blob: 7c9f9a6421b8f3acdf7ccff1a106402913c75dd9 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
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>
Patrick McHardycfacb052006-01-08 22:36:54 -080071#include <net/xfrm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#include <linux/skbuff.h>
73#include <net/sock.h>
74#include <net/arp.h>
75#include <net/icmp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#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>
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -070084#include <linux/tcp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Linus Torvalds1da177e2005-04-16 15:20:36 -070086int sysctl_ip_default_ttl = IPDEFTTL;
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. */
96static int ip_dev_loopback_xmit(struct sk_buff *newskb)
97{
98 newskb->mac.raw = newskb->data;
99 __skb_pull(newskb, newskb->nh.raw - newskb->data);
100 newskb->pkt_type = PACKET_LOOPBACK;
101 newskb->ip_summed = CHECKSUM_UNNECESSARY;
102 BUG_TRAP(newskb->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 netif_rx(newskb);
104 return 0;
105}
106
107static 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
116/*
117 * Add an ip header to a skbuff and send it out.
118 *
119 */
120int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
121 u32 saddr, u32 daddr, struct ip_options *opt)
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. */
128 if (opt)
129 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr) + opt->optlen);
130 else
131 iph=(struct iphdr *)skb_push(skb,sizeof(struct iphdr));
132
133 iph->version = 4;
134 iph->ihl = 5;
135 iph->tos = inet->tos;
136 if (ip_dont_fragment(sk, &rt->u.dst))
137 iph->frag_off = htons(IP_DF);
138 else
139 iph->frag_off = 0;
140 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
141 iph->daddr = rt->rt_dst;
142 iph->saddr = rt->rt_src;
143 iph->protocol = sk->sk_protocol;
144 iph->tot_len = htons(skb->len);
145 ip_select_ident(iph, &rt->u.dst, sk);
146 skb->nh.iph = iph;
147
148 if (opt && opt->optlen) {
149 iph->ihl += opt->optlen>>2;
150 ip_options_build(skb, opt, daddr, rt, 0);
151 }
152 ip_send_check(iph);
153
154 skb->priority = sk->sk_priority;
155
156 /* Send it out. */
157 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
158 dst_output);
159}
160
Arnaldo Carvalho de Melod8c97a92005-08-09 20:12:12 -0700161EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163static inline int ip_finish_output2(struct sk_buff *skb)
164{
165 struct dst_entry *dst = skb->dst;
166 struct hh_cache *hh = dst->hh;
167 struct net_device *dev = dst->dev;
168 int hh_len = LL_RESERVED_SPACE(dev);
169
170 /* Be paranoid, rather than too clever. */
171 if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
172 struct sk_buff *skb2;
173
174 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
175 if (skb2 == NULL) {
176 kfree_skb(skb);
177 return -ENOMEM;
178 }
179 if (skb->sk)
180 skb_set_owner_w(skb2, skb->sk);
181 kfree_skb(skb);
182 skb = skb2;
183 }
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 if (hh) {
186 int hh_alen;
187
188 read_lock_bh(&hh->hh_lock);
189 hh_alen = HH_DATA_ALIGN(hh->hh_len);
190 memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
191 read_unlock_bh(&hh->hh_lock);
192 skb_push(skb, hh->hh_len);
193 return hh->hh_output(skb);
194 } else if (dst->neighbour)
195 return dst->neighbour->output(skb);
196
197 if (net_ratelimit())
198 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
199 kfree_skb(skb);
200 return -EINVAL;
201}
202
Thomas Graf33d043d2005-08-20 17:27:34 -0700203static inline int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800205#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
206 /* Policy lookup after SNAT yielded a new policy */
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800207 if (skb->dst->xfrm != NULL) {
208 IPCB(skb)->flags |= IPSKB_REROUTED;
209 return dst_output(skb);
210 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800211#endif
Herbert Xu89114af2006-07-08 13:34:32 -0700212 if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800213 return ip_fragment(skb, ip_finish_output2);
214 else
215 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
218int ip_mc_output(struct sk_buff *skb)
219{
220 struct sock *sk = skb->sk;
221 struct rtable *rt = (struct rtable*)skb->dst;
222 struct net_device *dev = rt->u.dst.dev;
223
224 /*
225 * If the indicated interface is up and running, send the packet.
226 */
227 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
228
229 skb->dev = dev;
230 skb->protocol = htons(ETH_P_IP);
231
232 /*
233 * Multicasts are looped back for other local users
234 */
235
236 if (rt->rt_flags&RTCF_MULTICAST) {
237 if ((!sk || inet_sk(sk)->mc_loop)
238#ifdef CONFIG_IP_MROUTE
239 /* Small optimization: do not loopback not local frames,
240 which returned after forwarding; they will be dropped
241 by ip_mr_input in any case.
242 Note, that local frames are looped back to be delivered
243 to local recipients.
