blob: 81b2795a4c2053ddf1855b60dbec4357e9414845 [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
Brian Haleyab32ea52006-09-22 14:15:41 -070086int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
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 */
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -0700331 security_sk_classify_flow(sk, &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 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)) {
Wei Dong0668b472006-08-31 15:24:48 -0700444 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
446 htonl(dst_mtu(&rt->u.dst)));
447 kfree_skb(skb);
448 return -EMSGSIZE;
449 }
450
451 /*
452 * Setup starting values.
453 */
454
455 hlen = iph->ihl * 4;
456 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800457 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 /* When frag_list is given, use it. First, check its validity:
460 * some transformers could create wrong frag_list or break existing
461 * one, it is not prohibited. In this case fall back to copying.
462 *
463 * LATER: this step can be merged to real generation of fragments,
464 * we can switch to copy when see the first bad fragment.
465 */
466 if (skb_shinfo(skb)->frag_list) {
467 struct sk_buff *frag;
468 int first_len = skb_pagelen(skb);
469
470 if (first_len - hlen > mtu ||
471 ((first_len - hlen) & 7) ||
472 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
473 skb_cloned(skb))
474 goto slow_path;
475
476 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
477 /* Correct geometry. */
478 if (frag->len > mtu ||
479 ((frag->len & 7) && frag->next) ||
480 skb_headroom(frag) < hlen)
481 goto slow_path;
482
483 /* Partially cloned skb? */
484 if (skb_shared(frag))
485 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700486
487 BUG_ON(frag->sk);
488 if (skb->sk) {
489 sock_hold(skb->sk);
490 frag->sk = skb->sk;
491 frag->destructor = sock_wfree;
492 skb->truesize -= frag->truesize;
493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 }
495
496 /* Everything is OK. Generate! */
497
498 err = 0;
499 offset = 0;
500 frag = skb_shinfo(skb)->frag_list;
501 skb_shinfo(skb)->frag_list = NULL;
502 skb->data_len = first_len - skb_headlen(skb);
503 skb->len = first_len;
504 iph->tot_len = htons(first_len);
505 iph->frag_off = htons(IP_MF);
506 ip_send_check(iph);
507
508 for (;;) {
509 /* Prepare header of the next frame,
510 * before previous one went down. */
511 if (frag) {
512 frag->ip_summed = CHECKSUM_NONE;
513 frag->h.raw = frag->data;
514 frag->nh.raw = __skb_push(frag, hlen);
515 memcpy(frag->nh.raw, iph, hlen);
516 iph = frag->nh.iph;
517 iph->tot_len = htons(frag->len);
518 ip_copy_metadata(frag, skb);
519 if (offset == 0)
520 ip_options_fragment(frag);
521 offset += skb->len - hlen;
522 iph->frag_off = htons(offset>>3);
523 if (frag->next != NULL)
524 iph->frag_off |= htons(IP_MF);
525 /* Ready, complete checksum */
526 ip_send_check(iph);
527 }
528
529 err = output(skb);
530
Wei Dongdafee492006-08-02 13:41:21 -0700531 if (!err)
532 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (err || !frag)
534 break;
535
536 skb = frag;
537 frag = skb->next;
538 skb->next = NULL;
539 }
540
541 if (err == 0) {
542 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
543 return 0;
544 }
545
546 while (frag) {
547 skb = frag->next;
548 kfree_skb(frag);
549 frag = skb;
550 }
551 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
552 return err;
553 }
554
555slow_path:
556 left = skb->len - hlen; /* Space per frame */
557 ptr = raw + hlen; /* Where to start from */
558
559#ifdef CONFIG_BRIDGE_NETFILTER
560 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
561 * we need to make room for the encapsulating header */
562 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
563 mtu -= nf_bridge_pad(skb);
564#else
565 ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
566#endif
567 /*
568 * Fragment the datagram.
569 */
570
571 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
572 not_last_frag = iph->frag_off & htons(IP_MF);
573
574 /*
575 * Keep copying data until we run out.
