt->u.dst); | |
1254 | ||
1255 | /* Netfilter gets whole the not fragmented skb. */ | |
e905a9ed | 1256 | err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, |
1da177e4 LT |
1257 | skb->dst->dev, dst_output); |
1258 | if (err) { | |
1259 | if (err > 0) | |
1260 | err = inet->recverr ? net_xmit_errno(err) : 0; | |
1261 | if (err) | |
1262 | goto error; | |
1263 | } | |
1264 | ||
1265 | out: | |
1266 | inet->cork.flags &= ~IPCORK_OPT; | |
a51482bd JJ |
1267 | kfree(inet->cork.opt); |
1268 | inet->cork.opt = NULL; | |
1da177e4 LT |
1269 | if (inet->cork.rt) { |
1270 | ip_rt_put(inet->cork.rt); | |
1271 | inet->cork.rt = NULL; | |
1272 | } | |
1273 | return err; | |
1274 | ||
1275 | error: | |
1276 | IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS); | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Throw away all pending data on the socket. | |
1282 | */ | |
1283 | void ip_flush_pending_frames(struct sock *sk) | |
1284 | { | |
1285 | struct inet_sock *inet = inet_sk(sk); | |
1286 | struct sk_buff *skb; | |
1287 | ||
1288 | while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL) | |
1289 | kfree_skb(skb); | |
1290 | ||
1291 | inet->cork.flags &= ~IPCORK_OPT; | |
a51482bd JJ |
1292 | kfree(inet->cork.opt); |
1293 | inet->cork.opt = NULL; | |
1da177e4 LT |
1294 | if (inet->cork.rt) { |
1295 | ip_rt_put(inet->cork.rt); | |
1296 | inet->cork.rt = NULL; | |
1297 | } | |
1298 | } | |
1299 | ||
1300 | ||
1301 | /* | |
1302 | * Fetch data from kernel space and fill in checksum if needed. | |
1303 | */ | |
e905a9ed | 1304 | static int ip_reply_glue_bits(void *dptr, char *to, int offset, |
1da177e4 LT |
1305 | int len, int odd, struct sk_buff *skb) |
1306 | { | |
5084205f | 1307 | __wsum csum; |
1da177e4 LT |
1308 | |
1309 | csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0); | |
1310 | skb->csum = csum_block_add(skb->csum, csum, odd); | |
e905a9ed | 1311 | return 0; |
1da177e4 LT |
1312 | } |
1313 | ||
e905a9ed | 1314 | /* |
1da177e4 LT |
1315 | * Generic function to send a packet as reply to another packet. |
1316 | * Used to send TCP resets so far. ICMP should use this function too. | |
1317 | * | |
e905a9ed | 1318 | * Should run single threaded per socket because it uses the sock |
1da177e4 LT |
1319 | * structure to pass arguments. |
1320 | * | |
1321 | * LATER: switch from ip_build_xmit to ip_append_* | |
1322 | */ | |
1323 | void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg, | |
1324 | unsigned int len) | |
1325 | { | |
1326 | struct inet_sock *inet = inet_sk(sk); | |
1327 | struct { | |
1328 | struct ip_options opt; | |
1329 | char data[40]; | |
1330 | } replyopts; | |
1331 | struct ipcm_cookie ipc; | |
3ca3c68e | 1332 | __be32 daddr; |
1da177e4 LT |
1333 | struct rtable *rt = (struct rtable*)skb->dst; |
1334 | ||
1335 | if (ip_options_echo(&replyopts.opt, skb)) | |
1336 | return; | |
1337 | ||
1338 | daddr = ipc.addr = rt->rt_src; | |
1339 | ipc.opt = NULL; | |
1340 | ||
1341 | if (replyopts.opt.optlen) { | |
1342 | ipc.opt = &replyopts.opt; | |
1343 | ||
1344 | if (ipc.opt->srr) | |
1345 | daddr = replyopts.opt.faddr; | |
1346 | } | |
1347 | ||
1348 | { | |
1349 | struct flowi fl = { .nl_u = { .ip4_u = | |
1350 | { .daddr = daddr, | |
1351 | .saddr = rt->rt_spec_dst, | |
eddc9ec5 | 1352 | .tos = RT_TOS(ip_hdr(skb)->tos) } }, |
1da177e4 LT |
1353 | /* Not quite clean, but right. */ |
1354 | .uli_u = { .ports = | |
1355 | { .sport = skb->h.th->dest, | |
e905a9ed | 1356 | .dport = skb->h.th->source } }, |
1da177e4 | 1357 | .proto = sk->sk_protocol }; |
beb8d13b | 1358 | security_skb_classify_flow(skb, &fl); |
1da177e4 LT |
1359 | if (ip_route_output_key(&rt, &fl)) |
1360 | return; | |
1361 | } | |
1362 | ||
1363 | /* And let IP do all the hard work. | |
1364 | ||
1365 | This chunk is not reenterable, hence spinlock. | |
1366 | Note that it uses the fact, that this function is called | |
1367 | with locally disabled BH and that sk cannot be already spinlocked. | |
1368 | */ | |
1369 | bh_lock_sock(sk); | |
eddc9ec5 | 1370 | inet->tos = ip_hdr(skb)->tos; |
1da177e4 | 1371 | sk->sk_priority = skb->priority; |
eddc9ec5 | 1372 | sk->sk_protocol = ip_hdr(skb)->protocol; |
1da177e4 LT |
1373 | ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0, |
1374 | &ipc, rt, MSG_DONTWAIT); | |
1375 | if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) { | |
1376 | if (arg->csumoffset >= 0) | |
d3bc23e7 | 1377 | *((__sum16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum)); |
1da177e4 LT |
1378 | skb->ip_summed = CHECKSUM_NONE; |
1379 | ip_push_pending_frames(sk); | |
1380 | } | |
1381 | ||
1382 | bh_unlock_sock(sk); | |
1383 | ||
1384 | ip_rt_put(rt); | |
1385 | } | |
1386 | ||
1da177e4 LT |
1387 | void __init ip_init(void) |
1388 | { | |
1da177e4 LT |
1389 | ip_rt_init(); |
1390 | inet_initpeers(); | |
1391 | ||
1392 | #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS) | |
1393 | igmp_mc_proc_init(); | |
1394 | #endif | |
1395 | } | |
1396 | ||
1da177e4 LT |
1397 | EXPORT_SYMBOL(ip_generic_getfrag); |
1398 | EXPORT_SYMBOL(ip_queue_xmit); | |
1399 | EXPORT_SYMBOL(ip_send_check); |