244
245 This check is duplicated in ip_mr_input at the moment.
246 */
247 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
248#endif
249 ) {
250 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
251 if (newskb)
252 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
253 newskb->dev,
254 ip_dev_loopback_xmit);
255 }
256
257 /* Multicasts with ttl 0 must not go beyond the host */
258
259 if (skb->nh.iph->ttl == 0) {
260 kfree_skb(skb);
261 return 0;
262 }
263 }
264
265 if (rt->rt_flags&RTCF_BROADCAST) {
266 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
267 if (newskb)
268 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
269 newskb->dev, ip_dev_loopback_xmit);
270 }
271
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800272 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
273 ip_finish_output,
274 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275}
276
277int ip_output(struct sk_buff *skb)
278{
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800279 struct net_device *dev = skb->dst->dev;
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
282
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800283 skb->dev = dev;
284 skb->protocol = htons(ETH_P_IP);
285
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800286 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
287 ip_finish_output,
288 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
291int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
292{
293 struct sock *sk = skb->sk;
294 struct inet_sock *inet = inet_sk(sk);
295 struct ip_options *opt = inet->opt;
296 struct rtable *rt;
297 struct iphdr *iph;
298
299 /* Skip all of this if the packet is already routed,
300 * f.e. by something like SCTP.
301 */
302 rt = (struct rtable *) skb->dst;
303 if (rt != NULL)
304 goto packet_routed;
305
306 /* Make sure we can route this packet. */
307 rt = (struct rtable *)__sk_dst_check(sk, 0);
308 if (rt == NULL) {
309 u32 daddr;
310
311 /* Use correct destination address if we have options. */
312 daddr = inet->daddr;
313 if(opt && opt->srr)
314 daddr = opt->faddr;
315
316 {
317 struct flowi fl = { .oif = sk->sk_bound_dev_if,
318 .nl_u = { .ip4_u =
319 { .daddr = daddr,
320 .saddr = inet->saddr,
321 .tos = RT_CONN_FLAGS(sk) } },
322 .proto = sk->sk_protocol,
323 .uli_u = { .ports =
324 { .sport = inet->sport,
325 .dport = inet->dport } } };
326
327 /* If this fails, retransmit mechanism of transport layer will
328 * keep trying until route appears or the connection times
329 * itself out.
330 */
331 if (ip_route_output_flow(&rt, &fl, sk, 0))
332 goto no_route;
333 }
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -0700334 sk_setup_caps(sk, &rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
336 skb->dst = dst_clone(&rt->u.dst);
337
338packet_routed:
339 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
340 goto no_route;
341
342 /* OK, we know where to send it, allocate and build IP header. */
343 iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
344 *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
345 iph->tot_len = htons(skb->len);
346 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
347 iph->frag_off = htons(IP_DF);
348 else
349 iph->frag_off = 0;
350 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
351 iph->protocol = sk->sk_protocol;
352 iph->saddr = rt->rt_src;
353 iph->daddr = rt->rt_dst;
354 skb->nh.iph = iph;
355 /* Transport layer set skb->h.foo itself. */
356
357 if (opt && opt->optlen) {
358 iph->ihl += opt->optlen >> 2;
359 ip_options_build(skb, opt, inet->daddr, rt, 0);
360 }
361
Herbert Xu89f5f0a2005-11-08 09:41:56 -0800362 ip_select_ident_more(iph, &rt->u.dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700363 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 /* Add an IP checksum. */
366 ip_send_check(iph);
367
368 skb->priority = sk->sk_priority;
369
370 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
371 dst_output);
372
373no_route:
374 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
375 kfree_skb(skb);
376 return -EHOSTUNREACH;
377}
378
379
380static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
381{
382 to->pkt_type = from->pkt_type;
383 to->priority = from->priority;
384 to->protocol = from->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 dst_release(to->dst);
386 to->dst = dst_clone(from->dst);
387 to->dev = from->dev;
388
389 /* Copy the flags to each fragment. */
390 IPCB(to)->flags = IPCB(from)->flags;
391
392#ifdef CONFIG_NET_SCHED
393 to->tc_index = from->tc_index;
394#endif
395#ifdef CONFIG_NETFILTER
396 to->nfmark = from->nfmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 /* Connection association is same as pre-frag packet */
398 nf_conntrack_put(to->nfct);
399 to->nfct = from->nfct;
400 nf_conntrack_get(to->nfct);
401 to->nfctinfo = from->nfctinfo;
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300402#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
403 to->ipvs_property = from->ipvs_property;
404#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405#ifdef CONFIG_BRIDGE_NETFILTER
406 nf_bridge_put(to->nf_bridge);
407 to->nf_bridge = from->nf_bridge;
408 nf_bridge_get(to->nf_bridge);
409#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410#endif
James Morris984bc162006-06-09 00:29:17 -0700411 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413
414/*
415 * This IP datagram is too large to be sent in one piece. Break it up into
416 * smaller pieces (each of size equal to IP header plus
417 * a block of the data of the original IP data part) that will yet fit in a
418 * single device frame, and queue such a frame for sending.