576 */
577
578 while(left > 0) {
579 len = left;
580 /* IF: it doesn't fit, use 'mtu' - the data space left */
581 if (len > mtu)
582 len = mtu;
583 /* IF: we are not sending upto and including the packet end
584 then align the next start on an eight byte boundary */
585 if (len < left) {
586 len &= ~7;
587 }
588 /*
589 * Allocate buffer.
590 */
591
592 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700593 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 err = -ENOMEM;
595 goto fail;
596 }
597
598 /*
599 * Set up data on packet
600 */
601
602 ip_copy_metadata(skb2, skb);
603 skb_reserve(skb2, ll_rs);
604 skb_put(skb2, len + hlen);
605 skb2->nh.raw = skb2->data;
606 skb2->h.raw = skb2->data + hlen;
607
608 /*
609 * Charge the memory for the fragment to any owner
610 * it might possess
611 */
612
613 if (skb->sk)
614 skb_set_owner_w(skb2, skb->sk);
615
616 /*
617 * Copy the packet header into the new buffer.
618 */
619
620 memcpy(skb2->nh.raw, skb->data, hlen);
621
622 /*
623 * Copy a block of the IP datagram.
624 */
625 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
626 BUG();
627 left -= len;
628
629 /*
630 * Fill in the new header fields.
631 */
632 iph = skb2->nh.iph;
633 iph->frag_off = htons((offset >> 3));
634
635 /* ANK: dirty, but effective trick. Upgrade options only if
636 * the segment to be fragmented was THE FIRST (otherwise,
637 * options are already fixed) and make it ONCE
638 * on the initial skb, so that all the following fragments
639 * will inherit fixed options.
640 */
641 if (offset == 0)
642 ip_options_fragment(skb);
643
644 /*
645 * Added AC : If we are fragmenting a fragment that's not the
646 * last fragment then keep MF on each bit
647 */
648 if (left > 0 || not_last_frag)
649 iph->frag_off |= htons(IP_MF);
650 ptr += len;
651 offset += len;
652
653 /*
654 * Put this fragment into the sending queue.
655 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 iph->tot_len = htons(len + hlen);
657
658 ip_send_check(iph);
659
660 err = output(skb2);
661 if (err)
662 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700663
664 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 }
666 kfree_skb(skb);
667 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
668 return err;
669
670fail:
671 kfree_skb(skb);
672 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
673 return err;
674}
675
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700676EXPORT_SYMBOL(ip_fragment);
677
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678int
679ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
680{
681 struct iovec *iov = from;
682
Patrick McHardy84fa7932006-08-29 16:44:56 -0700683 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
685 return -EFAULT;
686 } else {
687 unsigned int csum = 0;
688 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
689 return -EFAULT;
690 skb->csum = csum_block_add(skb->csum, csum, odd);
691 }
692 return 0;
693}
694
695static inline unsigned int
696csum_page(struct page *page, int offset, int copy)
697{
698 char *kaddr;
699 unsigned int csum;
700 kaddr = kmap(page);
701 csum = csum_partial(kaddr + offset, copy, 0);
702 kunmap(page);
703 return csum;
704}
705
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800706static inline int ip_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700707 int getfrag(void *from, char *to, int offset, int len,
708 int odd, struct sk_buff *skb),
709 void *from, int length, int hh_len, int fragheaderlen,
710 int transhdrlen, int mtu,unsigned int flags)
711{
712 struct sk_buff *skb;
713 int err;
714
715 /* There is support for UDP fragmentation offload by network
716 * device, so create one single skb packet containing complete
717 * udp datagram
718 */
719 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
720 skb = sock_alloc_send_skb(sk,
721 hh_len + fragheaderlen + transhdrlen + 20,
722 (flags & MSG_DONTWAIT), &err);
723
724 if (skb == NULL)
725 return err;
726
727 /* reserve space for Hardware header */
728 skb_reserve(skb, hh_len);
729
730 /* create space for UDP/IP header */
731 skb_put(skb,fragheaderlen + transhdrlen);
732
733 /* initialize network header pointer */
734 skb->nh.raw = skb->data;
735
736 /* initialize protocol header pointer */
737 skb->h.raw = skb->data + fragheaderlen;
738
Patrick McHardy84fa7932006-08-29 16:44:56 -0700739 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700740 skb->csum = 0;
741 sk->sk_sndmsg_off = 0;
742 }
743
744 err = skb_append_datato_frags(sk,skb, getfrag, from,
745 (length - transhdrlen));
746 if (!err) {
747 /* specify the length of each IP datagram fragment*/
Herbert Xu79671682006-06-22 02:40:14 -0700748 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700749 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700750 __skb_queue_tail(&sk->sk_write_queue, skb);
751
752 return 0;
753 }
754 /* There is not enough support do UFO ,
755 * so follow normal path
756 */
757 kfree_skb(skb);
758 return err;
759}
760
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761/*
762 * ip_append_data() and ip_append_page() can make one large IP datagram
763 * from many pieces of data. Each pieces will be holded on the socket
764 * until ip_push_pending_frames() is called. Each piece can be a page
765 * or non-page data.
