blob: 7624fd1d8f9fe8860479db3d3e9d530a7cd69cda [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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.
25 * Bradford Johnson: Fix faulty handling of some frames when
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.
36 * Andi Kleen : Split fast and slow ip_build_xmit path
37 * for decreased register pressure on x86
38 * and more readibility.
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>
52#include <linux/sched.h>
53#include <linux/mm.h>
54#include <linux/string.h>
55#include <linux/errno.h>
56#include <linux/config.h>
57
58#include <linux/socket.h>
59#include <linux/sockios.h>
60#include <linux/in.h>
61#include <linux/inet.h>
62#include <linux/netdevice.h>
63#include <linux/etherdevice.h>
64#include <linux/proc_fs.h>
65#include <linux/stat.h>
66#include <linux/init.h>
67
68#include <net/snmp.h>
69#include <net/ip.h>
70#include <net/protocol.h>
71#include <net/route.h>
Patrick McHardycfacb052006-01-08 22:36:54 -080072#include <net/xfrm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <linux/skbuff.h>
74#include <net/sock.h>
75#include <net/arp.h>
76#include <net/icmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#include <net/checksum.h>
78#include <net/inetpeer.h>
79#include <net/checksum.h>
80#include <linux/igmp.h>
81#include <linux/netfilter_ipv4.h>
82#include <linux/netfilter_bridge.h>
83#include <linux/mroute.h>
84#include <linux/netlink.h>
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -070085#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087int sysctl_ip_default_ttl = IPDEFTTL;
88
89/* Generate a checksum for an outgoing IP datagram. */
90__inline__ void ip_send_check(struct iphdr *iph)
91{
92 iph->check = 0;
93 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
94}
95
96/* dev_loopback_xmit for use with netfilter. */
97static int ip_dev_loopback_xmit(struct sk_buff *newskb)
98{
99 newskb->mac.raw = newskb->data;
100 __skb_pull(newskb, newskb->nh.raw - newskb->data);
101 newskb->pkt_type = PACKET_LOOPBACK;
102 newskb->ip_summed = CHECKSUM_UNNECESSARY;
103 BUG_TRAP(newskb->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 netif_rx(newskb);
105 return 0;
106}
107
108static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
109{
110 int ttl = inet->uc_ttl;
111
112 if (ttl < 0)
113 ttl = dst_metric(dst, RTAX_HOPLIMIT);
114 return ttl;
115}
116
117/*
118 * Add an ip header to a skbuff and send it out.
119 *
120 */
121int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
122 u32 saddr, u32 daddr, struct ip_options *opt)
123{
124 struct inet_sock *inet = inet_sk(sk);
125 struct rtable *rt = (struct rtable *)skb->dst;
126 struct iphdr *iph;
127
128 /* Build the IP header. */
129 if (opt)
130 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen);
131 else
132 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
133
134 iph->version = 4;
135 iph->ihl = 5;
136 iph->tos = inet->tos;
137 if (ip_dont_fragment(sk, &rt->u.dst))
138 iph->frag_off = htons(IP_DF);
139 else
140 iph->frag_off = 0;
141 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
142 iph->daddr = rt->rt_dst;
143 iph->saddr = rt->rt_src;
144 iph->protocol = sk->sk_protocol;
145 iph->tot_len = htons(skb->len);
146 ip_select_ident(iph, &rt->u.dst, sk);
147 skb->nh.iph = iph;
148
149 if (opt && opt->optlen) {
150 iph->ihl += opt->optlen>>2;
151 ip_options_build(skb, opt, daddr, rt, 0);
152 }
153 ip_send_check(iph);
154
155 skb->priority = sk->sk_priority;
156
157 /* Send it out. */
158 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
159 dst_output);
160}
161
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -0700162EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
163
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164static inline int ip_finish_output2(struct sk_buff *skb)
165{
166 struct dst_entry *dst = skb->dst;
167 struct hh_cache *hh = dst->hh;
168 struct net_device *dev = dst->dev;
169 int hh_len = LL_RESERVED_SPACE(dev);
170
171 /* Be paranoid, rather than too clever. */
172 if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
173 struct sk_buff *skb2;
174
175 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
176 if (skb2 == NULL) {
177 kfree_skb(skb);
178 return -ENOMEM;
179 }
180 if (skb->sk)
181 skb_set_owner_w(skb2, skb->sk);
182 kfree_skb(skb);
183 skb = skb2;
184 }
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 if (hh) {
187 int hh_alen;
188
189 read_lock_bh(&hh->hh_lock);
190 hh_alen = HH_DATA_ALIGN(hh->hh_len);
191 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
192 read_unlock_bh(&hh->hh_lock);
193 skb_push(skb, hh->hh_len);
194 return hh->hh_output(skb);
195 } else if (dst->neighbour)
196 return dst->neighbour->output(skb);
197
198 if (net_ratelimit())
199 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
200 kfree_skb(skb);
201 return -EINVAL;
202}
203
Thomas Graf33d043d2005-08-20 17:27:34 -0700204static inline int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800206#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
207 /* Policy lookup after SNAT yielded a new policy */
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800208 if (skb->dst->xfrm != NULL) {
209 IPCB(skb)->flags |= IPSKB_REROUTED;
210 return dst_output(skb);
211 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800212#endif
Herbert Xu79671682006-06-22 02:40:14 -0700213 if (skb->len > dst_mtu(skb->dst) && !skb_shinfo(skb)->gso_size)
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800214 return ip_fragment(skb, ip_finish_output2);
215 else
216 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219int ip_mc_output(struct sk_buff *skb)
220{
221 struct sock *sk = skb->sk;
222 struct rtable *rt = (struct rtable*)skb->dst;
223 struct net_device *dev = rt->u.dst.dev;
224
225 /*
226 * If the indicated interface is up and running, send the packet.