419 */
420
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700421int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
423 struct iphdr *iph;
424 int raw = 0;
425 int ptr;
426 struct net_device *dev;
427 struct sk_buff *skb2;
428 unsigned int mtu, hlen, left, len, ll_rs;
429 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800430 __be16 not_last_frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 struct rtable *rt = (struct rtable*)skb->dst;
432 int err = 0;
433
434 dev = rt->u.dst.dev;
435
436 /*
437 * Point into the IP datagram header.
438 */
439
440 iph = skb->nh.iph;
441
442 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
443 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
444 htonl(dst_mtu(&rt->u.dst)));
445 kfree_skb(skb);
446 return -EMSGSIZE;
447 }
448
449 /*
450 * Setup starting values.
451 */
452
453 hlen = iph->ihl * 4;
454 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800455 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 /* When frag_list is given, use it. First, check its validity:
458 * some transformers could create wrong frag_list or break existing
459 * one, it is not prohibited. In this case fall back to copying.
460 *
461 * LATER: this step can be merged to real generation of fragments,
462 * we can switch to copy when see the first bad fragment.
463 */
464 if (skb_shinfo(skb)->frag_list) {
465 struct sk_buff *frag;
466 int first_len = skb_pagelen(skb);
467
468 if (first_len - hlen > mtu ||
469 ((first_len - hlen) & 7) ||
470 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
471 skb_cloned(skb))
472 goto slow_path;
473
474 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
475 /* Correct geometry. */
476 if (frag->len > mtu ||
477 ((frag->len & 7) && frag->next) ||
478 skb_headroom(frag) < hlen)
479 goto slow_path;
480
481 /* Partially cloned skb? */
482 if (skb_shared(frag))
483 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700484
485 BUG_ON(frag->sk);
486 if (skb->sk) {
487 sock_hold(skb->sk);
488 frag->sk = skb->sk;
489 frag->destructor = sock_wfree;
490 skb->truesize -= frag->truesize;
491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 }
493
494 /* Everything is OK. Generate! */
495
496 err = 0;
497 offset = 0;
498 frag = skb_shinfo(skb)->frag_list;
499 skb_shinfo(skb)->frag_list = NULL;
500 skb->data_len = first_len - skb_headlen(skb);
501 skb->len = first_len;
502 iph->tot_len = htons(first_len);
503 iph->frag_off = htons(IP_MF);
504 ip_send_check(iph);
505
506 for (;;) {
507 /* Prepare header of the next frame,
508 * before previous one went down. */
509 if (frag) {
510 frag->ip_summed = CHECKSUM_NONE;
511 frag->h.raw = frag->data;
512 frag->nh.raw = __skb_push(frag, hlen);
513 memcpy(frag->nh.raw, iph, hlen);
514 iph = frag->nh.iph;
515 iph->tot_len = htons(frag->len);
516 ip_copy_metadata(frag, skb);
517 if (offset == 0)
518 ip_options_fragment(frag);
519 offset += skb->len - hlen;
520 iph->frag_off = htons(offset>>3);
521 if (frag->next != NULL)
522 iph->frag_off |= htons(IP_MF);
523 /* Ready, complete checksum */
524 ip_send_check(iph);
525 }
526
527 err = output(skb);
528
529 if (err || !frag)
530 break;
531
532 skb = frag;
533 frag = skb->next;
534 skb->next = NULL;
535 }
536
537 if (err == 0) {
538 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
539 return 0;
540 }
541
542 while (frag) {
543 skb = frag->next;
544 kfree_skb(frag);
545 frag = skb;
546 }
547 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
548 return err;
549 }
550
551slow_path:
552 left = skb->len - hlen; /* Space per frame */
553 ptr = raw + hlen; /* Where to start from */
554
555#ifdef CONFIG_BRIDGE_NETFILTER
556 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
557 * we need to make room for the encapsulating header */
558 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
559 mtu -= nf_bridge_pad(skb);
560#else
561 ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
562#endif
563 /*
564 * Fragment the datagram.
565 */
566
567 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
568 not_last_frag = iph->frag_off & htons(IP_MF);
569
570 /*
571 * Keep copying data until we run out.
572 */
573
574 while(left > 0) {
575 len = left;
576 /* IF: it doesn't fit, use 'mtu' - the data space left */
577 if (len > mtu)
578 len = mtu;
579 /* IF: we are not sending upto and including the packet end
580 then align the next start on an eight byte boundary */
581 if (len < left) {
582 len &= ~7;
583 }
584 /*
585 * Allocate buffer.