766 *
767 * Not only UDP, other transport protocols - e.g. raw sockets - can use
768 * this interface potentially.
769 *
770 * LATER: length must be adjusted by pad at tail, when it is required.
771 */
772int ip_append_data(struct sock *sk,
773 int getfrag(void *from, char *to, int offset, int len,
774 int odd, struct sk_buff *skb),
775 void *from, int length, int transhdrlen,
776 struct ipcm_cookie *ipc, struct rtable *rt,
777 unsigned int flags)
778{
779 struct inet_sock *inet = inet_sk(sk);
780 struct sk_buff *skb;
781
782 struct ip_options *opt = NULL;
783 int hh_len;
784 int exthdrlen;
785 int mtu;
786 int copy;
787 int err;
788 int offset = 0;
789 unsigned int maxfraglen, fragheaderlen;
790 int csummode = CHECKSUM_NONE;
791
792 if (flags&MSG_PROBE)
793 return 0;
794
795 if (skb_queue_empty(&sk->sk_write_queue)) {
796 /*
797 * setup for corking.
798 */
799 opt = ipc->opt;
800 if (opt) {
801 if (inet->cork.opt == NULL) {
802 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
803 if (unlikely(inet->cork.opt == NULL))
804 return -ENOBUFS;
805 }
806 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
807 inet->cork.flags |= IPCORK_OPT;
808 inet->cork.addr = ipc->addr;
809 }
810 dst_hold(&rt->u.dst);
811 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
812 inet->cork.rt = rt;
813 inet->cork.length = 0;
814 sk->sk_sndmsg_page = NULL;
815 sk->sk_sndmsg_off = 0;
816 if ((exthdrlen = rt->u.dst.header_len) != 0) {
817 length += exthdrlen;
818 transhdrlen += exthdrlen;
819 }
820 } else {
821 rt = inet->cork.rt;
822 if (inet->cork.flags & IPCORK_OPT)
823 opt = inet->cork.opt;
824
825 transhdrlen = 0;
826 exthdrlen = 0;
827 mtu = inet->cork.fragsize;
828 }
829 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
830
831 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
832 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
833
834 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
835 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
836 return -EMSGSIZE;
837 }
838
839 /*
840 * transhdrlen > 0 means that this is the first fragment and we wish
841 * it won't be fragmented in the future.
842 */
843 if (transhdrlen &&
844 length + fragheaderlen <= mtu &&
Herbert Xu8648b302006-06-17 22:06:05 -0700845 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700847 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700850 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
851 (rt->u.dst.dev->features & NETIF_F_UFO)) {
852
Patrick McHardybaa829d2006-03-12 20:35:12 -0800853 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
854 fragheaderlen, transhdrlen, mtu,
855 flags);
856 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700857 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700858 return 0;
859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861 /* So, what's going on in the loop below?
862 *
863 * We use calculated fragment length to generate chained skb,
864 * each of segments is IP fragment ready for sending to network after
865 * adding appropriate IP header.