227 */
228 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
229
230 skb->dev = dev;
231 skb->protocol = htons(ETH_P_IP);
232
233 /*
234 * Multicasts are looped back for other local users
235 */
236
237 if (rt->rt_flags&RTCF_MULTICAST) {
238 if ((!sk || inet_sk(sk)->mc_loop)
239#ifdef CONFIG_IP_MROUTE
240 /* Small optimization: do not loopback not local frames,
241 which returned after forwarding; they will be dropped
242 by ip_mr_input in any case.
243 Note, that local frames are looped back to be delivered
244 to local recipients.
245
246 This check is duplicated in ip_mr_input at the moment.
247 */
248 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
249#endif
250 ) {
251 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
252 if (newskb)
253 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
254 newskb->dev,
255 ip_dev_loopback_xmit);
256 }
257
258 /* Multicasts with ttl 0 must not go beyond the host */
259
260 if (skb->nh.iph->ttl == 0) {
261 kfree_skb(skb);
262 return 0;
263 }
264 }
265
266 if (rt->rt_flags&RTCF_BROADCAST) {
267 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
268 if (newskb)
269 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
270 newskb->dev, ip_dev_loopback_xmit);
271 }
272
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800273 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
274 ip_finish_output,
275 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278int ip_output(struct sk_buff *skb)
279{
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800280 struct net_device *dev = skb->dst->dev;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
283
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800284 skb->dev = dev;
285 skb->protocol = htons(ETH_P_IP);
286
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800287 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
288 ip_finish_output,
289 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
292int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
293{
294 struct sock *sk = skb->sk;
295 struct inet_sock *inet = inet_sk(sk);
296 struct ip_options *opt = inet->opt;
297 struct rtable *rt;
298 struct iphdr *iph;
299
300 /* Skip all of this if the packet is already routed,
301 * f.e. by something like SCTP.
302 */
303 rt = (struct rtable *) skb->dst;
304 if (rt != NULL)
305 goto packet_routed;
306
307 /* Make sure we can route this packet. */
308 rt = (struct rtable *)__sk_dst_check(sk, 0);
309 if (rt == NULL) {
310 u32 daddr;
311
312 /* Use correct destination address if we have options. */
313 daddr = inet->daddr;
314 if(opt && opt->srr)
315 daddr = opt->faddr;
316
317 {
318 struct flowi fl = { .oif = sk->sk_bound_dev_if,
319 .nl_u = { .ip4_u =
320 { .daddr = daddr,
321 .saddr = inet->saddr,
322 .tos = RT_CONN_FLAGS(sk) } },
323 .proto = sk->sk_protocol,
324 .uli_u = { .ports =
325 { .sport = inet->sport,
326 .dport = inet->dport } } };
327
328 /* If this fails, retransmit mechanism of transport layer will
329 * keep trying until route appears or the connection times
330 * itself out.
331 */
332 if (ip_route_output_flow(&rt, &fl, sk, 0))
333 goto no_route;
334 }
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -0700335 sk_setup_caps(sk, &rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 }
337 skb->dst = dst_clone(&rt->u.dst);
338
339packet_routed:
340 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
341 goto no_route;
342
343 /* OK, we know where to send it, allocate and build IP header. */
344 iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
345 *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
346 iph->tot_len = htons(skb->len);
347 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
348 iph->frag_off = htons(IP_DF);
349 else
350 iph->frag_off = 0;
351 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
352 iph->protocol = sk->sk_protocol;
353 iph->saddr = rt->rt_src;
354 iph->daddr = rt->rt_dst;
355 skb->nh.iph = iph;
356 /* Transport layer set skb->h.foo itself. */
357
358 if (opt && opt->optlen) {
359 iph->ihl += opt->optlen >> 2;
360 ip_options_build(skb, opt, inet->daddr, rt, 0);
361 }
362
Herbert Xu89f5f0a2005-11-08 09:41:56 -0800363 ip_select_ident_more(iph, &rt->u.dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700364 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366 /* Add an IP checksum. */
367 ip_send_check(iph);
368
369 skb->priority = sk->sk_priority;
370
371 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
372 dst_output);
373
374no_route:
375 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
376 kfree_skb(skb);
377 return -EHOSTUNREACH;
378}
379
380
381static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
382{
383 to->pkt_type = from->pkt_type;
384 to->priority = from->priority;
385 to->protocol = from->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 dst_release(to->dst);
387 to->dst = dst_clone(from->dst);
388 to->dev = from->dev;
389
390 /* Copy the flags to each fragment. */
391 IPCB(to)->flags = IPCB(from)->flags;
392
393#ifdef CONFIG_NET_SCHED
394 to->tc_index = from->tc_index;
395#endif
396#ifdef CONFIG_NETFILTER
397 to->nfmark = from->nfmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 /* Connection association is same as pre-frag packet */
399 nf_conntrack_put(to->nfct);
400 to->nfct = from->nfct;
401 nf_conntrack_get(to->nfct);
402 to->nfctinfo = from->nfctinfo;
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300403#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
404 to->ipvs_property = from->ipvs_property;
405#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406#ifdef CONFIG_BRIDGE_NETFILTER
407 nf_bridge_put(to->nf_bridge);
408 to->nf_bridge = from->nf_bridge;
409 nf_bridge_get(to->nf_bridge);
410#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411#endif
James Morris984bc162006-06-09 00:29:17 -0700412 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413}
414
415/*
416 * This IP datagram is too large to be sent in one piece. Break it up into
417 * smaller pieces (each of size equal to IP header plus
418 * a block of the data of the original IP data part) that will yet fit in a
419 * single device frame, and queue such a frame for sending.