586 */
587
588 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700589 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 err = -ENOMEM;
591 goto fail;
592 }
593
594 /*
595 * Set up data on packet
596 */
597
598 ip_copy_metadata(skb2, skb);
599 skb_reserve(skb2, ll_rs);
600 skb_put(skb2, len + hlen);
601 skb2->nh.raw = skb2->data;
602 skb2->h.raw = skb2->data + hlen;
603
604 /*
605 * Charge the memory for the fragment to any owner
606 * it might possess
607 */
608
609 if (skb->sk)
610 skb_set_owner_w(skb2, skb->sk);
611
612 /*
613 * Copy the packet header into the new buffer.
614 */
615
616 memcpy(skb2->nh.raw, skb->data, hlen);
617
618 /*
619 * Copy a block of the IP datagram.
620 */
621 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
622 BUG();
623 left -= len;
624
625 /*
626 * Fill in the new header fields.
627 */
628 iph = skb2->nh.iph;
629 iph->frag_off = htons((offset >> 3));
630
631 /* ANK: dirty, but effective trick. Upgrade options only if
632 * the segment to be fragmented was THE FIRST (otherwise,
633 * options are already fixed) and make it ONCE
634 * on the initial skb, so that all the following fragments
635 * will inherit fixed options.
636 */
637 if (offset == 0)
638 ip_options_fragment(skb);
639
640 /*
641 * Added AC : If we are fragmenting a fragment that's not the
642 * last fragment then keep MF on each bit
643 */
644 if (left > 0 || not_last_frag)
645 iph->frag_off |= htons(IP_MF);
646 ptr += len;
647 offset += len;
648
649 /*
650 * Put this fragment into the sending queue.
651 */
652
653 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
654
655 iph->tot_len = htons(len + hlen);
656
657 ip_send_check(iph);
658
659 err = output(skb2);
660 if (err)
661 goto fail;
662 }
663 kfree_skb(skb);
664 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
665 return err;
666
667fail:
668 kfree_skb(skb);
669 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
670 return err;
671}
672
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700673EXPORT_SYMBOL(ip_fragment);
674
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675int
676ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
677{
678 struct iovec *iov = from;
679
680 if (skb->ip_summed == CHECKSUM_HW) {
681 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
682 return -EFAULT;
683 } else {
684 unsigned int csum = 0;
685 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
686 return -EFAULT;
687 skb->csum = csum_block_add(skb->csum, csum, odd);
688 }
689 return 0;
690}
691
692static inline unsigned int
693csum_page(struct page *page, int offset, int copy)
694{
695 char *kaddr;
696 unsigned int csum;
697 kaddr = kmap(page);
698 csum = csum_partial(kaddr + offset, copy, 0);
699 kunmap(page);
700 return csum;
701}
702
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800703static inline int ip_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700704 int getfrag(void *from, char *to, int offset, int len,
705 int odd, struct sk_buff *skb),
706 void *from, int length, int hh_len, int fragheaderlen,
707 int transhdrlen, int mtu,unsigned int flags)
708{
709 struct sk_buff *skb;
710 int err;
711
712 /* There is support for UDP fragmentation offload by network
713 * device, so create one single skb packet containing complete
714 * udp datagram
715 */
716 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
717 skb = sock_alloc_send_skb(sk,
718 hh_len + fragheaderlen + transhdrlen + 20,
719 (flags & MSG_DONTWAIT), &err);
720
721 if (skb == NULL)
722 return err;
723
724 /* reserve space for Hardware header */
725 skb_reserve(skb, hh_len);
726
727 /* create space for UDP/IP header */
728 skb_put(skb,fragheaderlen + transhdrlen);
729
730 /* initialize network header pointer */
731 skb->nh.raw = skb->data;
732
733 /* initialize protocol header pointer */
734 skb->h.raw = skb->data + fragheaderlen;
735
736 skb->ip_summed = CHECKSUM_HW;
737 skb->csum = 0;
738 sk->sk_sndmsg_off = 0;
739 }
740
741 err = skb_append_datato_frags(sk,skb, getfrag, from,
742 (length - transhdrlen));
743 if (!err) {
744 /* specify the length of each IP datagram fragment*/
Herbert Xu79671682006-06-22 02:40:14 -0700745 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700746 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700747 __skb_queue_tail(&sk->sk_write_queue, skb);
748
749 return 0;
750 }
751 /* There is not enough support do UFO ,
752 * so follow normal path
753 */
754 kfree_skb(skb);
755 return err;
756}
757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758/*
759 * ip_append_data() and ip_append_page() can make one large IP datagram
760 * from many pieces of data. Each pieces will be holded on the socket
761 * until ip_push_pending_frames() is called. Each piece can be a page
762 * or non-page data.
763 *
764 * Not only UDP, other transport protocols - e.g. raw sockets - can use
765 * this interface potentially.
766 *
767 * LATER: length must be adjusted by pad at tail, when it is required.