866 */
867
868 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
869 goto alloc_new_skb;
870
871 while (length > 0) {
872 /* Check if the remaining data fits into current packet. */
873 copy = mtu - skb->len;
874 if (copy < length)
875 copy = maxfraglen - skb->len;
876 if (copy <= 0) {
877 char *data;
878 unsigned int datalen;
879 unsigned int fraglen;
880 unsigned int fraggap;
881 unsigned int alloclen;
882 struct sk_buff *skb_prev;
883alloc_new_skb:
884 skb_prev = skb;
885 if (skb_prev)
886 fraggap = skb_prev->len - maxfraglen;
887 else
888 fraggap = 0;
889
890 /*
891 * If remaining data exceeds the mtu,
892 * we know we need more fragment(s).
893 */
894 datalen = length + fraggap;
895 if (datalen > mtu - fragheaderlen)
896 datalen = maxfraglen - fragheaderlen;
897 fraglen = datalen + fragheaderlen;
898
899 if ((flags & MSG_MORE) &&
900 !(rt->u.dst.dev->features&NETIF_F_SG))
901 alloclen = mtu;
902 else
903 alloclen = datalen + fragheaderlen;
904
905 /* The last fragment gets additional space at tail.
906 * Note, with MSG_MORE we overallocate on fragments,
907 * because we have no idea what fragment will be
908 * the last.
909 */
Zach Brown3d9dd752006-04-14 16:04:18 -0700910 if (datalen == length + fraggap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 alloclen += rt->u.dst.trailer_len;
912
913 if (transhdrlen) {
914 skb = sock_alloc_send_skb(sk,
915 alloclen + hh_len + 15,
916 (flags & MSG_DONTWAIT), &err);
917 } else {
918 skb = NULL;
919 if (atomic_read(&sk->sk_wmem_alloc) <=
920 2 * sk->sk_sndbuf)
921 skb = sock_wmalloc(sk,
922 alloclen + hh_len + 15, 1,
923 sk->sk_allocation);
924 if (unlikely(skb == NULL))
925 err = -ENOBUFS;
926 }
927 if (skb == NULL)
928 goto error;
929
930 /*
931 * Fill in the control structures
932 */
933 skb->ip_summed = csummode;
934 skb->csum = 0;
935 skb_reserve(skb, hh_len);
936
937 /*
938 * Find where to start putting bytes.
939 */
940 data = skb_put(skb, fraglen);
941 skb->nh.raw = data + exthdrlen;
942 data += fragheaderlen;
943 skb->h.raw = data + exthdrlen;
944
945 if (fraggap) {
946 skb->csum = skb_copy_and_csum_bits(
947 skb_prev, maxfraglen,
948 data + transhdrlen, fraggap, 0);
949 skb_prev->csum = csum_sub(skb_prev->csum,
950 skb->csum);
951 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700952 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
954
955 copy = datalen - transhdrlen - fraggap;
956 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
957 err = -EFAULT;
958 kfree_skb(skb);
959 goto error;
960 }
961
962 offset += copy;
963 length -= datalen - fraggap;
964 transhdrlen = 0;
965 exthdrlen = 0;
966 csummode = CHECKSUM_NONE;
967
968 /*
969 * Put the packet on the pending queue.