420 */
421
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700422int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
424 struct iphdr *iph;
425 int raw = 0;
426 int ptr;
427 struct net_device *dev;
428 struct sk_buff *skb2;
429 unsigned int mtu, hlen, left, len, ll_rs;
430 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800431 __be16 not_last_frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct rtable *rt = (struct rtable*)skb->dst;
433 int err = 0;
434
435 dev = rt->u.dst.dev;
436
437 /*
438 * Point into the IP datagram header.
439 */
440
441 iph = skb->nh.iph;
442
443 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
444 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
445 htonl(dst_mtu(&rt->u.dst)));
446 kfree_skb(skb);
447 return -EMSGSIZE;
448 }
449
450 /*
451 * Setup starting values.
452 */
453
454 hlen = iph->ihl * 4;
455 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800456 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 /* When frag_list is given, use it. First, check its validity:
459 * some transformers could create wrong frag_list or break existing
460 * one, it is not prohibited. In this case fall back to copying.
461 *
462 * LATER: this step can be merged to real generation of fragments,
463 * we can switch to copy when see the first bad fragment.
464 */
465 if (skb_shinfo(skb)->frag_list) {
466 struct sk_buff *frag;
467 int first_len = skb_pagelen(skb);
468
469 if (first_len - hlen > mtu ||
470 ((first_len - hlen) & 7) ||
471 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
472 skb_cloned(skb))
473 goto slow_path;
474
475 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
476 /* Correct geometry. */
477 if (frag->len > mtu ||
478 ((frag->len & 7) && frag->next) ||
479 skb_headroom(frag) < hlen)
480 goto slow_path;
481
482 /* Partially cloned skb? */
483 if (skb_shared(frag))
484 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700485
486 BUG_ON(frag->sk);
487 if (skb->sk) {
488 sock_hold(skb->sk);
489 frag->sk = skb->sk;
490 frag->destructor = sock_wfree;
491 skb->truesize -= frag->truesize;
492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 }
494
495 /* Everything is OK. Generate! */
496
497 err = 0;
498 offset = 0;
499 frag = skb_shinfo(skb)->frag_list;
500 skb_shinfo(skb)->frag_list = NULL;
501 skb->data_len = first_len - skb_headlen(skb);
502 skb->len = first_len;
503 iph->tot_len = htons(first_len);
504 iph->frag_off = htons(IP_MF);
505 ip_send_check(iph);
506
507 for (;;) {
508 /* Prepare header of the next frame,
509 * before previous one went down. */
510 if (frag) {
511 frag->ip_summed = CHECKSUM_NONE;
512 frag->h.raw = frag->data;
513 frag->nh.raw = __skb_push(frag, hlen);
514 memcpy(frag->nh.raw, iph, hlen);
515 iph = frag->nh.iph;
516 iph->tot_len = htons(frag->len);
517 ip_copy_metadata(frag, skb);
518 if (offset == 0)
519 ip_options_fragment(frag);
520 offset += skb->len - hlen;
521 iph->frag_off = htons(offset>>3);
522 if (frag->next != NULL)
523 iph->frag_off |= htons(IP_MF);
524 /* Ready, complete checksum */
525 ip_send_check(iph);
526 }
527
528 err = output(skb);
529
530 if (err || !frag)
531 break;
532
533 skb = frag;
534 frag = skb->next;
535 skb->next = NULL;
536 }
537
538 if (err == 0) {
539 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
540 return 0;
541 }
542
543 while (frag) {
544 skb = frag->next;
545 kfree_skb(frag);
546 frag = skb;
547 }
548 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
549 return err;
550 }
551
552slow_path:
553 left = skb->len - hlen; /* Space per frame */
554 ptr = raw + hlen; /* Where to start from */
555
556#ifdef CONFIG_BRIDGE_NETFILTER
557 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
558 * we need to make room for the encapsulating header */
559 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
560 mtu -= nf_bridge_pad(skb);
561#else
562 ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
563#endif
564 /*
565 * Fragment the datagram.