768 */
769int ip_append_data(struct sock *sk,
770 int getfrag(void *from, char *to, int offset, int len,
771 int odd, struct sk_buff *skb),
772 void *from, int length, int transhdrlen,
773 struct ipcm_cookie *ipc, struct rtable *rt,
774 unsigned int flags)
775{
776 struct inet_sock *inet = inet_sk(sk);
777 struct sk_buff *skb;
778
779 struct ip_options *opt = NULL;
780 int hh_len;
781 int exthdrlen;
782 int mtu;
783 int copy;
784 int err;
785 int offset = 0;
786 unsigned int maxfraglen, fragheaderlen;
787 int csummode = CHECKSUM_NONE;
788
789 if (flags&MSG_PROBE)
790 return 0;
791
792 if (skb_queue_empty(&sk->sk_write_queue)) {
793 /*
794 * setup for corking.
795 */
796 opt = ipc->opt;
797 if (opt) {
798 if (inet->cork.opt == NULL) {
799 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
800 if (unlikely(inet->cork.opt == NULL))
801 return -ENOBUFS;
802 }
803 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
804 inet->cork.flags |= IPCORK_OPT;
805 inet->cork.addr = ipc->addr;
806 }
807 dst_hold(&rt->u.dst);
808 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
809 inet->cork.rt = rt;
810 inet->cork.length = 0;
811 sk->sk_sndmsg_page = NULL;
812 sk->sk_sndmsg_off = 0;
813 if ((exthdrlen = rt->u.dst.header_len) != 0) {
814 length += exthdrlen;
815 transhdrlen += exthdrlen;
816 }
817 } else {
818 rt = inet->cork.rt;
819 if (inet->cork.flags & IPCORK_OPT)
820 opt = inet->cork.opt;
821
822 transhdrlen = 0;
823 exthdrlen = 0;
824 mtu = inet->cork.fragsize;
825 }
826 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
827
828 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
829 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
830
831 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
832 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
833 return -EMSGSIZE;
834 }
835
836 /*
837 * transhdrlen > 0 means that this is the first fragment and we wish
838 * it won't be fragmented in the future.
839 */
840 if (transhdrlen &&
841 length + fragheaderlen <= mtu &&
Herbert Xu8648b302006-06-17 22:06:05 -0700842 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 !exthdrlen)
844 csummode = CHECKSUM_HW;
845
846 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700847 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
848 (rt->u.dst.dev->features & NETIF_F_UFO)) {
849
Patrick McHardybaa829d2006-03-12 20:35:12 -0800850 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
851 fragheaderlen, transhdrlen, mtu,
852 flags);
853 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700854 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700855 return 0;
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
858 /* So, what's going on in the loop below?
859 *
860 * We use calculated fragment length to generate chained skb,
861 * each of segments is IP fragment ready for sending to network after
862 * adding appropriate IP header.
863 */
864
865 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
866 goto alloc_new_skb;
867
868 while (length > 0) {
869 /* Check if the remaining data fits into current packet. */
870 copy = mtu - skb->len;
871 if (copy < length)
872 copy = maxfraglen - skb->len;
873 if (copy <= 0) {
874 char *data;
875 unsigned int datalen;
876 unsigned int fraglen;
877 unsigned int fraggap;
878 unsigned int alloclen;
879 struct sk_buff *skb_prev;
880alloc_new_skb:
881 skb_prev = skb;
882 if (skb_prev)
883 fraggap = skb_prev->len - maxfraglen;
884 else
885 fraggap = 0;
886
887 /*
888 * If remaining data exceeds the mtu,
889 * we know we need more fragment(s).
890 */
891 datalen = length + fraggap;
892 if (datalen > mtu - fragheaderlen)
893 datalen = maxfraglen - fragheaderlen;
894 fraglen = datalen + fragheaderlen;
895
896 if ((flags & MSG_MORE) &&
897 !(rt->u.dst.dev->features&NETIF_F_SG))
898 alloclen = mtu;
899 else
900 alloclen = datalen + fragheaderlen;
901
902 /* The last fragment gets additional space at tail.
903 * Note, with MSG_MORE we overallocate on fragments,
904 * because we have no idea what fragment will be
905 * the last.
906 */
Zach Brown3d9dd752006-04-14 16:04:18 -0700907 if (datalen == length + fraggap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 alloclen += rt->u.dst.trailer_len;
909
910 if (transhdrlen) {
911 skb = sock_alloc_send_skb(sk,
912 alloclen + hh_len + 15,
913 (flags & MSG_DONTWAIT), &err);
914 } else {
915 skb = NULL;
916 if (atomic_read(&sk->sk_wmem_alloc) <=
917 2 * sk->sk_sndbuf)
918 skb = sock_wmalloc(sk,
919 alloclen + hh_len + 15, 1,
920 sk->sk_allocation);
921 if (unlikely(skb == NULL))
922 err = -ENOBUFS;
923 }
924 if (skb == NULL)
925 goto error;
926
927 /*
928 * Fill in the control structures
929 */
930 skb->ip_summed = csummode;
931 skb->csum = 0;
932 skb_reserve(skb, hh_len);
933
934 /*
935 * Find where to start putting bytes.