970 */
971 __skb_queue_tail(&sk->sk_write_queue, skb);
972 continue;
973 }
974
975 if (copy > length)
976 copy = length;
977
978 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
979 unsigned int off;
980
981 off = skb->len;
982 if (getfrag(from, skb_put(skb, copy),
983 offset, copy, off, skb) < 0) {
984 __skb_trim(skb, off);
985 err = -EFAULT;
986 goto error;
987 }
988 } else {
989 int i = skb_shinfo(skb)->nr_frags;
990 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
991 struct page *page = sk->sk_sndmsg_page;
992 int off = sk->sk_sndmsg_off;
993 unsigned int left;
994
995 if (page && (left = PAGE_SIZE - off) > 0) {
996 if (copy >= left)
997 copy = left;
998 if (page != frag->page) {
999 if (i == MAX_SKB_FRAGS) {
1000 err = -EMSGSIZE;
1001 goto error;
1002 }
1003 get_page(page);
1004 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1005 frag = &skb_shinfo(skb)->frags[i];
1006 }
1007 } else if (i < MAX_SKB_FRAGS) {
1008 if (copy > PAGE_SIZE)
1009 copy = PAGE_SIZE;
1010 page = alloc_pages(sk->sk_allocation, 0);
1011 if (page == NULL) {
1012 err = -ENOMEM;
1013 goto error;
1014 }
1015 sk->sk_sndmsg_page = page;
1016 sk->sk_sndmsg_off = 0;
1017
1018 skb_fill_page_desc(skb, i, page, 0, 0);
1019 frag = &skb_shinfo(skb)->frags[i];
1020 skb->truesize += PAGE_SIZE;
1021 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1022 } else {
1023 err = -EMSGSIZE;
1024 goto error;
1025 }
1026 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1027 err = -EFAULT;
1028 goto error;
1029 }
1030 sk->sk_sndmsg_off += copy;
1031 frag->size += copy;
1032 skb->len += copy;
1033 skb->data_len += copy;
1034 }
1035 offset += copy;
1036 length -= copy;
1037 }
1038
1039 return 0;
1040
1041error:
1042 inet->cork.length -= length;
1043 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1044 return err;
1045}
1046
1047ssize_t ip_append_page(struct sock *sk, struct page *page,
1048 int offset, size_t size, int flags)
1049{
1050 struct inet_sock *inet = inet_sk(sk);
1051 struct sk_buff *skb;
1052 struct rtable *rt;
1053 struct ip_options *opt = NULL;
1054 int hh_len;
1055 int mtu;
1056 int len;
1057 int err;
1058 unsigned int maxfraglen, fragheaderlen, fraggap;
1059
1060 if (inet->hdrincl)
1061 return -EPERM;
1062
1063 if (flags&MSG_PROBE)
1064 return 0;
1065
1066 if (skb_queue_empty(&sk->sk_write_queue))
1067 return -EINVAL;
1068
1069 rt = inet->cork.rt;
1070 if (inet->cork.flags & IPCORK_OPT)
1071 opt = inet->cork.opt;
1072
1073 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1074 return -EOPNOTSUPP;
1075
1076 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1077 mtu = inet->cork.fragsize;
1078
1079 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1080 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1081
1082 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1083 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1084 return -EMSGSIZE;
1085 }
1086
1087 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1088 return -EINVAL;
1089
1090 inet->cork.length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001091 if ((sk->sk_protocol == IPPROTO_UDP) &&
Herbert Xu79671682006-06-22 02:40:14 -07001092 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1093 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001094 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001095 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001096
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097
1098 while (size > 0) {
1099 int i;
1100
Herbert Xu89114af2006-07-08 13:34:32 -07001101 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001102 len = size;
1103 else {
1104
1105 /* Check if the remaining data fits into current packet. */
1106 len = mtu - skb->len;
1107 if (len < size)
1108 len = maxfraglen - skb->len;
1109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (len <= 0) {
1111 struct sk_buff *skb_prev;
1112 char *data;
1113 struct iphdr *iph;
1114 int alloclen;
1115
1116 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001117 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 alloclen = fragheaderlen + hh_len + fraggap + 15;
1120 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1121 if (unlikely(!skb)) {
1122 err = -ENOBUFS;
1123 goto error;
1124 }
1125
1126 /*
1127 * Fill in the control structures
1128 */
1129 skb->ip_summed = CHECKSUM_NONE;
1130 skb->csum = 0;
1131 skb_reserve(skb, hh_len);
1132
1133 /*
1134 * Find where to start putting bytes.
1135 */
1136 data = skb_put(skb, fragheaderlen + fraggap);
1137 skb->nh.iph = iph = (struct iphdr *)data;
1138 data += fragheaderlen;
1139 skb->h.raw = data;
1140
1141 if (fraggap) {
1142 skb->csum = skb_copy_and_csum_bits(
1143 skb_prev, maxfraglen,
1144 data, fraggap, 0);
1145 skb_prev->csum = csum_sub(skb_prev->csum,
1146 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001147 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
1149
1150 /*
1151 * Put the packet on the pending queue.