566 */
567
568 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
569 not_last_frag = iph->frag_off & htons(IP_MF);
570
571 /*
572 * Keep copying data until we run out.
573 */
574
575 while(left > 0) {
576 len = left;
577 /* IF: it doesn't fit, use 'mtu' - the data space left */
578 if (len > mtu)
579 len = mtu;
580 /* IF: we are not sending upto and including the packet end
581 then align the next start on an eight byte boundary */
582 if (len < left) {
583 len &= ~7;
584 }
585 /*
586 * Allocate buffer.
587 */
588
589 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700590 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 err = -ENOMEM;
592 goto fail;
593 }
594
595 /*
596 * Set up data on packet
597 */
598
599 ip_copy_metadata(skb2, skb);
600 skb_reserve(skb2, ll_rs);
601 skb_put(skb2, len + hlen);
602 skb2->nh.raw = skb2->data;
603 skb2->h.raw = skb2->data + hlen;
604
605 /*
606 * Charge the memory for the fragment to any owner
607 * it might possess
608 */
609
610 if (skb->sk)
611 skb_set_owner_w(skb2, skb->sk);
612
613 /*
614 * Copy the packet header into the new buffer.
615 */
616
617 memcpy(skb2->nh.raw, skb->data, hlen);
618
619 /*
620 * Copy a block of the IP datagram.
621 */
622 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
623 BUG();
624 left -= len;
625
626 /*
627 * Fill in the new header fields.
628 */
629 iph = skb2->nh.iph;
630 iph->frag_off = htons((offset >> 3));
631
632 /* ANK: dirty, but effective trick. Upgrade options only if
633 * the segment to be fragmented was THE FIRST (otherwise,
634 * options are already fixed) and make it ONCE
635 * on the initial skb, so that all the following fragments
636 * will inherit fixed options.
637 */
638 if (offset == 0)
639 ip_options_fragment(skb);
640
641 /*
642 * Added AC : If we are fragmenting a fragment that's not the
643 * last fragment then keep MF on each bit
644 */
645 if (left > 0 || not_last_frag)
646 iph->frag_off |= htons(IP_MF);
647 ptr += len;
648 offset += len;
649
650 /*
651 * Put this fragment into the sending queue.
652 */
653
654 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
655
656 iph->tot_len = htons(len + hlen);
657
658 ip_send_check(iph);
659
660 err = output(skb2);
661 if (err)
662 goto fail;
663 }
664 kfree_skb(skb);
665 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
666 return err;
667
668fail:
669 kfree_skb(skb);
670 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
671 return err;
672}
673
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700674EXPORT_SYMBOL(ip_fragment);
675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676int
677ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
678{
679 struct iovec *iov = from;
680
681 if (skb->ip_summed == CHECKSUM_HW) {
682 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
683 return -EFAULT;
684 } else {
685 unsigned int csum = 0;
686 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
687 return -EFAULT;
688 skb->csum = csum_block_add(skb->csum, csum, odd);
689 }
690 return 0;
691}
692
693static inline unsigned int
694csum_page(struct page *page, int offset, int copy)
695{
696 char *kaddr;
697 unsigned int csum;
698 kaddr = kmap(page);
699 csum = csum_partial(kaddr + offset, copy, 0);
700 kunmap(page);
701 return csum;
702}
703
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800704static inline int ip_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700705 int getfrag(void *from, char *to, int offset, int len,
706 int odd, struct sk_buff *skb),
707 void *from, int length, int hh_len, int fragheaderlen,
708 int transhdrlen, int mtu,unsigned int flags)
709{
710 struct sk_buff *skb;
711 int err;
712
713 /* There is support for UDP fragmentation offload by network
714 * device, so create one single skb packet containing complete
715 * udp datagram
716 */
717 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
718 skb = sock_alloc_send_skb(sk,
719 hh_len + fragheaderlen + transhdrlen + 20,
720 (flags & MSG_DONTWAIT), &err);
721
722 if (skb == NULL)
723 return err;
724
725 /* reserve space for Hardware header */
726 skb_reserve(skb, hh_len);
727
728 /* create space for UDP/IP header */
729 skb_put(skb,fragheaderlen + transhdrlen);
730
731 /* initialize network header pointer */
732 skb->nh.raw = skb->data;
733
734 /* initialize protocol header pointer */
735 skb->h.raw = skb->data + fragheaderlen;
736
737 skb->ip_summed = CHECKSUM_HW;
738 skb->csum = 0;
739 sk->sk_sndmsg_off = 0;
740 }
741
742 err = skb_append_datato_frags(sk,skb, getfrag, from,
743 (length - transhdrlen));
744 if (!err) {
745 /* specify the length of each IP datagram fragment*/
Herbert Xu79671682006-06-22 02:40:14 -0700746 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
747 skb_shinfo(skb)->gso_type = SKB_GSO_UDPV4;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700748 __skb_queue_tail(&sk->sk_write_queue, skb);
749
750 return 0;
751 }
752 /* There is not enough support do UFO ,
753 * so follow normal path
754 */
755 kfree_skb(skb);
756 return err;
757}
758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759/*
760 * ip_append_data() and ip_append_page() can make one large IP datagram
761 * from many pieces of data. Each pieces will be holded on the socket
762 * until ip_push_pending_frames() is called. Each piece can be a page
763 * or non-page data.