936 */
937 data = skb_put(skb, fraglen);
938 skb->nh.raw = data + exthdrlen;
939 data += fragheaderlen;
940 skb->h.raw = data + exthdrlen;
941
942 if (fraggap) {
943 skb->csum = skb_copy_and_csum_bits(
944 skb_prev, maxfraglen,
945 data + transhdrlen, fraggap, 0);
946 skb_prev->csum = csum_sub(skb_prev->csum,
947 skb->csum);
948 data += fraggap;
949 skb_trim(skb_prev, maxfraglen);
950 }
951
952 copy = datalen - transhdrlen - fraggap;
953 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
954 err = -EFAULT;
955 kfree_skb(skb);
956 goto error;
957 }
958
959 offset += copy;
960 length -= datalen - fraggap;
961 transhdrlen = 0;
962 exthdrlen = 0;
963 csummode = CHECKSUM_NONE;
964
965 /*
966 * Put the packet on the pending queue.
967 */
968 __skb_queue_tail(&sk->sk_write_queue, skb);
969 continue;
970 }
971
972 if (copy > length)
973 copy = length;
974
975 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
976 unsigned int off;
977
978 off = skb->len;
979 if (getfrag(from, skb_put(skb, copy),
980 offset, copy, off, skb) < 0) {
981 __skb_trim(skb, off);
982 err = -EFAULT;
983 goto error;
984 }
985 } else {
986 int i = skb_shinfo(skb)->nr_frags;
987 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
988 struct page *page = sk->sk_sndmsg_page;
989 int off = sk->sk_sndmsg_off;
990 unsigned int left;
991
992 if (page && (left = PAGE_SIZE - off) > 0) {
993 if (copy >= left)
994 copy = left;
995 if (page != frag->page) {
996 if (i == MAX_SKB_FRAGS) {
997 err = -EMSGSIZE;
998 goto error;
999 }
1000 get_page(page);
1001 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1002 frag = &skb_shinfo(skb)->frags[i];
1003 }
1004 } else if (i < MAX_SKB_FRAGS) {
1005 if (copy > PAGE_SIZE)
1006 copy = PAGE_SIZE;
1007 page = alloc_pages(sk->sk_allocation, 0);
1008 if (page == NULL) {
1009 err = -ENOMEM;
1010 goto error;
1011 }
1012 sk->sk_sndmsg_page = page;
1013 sk->sk_sndmsg_off = 0;
1014
1015 skb_fill_page_desc(skb, i, page, 0, 0);
1016 frag = &skb_shinfo(skb)->frags[i];
1017 skb->truesize += PAGE_SIZE;
1018 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1019 } else {
1020 err = -EMSGSIZE;
1021 goto error;
1022 }
1023 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1024 err = -EFAULT;
1025 goto error;
1026 }
1027 sk->sk_sndmsg_off += copy;
1028 frag->size += copy;
1029 skb->len += copy;
1030 skb->data_len += copy;
1031 }
1032 offset += copy;
1033 length -= copy;
1034 }
1035
1036 return 0;
1037
1038error:
1039 inet->cork.length -= length;
1040 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1041 return err;
1042}
1043
1044ssize_t ip_append_page(struct sock *sk, struct page *page,
1045 int offset, size_t size, int flags)
1046{
1047 struct inet_sock *inet = inet_sk(sk);
1048 struct sk_buff *skb;
1049 struct rtable *rt;
1050 struct ip_options *opt = NULL;
1051 int hh_len;
1052 int mtu;
1053 int len;
1054 int err;
1055 unsigned int maxfraglen, fragheaderlen, fraggap;
1056
1057 if (inet->hdrincl)
1058 return -EPERM;
1059
1060 if (flags&MSG_PROBE)
1061 return 0;
1062
1063 if (skb_queue_empty(&sk->sk_write_queue))
1064 return -EINVAL;
1065
1066 rt = inet->cork.rt;
1067 if (inet->cork.flags & IPCORK_OPT)
1068 opt = inet->cork.opt;
1069
1070 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1071 return -EOPNOTSUPP;
1072
1073 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1074 mtu = inet->cork.fragsize;
1075
1076 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1077 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1078
1079 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1080 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1081 return -EMSGSIZE;
1082 }
1083
1084 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1085 return -EINVAL;
1086
1087 inet->cork.length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001088 if ((sk->sk_protocol == IPPROTO_UDP) &&
Herbert Xu79671682006-06-22 02:40:14 -07001089 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1090 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001091 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001092 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 while (size > 0) {
1096 int i;
1097
Herbert Xu89114af2006-07-08 13:34:32 -07001098 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001099 len = size;
1100 else {
1101
1102 /* Check if the remaining data fits into current packet. */
1103 len = mtu - skb->len;
1104 if (len < size)
1105 len = maxfraglen - skb->len;
1106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 if (len <= 0) {
1108 struct sk_buff *skb_prev;
1109 char *data;
1110 struct iphdr *iph;
1111 int alloclen;
1112
1113 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001114 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 alloclen = fragheaderlen + hh_len + fraggap + 15;
1117 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1118 if (unlikely(!skb)) {
1119 err = -ENOBUFS;
1120 goto error;
1121 }
1122
1123 /*
1124 * Fill in the control structures
1125 */
1126 skb->ip_summed = CHECKSUM_NONE;
1127 skb->csum = 0;
1128 skb_reserve(skb, hh_len);
1129
1130 /*
1131 * Find where to start putting bytes.