1152 */
1153 __skb_queue_tail(&sk->sk_write_queue, skb);
1154 continue;
1155 }
1156
1157 i = skb_shinfo(skb)->nr_frags;
1158 if (len > size)
1159 len = size;
1160 if (skb_can_coalesce(skb, i, page, offset)) {
1161 skb_shinfo(skb)->frags[i-1].size += len;
1162 } else if (i < MAX_SKB_FRAGS) {
1163 get_page(page);
1164 skb_fill_page_desc(skb, i, page, offset, len);
1165 } else {
1166 err = -EMSGSIZE;
1167 goto error;
1168 }
1169
1170 if (skb->ip_summed == CHECKSUM_NONE) {
1171 unsigned int csum;
1172 csum = csum_page(page, offset, len);
1173 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1174 }
1175
1176 skb->len += len;
1177 skb->data_len += len;
1178 offset += len;
1179 size -= len;
1180 }
1181 return 0;
1182
1183error:
1184 inet->cork.length -= size;
1185 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1186 return err;
1187}
1188
1189/*
1190 * Combined all pending IP fragments on the socket as one IP datagram
1191 * and push them out.
1192 */
1193int ip_push_pending_frames(struct sock *sk)
1194{
1195 struct sk_buff *skb, *tmp_skb;
1196 struct sk_buff **tail_skb;
1197 struct inet_sock *inet = inet_sk(sk);
1198 struct ip_options *opt = NULL;
1199 struct rtable *rt = inet->cork.rt;
1200 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001201 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 __u8 ttl;
1203 int err = 0;
1204
1205 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1206 goto out;
1207 tail_skb = &(skb_shinfo(skb)->frag_list);
1208
1209 /* move skb->data to ip header from ext header */
1210 if (skb->data < skb->nh.raw)
1211 __skb_pull(skb, skb->nh.raw - skb->data);
1212 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1213 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1214 *tail_skb = tmp_skb;
1215 tail_skb = &(tmp_skb->next);
1216 skb->len += tmp_skb->len;
1217 skb->data_len += tmp_skb->len;
1218 skb->truesize += tmp_skb->truesize;
1219 __sock_put(tmp_skb->sk);
1220 tmp_skb->destructor = NULL;
1221 tmp_skb->sk = NULL;
1222 }
1223
1224 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1225 * to fragment the frame generated here. No matter, what transforms
1226 * how transforms change size of the packet, it will come out.
1227 */
1228 if (inet->pmtudisc != IP_PMTUDISC_DO)
1229 skb->local_df = 1;
1230
1231 /* DF bit is set when we want to see DF on outgoing frames.
1232 * If local_df is set too, we still allow to fragment this frame
1233 * locally. */
1234 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1235 (skb->len <= dst_mtu(&rt->u.dst) &&
1236 ip_dont_fragment(sk, &rt->u.dst)))
1237 df = htons(IP_DF);
1238
1239 if (inet->cork.flags & IPCORK_OPT)
1240 opt = inet->cork.opt;
1241
1242 if (rt->rt_type == RTN_MULTICAST)
1243 ttl = inet->mc_ttl;
1244 else
1245 ttl = ip_select_ttl(inet, &rt->u.dst);
1246
1247 iph = (struct iphdr *)skb->data;
1248 iph->version = 4;
1249 iph->ihl = 5;
1250 if (opt) {
1251 iph->ihl += opt->optlen>>2;
1252 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1253 }
1254 iph->tos = inet->tos;
1255 iph->tot_len = htons(skb->len);
1256 iph->frag_off = df;
Alexey Kuznetsov1a55d572006-03-22 14:27:59 -08001257 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 iph->ttl = ttl;
1259 iph->protocol = sk->sk_protocol;
1260 iph->saddr = rt->rt_src;
1261 iph->daddr = rt->rt_dst;
1262 ip_send_check(iph);
1263
1264 skb->priority = sk->sk_priority;
1265 skb->dst = dst_clone(&rt->u.dst);
1266
1267 /* Netfilter gets whole the not fragmented skb. */
1268 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
1269 skb->dst->dev, dst_output);
1270 if (err) {
1271 if (err > 0)
1272 err = inet->recverr ? net_xmit_errno(err) : 0;
1273 if (err)
1274 goto error;
1275 }
1276
1277out:
1278 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001279 kfree(inet->cork.opt);
1280 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (inet->cork.rt) {
1282 ip_rt_put(inet->cork.rt);
1283 inet->cork.rt = NULL;
1284 }
1285 return err;
1286
1287error:
1288 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1289 goto out;
1290}
1291
1292/*
1293 * Throw away all pending data on the socket.