764 *
765 * Not only UDP, other transport protocols - e.g. raw sockets - can use
766 * this interface potentially.
767 *
768 * LATER: length must be adjusted by pad at tail, when it is required.
769 */
770int ip_append_data(struct sock *sk,
771 int getfrag(void *from, char *to, int offset, int len,
772 int odd, struct sk_buff *skb),
773 void *from, int length, int transhdrlen,
774 struct ipcm_cookie *ipc, struct rtable *rt,
775 unsigned int flags)
776{
777 struct inet_sock *inet = inet_sk(sk);
778 struct sk_buff *skb;
779
780 struct ip_options *opt = NULL;
781 int hh_len;
782 int exthdrlen;
783 int mtu;
784 int copy;
785 int err;
786 int offset = 0;
787 unsigned int maxfraglen, fragheaderlen;
788 int csummode = CHECKSUM_NONE;
789
790 if (flags&MSG_PROBE)
791 return 0;
792
793 if (skb_queue_empty(&sk->sk_write_queue)) {
794 /*
795 * setup for corking.
796 */
797 opt = ipc->opt;
798 if (opt) {
799 if (inet->cork.opt == NULL) {
800 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
801 if (unlikely(inet->cork.opt == NULL))
802 return -ENOBUFS;
803 }
804 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
805 inet->cork.flags |= IPCORK_OPT;
806 inet->cork.addr = ipc->addr;
807 }
808 dst_hold(&rt->u.dst);
809 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
810 inet->cork.rt = rt;
811 inet->cork.length = 0;
812 sk->sk_sndmsg_page = NULL;
813 sk->sk_sndmsg_off = 0;
814 if ((exthdrlen = rt->u.dst.header_len) != 0) {
815 length += exthdrlen;
816 transhdrlen += exthdrlen;
817 }
818 } else {
819 rt = inet->cork.rt;
820 if (inet->cork.flags & IPCORK_OPT)
821 opt = inet->cork.opt;
822
823 transhdrlen = 0;
824 exthdrlen = 0;
825 mtu = inet->cork.fragsize;
826 }
827 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
828
829 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
830 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
831
832 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
833 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
834 return -EMSGSIZE;
835 }
836
837 /*
838 * transhdrlen > 0 means that this is the first fragment and we wish
839 * it won't be fragmented in the future.
840 */
841 if (transhdrlen &&
842 length + fragheaderlen <= mtu &&
Herbert Xu8648b302006-06-17 22:06:05 -0700843 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 !exthdrlen)
845 csummode = CHECKSUM_HW;
846
847 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700848 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
849 (rt->u.dst.dev->features & NETIF_F_UFO)) {
850
Patrick McHardybaa829d2006-03-12 20:35:12 -0800851 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
852 fragheaderlen, transhdrlen, mtu,
853 flags);
854 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700855 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700856 return 0;
857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 /* So, what's going on in the loop below?
860 *
861 * We use calculated fragment length to generate chained skb,
862 * each of segments is IP fragment ready for sending to network after
863 * adding appropriate IP header.
864 */
865
866 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
867 goto alloc_new_skb;
868
869 while (length > 0) {
870 /* Check if the remaining data fits into current packet. */
871 copy = mtu - skb->len;
872 if (copy < length)
873 copy = maxfraglen - skb->len;
874 if (copy <= 0) {
875 char *data;
876 unsigned int datalen;
877 unsigned int fraglen;
878 unsigned int fraggap;
879 unsigned int alloclen;
880 struct sk_buff *skb_prev;
881alloc_new_skb:
882 skb_prev = skb;
883 if (skb_prev)
884 fraggap = skb_prev->len - maxfraglen;
885 else
886 fraggap = 0;
887
888 /*
889 * If remaining data exceeds the mtu,
890 * we know we need more fragment(s).
891 */
892 datalen = length + fraggap;
893 if (datalen > mtu - fragheaderlen)
894 datalen = maxfraglen - fragheaderlen;
895 fraglen = datalen + fragheaderlen;
896
897 if ((flags & MSG_MORE) &&
898 !(rt->u.dst.dev->features&NETIF_F_SG))
899 alloclen = mtu;
900 else
901 alloclen = datalen + fragheaderlen;
902
903 /* The last fragment gets additional space at tail.
904 * Note, with MSG_MORE we overallocate on fragments,
905 * because we have no idea what fragment will be
906 * the last.