1132 */
1133 data = skb_put(skb, fragheaderlen + fraggap);
1134 skb->nh.iph = iph = (struct iphdr *)data;
1135 data += fragheaderlen;
1136 skb->h.raw = data;
1137
1138 if (fraggap) {
1139 skb->csum = skb_copy_and_csum_bits(
1140 skb_prev, maxfraglen,
1141 data, fraggap, 0);
1142 skb_prev->csum = csum_sub(skb_prev->csum,
1143 skb->csum);
1144 skb_trim(skb_prev, maxfraglen);
1145 }
1146
1147 /*
1148 * Put the packet on the pending queue.
1149 */
1150 __skb_queue_tail(&sk->sk_write_queue, skb);
1151 continue;
1152 }
1153
1154 i = skb_shinfo(skb)->nr_frags;
1155 if (len > size)
1156 len = size;
1157 if (skb_can_coalesce(skb, i, page, offset)) {
1158 skb_shinfo(skb)->frags[i-1].size += len;
1159 } else if (i < MAX_SKB_FRAGS) {
1160 get_page(page);
1161 skb_fill_page_desc(skb, i, page, offset, len);
1162 } else {
1163 err = -EMSGSIZE;
1164 goto error;
1165 }
1166
1167 if (skb->ip_summed == CHECKSUM_NONE) {
1168 unsigned int csum;
1169 csum = csum_page(page, offset, len);
1170 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1171 }
1172
1173 skb->len += len;
1174 skb->data_len += len;
1175 offset += len;
1176 size -= len;
1177 }
1178 return 0;
1179
1180error:
1181 inet->cork.length -= size;
1182 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1183 return err;
1184}
1185
1186/*
1187 * Combined all pending IP fragments on the socket as one IP datagram
1188 * and push them out.
1189 */
1190int ip_push_pending_frames(struct sock *sk)
1191{
1192 struct sk_buff *skb, *tmp_skb;
1193 struct sk_buff **tail_skb;
1194 struct inet_sock *inet = inet_sk(sk);
1195 struct ip_options *opt = NULL;
1196 struct rtable *rt = inet->cork.rt;
1197 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001198 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 __u8 ttl;
1200 int err = 0;
1201
1202 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1203 goto out;
1204 tail_skb = &(skb_shinfo(skb)->frag_list);
1205
1206 /* move skb->data to ip header from ext header */
1207 if (skb->data < skb->nh.raw)
1208 __skb_pull(skb, skb->nh.raw - skb->data);
1209 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1210 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1211 *tail_skb = tmp_skb;
1212 tail_skb = &(tmp_skb->next);
1213 skb->len += tmp_skb->len;
1214 skb->data_len += tmp_skb->len;
1215 skb->truesize += tmp_skb->truesize;
1216 __sock_put(tmp_skb->sk);
1217 tmp_skb->destructor = NULL;
1218 tmp_skb->sk = NULL;
1219 }
1220
1221 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1222 * to fragment the frame generated here. No matter, what transforms
1223 * how transforms change size of the packet, it will come out.
1224 */
1225 if (inet->pmtudisc != IP_PMTUDISC_DO)
1226 skb->local_df = 1;
1227
1228 /* DF bit is set when we want to see DF on outgoing frames.