1294 */
1295void ip_flush_pending_frames(struct sock *sk)
1296{
1297 struct inet_sock *inet = inet_sk(sk);
1298 struct sk_buff *skb;
1299
1300 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1301 kfree_skb(skb);
1302
1303 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001304 kfree(inet->cork.opt);
1305 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if (inet->cork.rt) {
1307 ip_rt_put(inet->cork.rt);
1308 inet->cork.rt = NULL;
1309 }
1310}
1311
1312
1313/*
1314 * Fetch data from kernel space and fill in checksum if needed.
1315 */
1316static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1317 int len, int odd, struct sk_buff *skb)
1318{
1319 unsigned int csum;
1320
1321 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1322 skb->csum = csum_block_add(skb->csum, csum, odd);
1323 return 0;
1324}
1325
1326/*
1327 * Generic function to send a packet as reply to another packet.
1328 * Used to send TCP resets so far. ICMP should use this function too.
1329 *
1330 * Should run single threaded per socket because it uses the sock
1331 * structure to pass arguments.
1332 *
1333 * LATER: switch from ip_build_xmit to ip_append_*
1334 */
1335void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1336 unsigned int len)
1337{
1338 struct inet_sock *inet = inet_sk(sk);
1339 struct {
1340 struct ip_options opt;
1341 char data[40];
1342 } replyopts;
1343 struct ipcm_cookie ipc;
1344 u32 daddr;
1345 struct rtable *rt = (struct rtable*)skb->dst;
1346
1347 if (ip_options_echo(&replyopts.opt, skb))
1348 return;
1349
1350 daddr = ipc.addr = rt->rt_src;
1351 ipc.opt = NULL;
1352
1353 if (replyopts.opt.optlen) {
1354 ipc.opt = &replyopts.opt;
1355
1356 if (ipc.opt->srr)
1357 daddr = replyopts.opt.faddr;
1358 }
1359
1360 {
1361 struct flowi fl = { .nl_u = { .ip4_u =
1362 { .daddr = daddr,
1363 .saddr = rt->rt_spec_dst,
1364 .tos = RT_TOS(skb->nh.iph->tos) } },
1365 /* Not quite clean, but right. */
1366 .uli_u = { .ports =
1367 { .sport = skb->h.th->dest,
1368 .dport = skb->h.th->source } },
1369 .proto = sk->sk_protocol };
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001370 security_skb_classify_flow(skb, &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 if (ip_route_output_key(&rt, &fl))
1372 return;
1373 }
1374
1375 /* And let IP do all the hard work.
1376
1377 This chunk is not reenterable, hence spinlock.
1378 Note that it uses the fact, that this function is called
1379 with locally disabled BH and that sk cannot be already spinlocked.
1380 */
1381 bh_lock_sock(sk);
1382 inet->tos = skb->nh.iph->tos;
1383 sk->sk_priority = skb->priority;
1384 sk->sk_protocol = skb->nh.iph->protocol;
1385 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1386 &ipc, rt, MSG_DONTWAIT);
1387 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1388 if (arg->csumoffset >= 0)
1389 *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
1390 skb->ip_summed = CHECKSUM_NONE;
1391 ip_push_pending_frames(sk);
1392 }
1393
1394 bh_unlock_sock(sk);
1395
1396 ip_rt_put(rt);
1397}
1398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399void __init ip_init(void)
1400{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 ip_rt_init();
1402 inet_initpeers();
1403
1404#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1405 igmp_mc_proc_init();
1406#endif
1407}
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409EXPORT_SYMBOL(ip_generic_getfrag);
1410EXPORT_SYMBOL(ip_queue_xmit);
1411EXPORT_SYMBOL(ip_send_check);