907 */
Zach Brown3d9dd752006-04-14 16:04:18 -0700908 if (datalen == length + fraggap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 alloclen += rt->u.dst.trailer_len;
910
911 if (transhdrlen) {
912 skb = sock_alloc_send_skb(sk,
913 alloclen + hh_len + 15,
914 (flags & MSG_DONTWAIT), &err);
915 } else {
916 skb = NULL;
917 if (atomic_read(&sk->sk_wmem_alloc) <=
918 2 * sk->sk_sndbuf)
919 skb = sock_wmalloc(sk,
920 alloclen + hh_len + 15, 1,
921 sk->sk_allocation);
922 if (unlikely(skb == NULL))
923 err = -ENOBUFS;
924 }
925 if (skb == NULL)
926 goto error;
927
928 /*
929 * Fill in the control structures
930 */
931 skb->ip_summed = csummode;
932 skb->csum = 0;
933 skb_reserve(skb, hh_len);
934
935 /*
936 * Find where to start putting bytes.
937 */
938 data = skb_put(skb, fraglen);
939 skb->nh.raw = data + exthdrlen;
940 data += fragheaderlen;
941 skb->h.raw = data + exthdrlen;
942
943 if (fraggap) {
944 skb->csum = skb_copy_and_csum_bits(
945 skb_prev, maxfraglen,
946 data + transhdrlen, fraggap, 0);
947 skb_prev->csum = csum_sub(skb_prev->csum,
948 skb->csum);
949 data += fraggap;
950 skb_trim(skb_prev, maxfraglen);
951 }
952
953 copy = datalen - transhdrlen - fraggap;
954 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
955 err = -EFAULT;
956 kfree_skb(skb);
957 goto error;
958 }
959
960 offset += copy;
961 length -= datalen - fraggap;
962 transhdrlen = 0;
963 exthdrlen = 0;
964 csummode = CHECKSUM_NONE;
965
966 /*
967 * Put the packet on the pending queue.
968 */
969 __skb_queue_tail(&sk->sk_write_queue, skb);
970 continue;
971 }
972
973 if (copy > length)
974 copy = length;
975
976 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
977 unsigned int off;
978
979 off = skb->len;
980 if (getfrag(from, skb_put(skb, copy),
981 offset, copy, off, skb) < 0) {
982 __skb_trim(skb, off);
983 err = -EFAULT;
984 goto error;
985 }
986 } else {
987 int i = skb_shinfo(skb)->nr_frags;
988 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
989 struct page *page = sk->sk_sndmsg_page;
990 int off = sk->sk_sndmsg_off;
991 unsigned int left;
992
993 if (page && (left = PAGE_SIZE - off) > 0) {
994 if (copy >= left)
995 copy = left;
996 if (page != frag->page) {
997 if (i == MAX_SKB_FRAGS) {
998 err = -EMSGSIZE;
999 goto error;
1000 }
1001 get_page(page);
1002 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1003 frag = &skb_shinfo(skb)->frags[i];
1004 }
1005 } else if (i < MAX_SKB_FRAGS) {
1006 if (copy > PAGE_SIZE)
1007 copy = PAGE_SIZE;
1008 page = alloc_pages(sk->sk_allocation, 0);
1009 if (page == NULL) {
1010 err = -ENOMEM;
1011 goto error;
1012 }
1013 sk->sk_sndmsg_page = page;
1014 sk->sk_sndmsg_off = 0;
1015
1016 skb_fill_page_desc(skb, i, page, 0, 0);
1017 frag = &skb_shinfo(skb)->frags[i];
1018 skb->truesize += PAGE_SIZE;
1019 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1020 } else {
1021 err = -EMSGSIZE;
1022 goto error;
1023 }
1024 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1025 err = -EFAULT;
1026 goto error;
1027 }
1028 sk->sk_sndmsg_off += copy;
1029 frag->size += copy;
1030 skb->len += copy;
1031 skb->data_len += copy;
1032 }
1033 offset += copy;
1034 length -= copy;
1035 }
1036
1037 return 0;
1038
1039error:
1040 inet->cork.length -= length;
1041 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1042 return err;
1043}
1044
1045ssize_t ip_append_page(struct sock *sk, struct page *page,
1046 int offset, size_t size, int flags)
1047{
1048 struct inet_sock *inet = inet_sk(sk);
1049 struct sk_buff *skb;
1050 struct rtable *rt;
1051 struct ip_options *opt = NULL;
1052 int hh_len;
1053 int mtu;
1054 int len;
1055 int err;
1056 unsigned int maxfraglen, fragheaderlen, fraggap;
1057
1058 if (inet->hdrincl)
1059 return -EPERM;
1060
1061 if (flags&MSG_PROBE)
1062 return 0;
1063
1064 if (skb_queue_empty(&sk->sk_write_queue))
1065 return -EINVAL;
1066
1067 rt = inet->cork.rt;
1068 if (inet->cork.flags & IPCORK_OPT)
1069 opt = inet->cork.opt;
1070
1071 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1072 return -EOPNOTSUPP;
1073
1074 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1075 mtu = inet->cork.fragsize;
1076
1077 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1078 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1079
1080 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1081 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1082 return -EMSGSIZE;
1083 }
1084
1085 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1086 return -EINVAL;
1087
1088 inet->cork.length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001089 if ((sk->sk_protocol == IPPROTO_UDP) &&
Herbert Xu79671682006-06-22 02:40:14 -07001090 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1091 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1092 skb_shinfo(skb)->gso_type = SKB_GSO_UDPV4;
1093 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 while (size > 0) {
1097 int i;
1098
Herbert Xu79671682006-06-22 02:40:14 -07001099 if (skb_shinfo(skb)->gso_size)
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001100 len = size;
1101 else {
1102
1103 /* Check if the remaining data fits into current packet. */
1104 len = mtu - skb->len;
1105 if (len < size)
1106 len = maxfraglen - skb->len;
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 if (len <= 0) {
1109 struct sk_buff *skb_prev;
1110 char *data;
1111 struct iphdr *iph;
1112 int alloclen;
1113
1114 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001115 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
1117 alloclen = fragheaderlen + hh_len + fraggap + 15;
1118 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1119 if (unlikely(!skb)) {
1120 err = -ENOBUFS;
1121 goto error;
1122 }
1123
1124 /*
1125 * Fill in the control structures
1126 */
1127 skb->ip_summed = CHECKSUM_NONE;
1128 skb->csum = 0;
1129 skb_reserve(skb, hh_len);
1130
1131 /*
1132 * Find where to start putting bytes.