1229 * If local_df is set too, we still allow to fragment this frame
1230 * locally. */
1231 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1232 (skb->len <= dst_mtu(&rt->u.dst) &&
1233 ip_dont_fragment(sk, &rt->u.dst)))
1234 df = htons(IP_DF);
1235
1236 if (inet->cork.flags & IPCORK_OPT)
1237 opt = inet->cork.opt;
1238
1239 if (rt->rt_type == RTN_MULTICAST)
1240 ttl = inet->mc_ttl;
1241 else
1242 ttl = ip_select_ttl(inet, &rt->u.dst);
1243
1244 iph = (struct iphdr *)skb->data;
1245 iph->version = 4;
1246 iph->ihl = 5;
1247 if (opt) {
1248 iph->ihl += opt->optlen>>2;
1249 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1250 }
1251 iph->tos = inet->tos;
1252 iph->tot_len = htons(skb->len);
1253 iph->frag_off = df;
Alexey Kuznetsov1a55d572006-03-22 14:27:59 -08001254 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 iph->ttl = ttl;
1256 iph->protocol = sk->sk_protocol;
1257 iph->saddr = rt->rt_src;
1258 iph->daddr = rt->rt_dst;
1259 ip_send_check(iph);
1260
1261 skb->priority = sk->sk_priority;
1262 skb->dst = dst_clone(&rt->u.dst);
1263
1264 /* Netfilter gets whole the not fragmented skb. */
1265 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1266 skb->dst->dev, dst_output);
1267 if (err) {
1268 if (err > 0)
1269 err = inet->recverr ? net_xmit_errno(err) : 0;
1270 if (err)
1271 goto error;
1272 }
1273
1274out:
1275 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001276 kfree(inet->cork.opt);
1277 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (inet->cork.rt) {
1279 ip_rt_put(inet->cork.rt);
1280 inet->cork.rt = NULL;
1281 }
1282 return err;
1283
1284error:
1285 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1286 goto out;
1287}
1288
1289/*
1290 * Throw away all pending data on the socket.
1291 */
1292void ip_flush_pending_frames(struct sock *sk)
1293{
1294 struct inet_sock *inet = inet_sk(sk);
1295 struct sk_buff *skb;
1296
1297 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1298 kfree_skb(skb);
1299
1300 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001301 kfree(inet->cork.opt);
1302 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 if (inet->cork.rt) {
1304 ip_rt_put(inet->cork.rt);
1305 inet->cork.rt = NULL;
1306 }
1307}
1308
1309
1310/*
1311 * Fetch data from kernel space and fill in checksum if needed.
1312 */
1313static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1314 int len, int odd, struct sk_buff *skb)
1315{
1316 unsigned int csum;
1317
1318 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1319 skb->csum = csum_block_add(skb->csum, csum, odd);
1320 return 0;
1321}
1322
1323/*
1324 * Generic function to send a packet as reply to another packet.
1325 * Used to send TCP resets so far. ICMP should use this function too.
1326 *
1327 * Should run single threaded per socket because it uses the sock
1328 * structure to pass arguments.
1329 *
1330 * LATER: switch from ip_build_xmit to ip_append_*
1331 */
1332void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1333 unsigned int len)
1334{
1335 struct inet_sock *inet = inet_sk(sk);
1336 struct {
1337 struct ip_options opt;
1338 char data[40];
1339 } replyopts;
1340 struct ipcm_cookie ipc;
1341 u32 daddr;
1342 struct rtable *rt = (struct rtable*)skb->dst;
1343
1344 if (ip_options_echo(&replyopts.opt, skb))
1345 return;
1346
1347 daddr = ipc.addr = rt->rt_src;
1348 ipc.opt = NULL;
1349
1350 if (replyopts.opt.optlen) {
1351 ipc.opt = &replyopts.opt;
1352
1353 if (ipc.opt->srr)
1354 daddr = replyopts.opt.faddr;
1355 }
1356
1357 {
1358 struct flowi fl = { .nl_u = { .ip4_u =
1359 { .daddr = daddr,
1360 .saddr = rt->rt_spec_dst,
1361 .tos = RT_TOS(skb->nh.iph->tos) } },
1362 /* Not quite clean, but right. */
1363 .uli_u = { .ports =
1364 { .sport = skb->h.th->dest,
1365 .dport = skb->h.th->source } },
1366 .proto = sk->sk_protocol };
1367 if (ip_route_output_key(&rt, &fl))
1368 return;
1369 }
1370
1371 /* And let IP do all the hard work.
1372
1373 This chunk is not reenterable, hence spinlock.
1374 Note that it uses the fact, that this function is called
1375 with locally disabled BH and that sk cannot be already spinlocked.
1376 */
1377 bh_lock_sock(sk);
1378 inet->tos = skb->nh.iph->tos;
1379 sk->sk_priority = skb->priority;
1380 sk->sk_protocol = skb->nh.iph->protocol;
1381 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1382 &ipc, rt, MSG_DONTWAIT);
1383 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1384 if (arg->csumoffset >= 0)
1385 *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1386 skb->ip_summed = CHECKSUM_NONE;
1387 ip_push_pending_frames(sk);
1388 }
1389
1390 bh_unlock_sock(sk);
1391
1392 ip_rt_put(rt);
1393}
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395void __init ip_init(void)
1396{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 ip_rt_init();
1398 inet_initpeers();
1399
1400#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1401 igmp_mc_proc_init();
1402#endif
1403}
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405EXPORT_SYMBOL(ip_generic_getfrag);
1406EXPORT_SYMBOL(ip_queue_xmit);
1407EXPORT_SYMBOL(ip_send_check);