1133 */
1134 data = skb_put(skb, fragheaderlen + fraggap);
1135 skb->nh.iph = iph = (struct iphdr *)data;
1136 data += fragheaderlen;
1137 skb->h.raw = data;
1138
1139 if (fraggap) {
1140 skb->csum = skb_copy_and_csum_bits(
1141 skb_prev, maxfraglen,
1142 data, fraggap, 0);
1143 skb_prev->csum = csum_sub(skb_prev->csum,
1144 skb->csum);
1145 skb_trim(skb_prev, maxfraglen);
1146 }
1147
1148 /*
1149 * Put the packet on the pending queue.
1150 */
1151 __skb_queue_tail(&sk->sk_write_queue, skb);
1152 continue;
1153 }
1154
1155 i = skb_shinfo(skb)->nr_frags;
1156 if (len > size)
1157 len = size;
1158 if (skb_can_coalesce(skb, i, page, offset)) {
1159 skb_shinfo(skb)->frags[i-1].size += len;
1160 } else if (i < MAX_SKB_FRAGS) {
1161 get_page(page);
1162 skb_fill_page_desc(skb, i, page, offset, len);
1163 } else {
1164 err = -EMSGSIZE;
1165 goto error;
1166 }
1167
1168 if (skb->ip_summed == CHECKSUM_NONE) {
1169 unsigned int csum;
1170 csum = csum_page(page, offset, len);
1171 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1172 }
1173
1174 skb->len += len;
1175 skb->data_len += len;
1176 offset += len;
1177 size -= len;
1178 }
1179 return 0;
1180
1181error:
1182 inet->cork.length -= size;
1183 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1184 return err;
1185}
1186
1187/*
1188 * Combined all pending IP fragments on the socket as one IP datagram
1189 * and push them out.
1190 */
1191int ip_push_pending_frames(struct sock *sk)
1192{
1193 struct sk_buff *skb, *tmp_skb;
1194 struct sk_buff **tail_skb;
1195 struct inet_sock *inet = inet_sk(sk);
1196 struct ip_options *opt = NULL;
1197 struct rtable *rt = inet->cork.rt;
1198 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001199 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 __u8 ttl;
1201 int err = 0;
1202
1203 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1204 goto out;
1205 tail_skb = &(skb_shinfo(skb)->frag_list);
1206
1207 /* move skb->data to ip header from ext header */
1208 if (skb->data < skb->nh.raw)
1209 __skb_pull(skb, skb->nh.raw - skb->data);
1210 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1211 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1212 *tail_skb = tmp_skb;
1213 tail_skb = &(tmp_skb->next);
1214 skb->len += tmp_skb->len;
1215 skb->data_len += tmp_skb->len;
1216 skb->truesize += tmp_skb->truesize;
1217 __sock_put(tmp_skb->sk);
1218 tmp_skb->destructor = NULL;
1219 tmp_skb->sk = NULL;
1220 }
1221
1222 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1223 * to fragment the frame generated here. No matter, what transforms
1224 * how transforms change size of the packet, it will come out.
1225 */
1226 if (inet->pmtudisc != IP_PMTUDISC_DO)
1227 skb->local_df = 1;
1228
1229 /* DF bit is set when we want to see DF on outgoing frames.
1230 * If local_df is set too, we still allow to fragment this frame
1231 * locally. */
1232 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1233 (skb->len <= dst_mtu(&rt->u.dst) &&
1234 ip_dont_fragment(sk, &rt->u.dst)))
1235 df = htons(IP_DF);
1236
1237 if (inet->cork.flags & IPCORK_OPT)
1238 opt = inet->cork.opt;
1239
1240 if (rt->rt_type == RTN_MULTICAST)
1241 ttl = inet->mc_ttl;
1242 else
1243 ttl = ip_select_ttl(inet, &rt->u.dst);
1244
1245 iph = (struct iphdr *)skb->data;
1246 iph->version = 4;
1247 iph->ihl = 5;
1248 if (opt) {
1249 iph->ihl += opt->optlen>>2;
1250 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1251 }
1252 iph->tos = inet->tos;
1253 iph->tot_len = htons(skb->len);
1254 iph->frag_off = df;
Alexey Kuznetsov1a55d572006-03-22 14:27:59 -08001255 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 iph->ttl = ttl;
1257 iph->protocol = sk->sk_protocol;
1258 iph->saddr = rt->rt_src;
1259 iph->daddr = rt->rt_dst;
1260 ip_send_check(iph);
1261
1262 skb->priority = sk->sk_priority;
1263 skb->dst = dst_clone(&rt->u.dst);
1264
1265 /* Netfilter gets whole the not fragmented skb. */
1266 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1267 skb->dst->dev, dst_output);
1268 if (err) {
1269 if (err > 0)
1270 err = inet->recverr ? net_xmit_errno(err) : 0;
1271 if (err)
1272 goto error;
1273 }
1274
1275out:
1276 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001277 kfree(inet->cork.opt);
1278 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (inet->cork.rt) {
1280 ip_rt_put(inet->cork.rt);
1281 inet->cork.rt = NULL;
1282 }
1283 return err;
1284
1285error:
1286 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1287 goto out;
1288}
1289
1290/*
1291 * Throw away all pending data on the socket.
1292 */
1293void ip_flush_pending_frames(struct sock *sk)
1294{
1295 struct inet_sock *inet = inet_sk(sk);
1296 struct sk_buff *skb;
1297
1298 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1299 kfree_skb(skb);
1300
1301 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001302 kfree(inet->cork.opt);
1303 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (inet->cork.rt) {
1305 ip_rt_put(inet->cork.rt);
1306 inet->cork.rt = NULL;
1307 }
1308}
1309
1310
1311/*
1312 * Fetch data from kernel space and fill in checksum if needed.
1313 */
1314static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1315 int len, int odd, struct sk_buff *skb)
1316{
1317 unsigned int csum;
1318
1319 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1320 skb->csum = csum_block_add(skb->csum, csum, odd);
1321 return 0;
1322}
1323
1324/*
1325 * Generic function to send a packet as reply to another packet.
1326 * Used to send TCP resets so far. ICMP should use this function too.
1327 *
1328 * Should run single threaded per socket because it uses the sock
1329 * structure to pass arguments.
1330 *
1331 * LATER: switch from ip_build_xmit to ip_append_*
1332 */
1333void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1334 unsigned int len)
1335{
1336 struct inet_sock *inet = inet_sk(sk);
1337 struct {
1338 struct ip_options opt;
1339 char data[40];
1340 } replyopts;
1341 struct ipcm_cookie ipc;
1342 u32 daddr;
1343 struct rtable *rt = (struct rtable*)skb->dst;
1344
1345 if (ip_options_echo(&replyopts.opt, skb))
1346 return;
1347
1348 daddr = ipc.addr = rt->rt_src;
1349 ipc.opt = NULL;
1350
1351 if (replyopts.opt.optlen) {
1352 ipc.opt = &replyopts.opt;
1353
1354 if (ipc.opt->srr)
1355 daddr = replyopts.opt.faddr;
1356 }
1357
1358 {
1359 struct flowi fl = { .nl_u = { .ip4_u =
1360 { .daddr = daddr,
1361 .saddr = rt->rt_spec_dst,
1362 .tos = RT_TOS(skb->nh.iph->tos) } },
1363 /* Not quite clean, but right. */
1364 .uli_u = { .ports =
1365 { .sport = skb->h.th->dest,
1366 .dport = skb->h.th->source } },
1367 .proto = sk->sk_protocol };
1368 if (ip_route_output_key(&rt, &fl))
1369 return;
1370 }
1371
1372 /* And let IP do all the hard work.
1373
1374 This chunk is not reenterable, hence spinlock.
1375 Note that it uses the fact, that this function is called
1376 with locally disabled BH and that sk cannot be already spinlocked.
1377 */
1378 bh_lock_sock(sk);
1379 inet->tos = skb->nh.iph->tos;
1380 sk->sk_priority = skb->priority;
1381 sk->sk_protocol = skb->nh.iph->protocol;
1382 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1383 &ipc, rt, MSG_DONTWAIT);
1384 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1385 if (arg->csumoffset >= 0)
1386 *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1387 skb->ip_summed = CHECKSUM_NONE;
1388 ip_push_pending_frames(sk);
1389 }
1390
1391 bh_unlock_sock(sk);
1392
1393 ip_rt_put(rt);
1394}
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396void __init ip_init(void)
1397{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 ip_rt_init();
1399 inet_initpeers();
1400
1401#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1402 igmp_mc_proc_init();
1403#endif
1404}
1405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406EXPORT_SYMBOL(ip_generic_getfrag);
1407EXPORT_SYMBOL(ip_queue_xmit);
1408EXPORT_SYMBOL(ip_send_check);