blob: d329718a8e8cb2985d9e98d8c2ed6d1c512fafcd [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.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090025 * Bradford Johnson: Fix faulty handling of some frames when
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * 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.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090036 * Andi Kleen : Split fast and slow ip_build_xmit path
37 * for decreased register pressure on x86
38 * and more readibility.
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/mm.h>
53#include <linux/string.h>
54#include <linux/errno.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040055#include <linux/highmem.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{
Arnaldo Carvalho de Melo459a98e2007-03-19 15:30:44 -070098 skb_reset_mac_header(newskb);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 __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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900116/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 * 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,
Al Viro13d8eaa2006-09-26 22:27:30 -0700121 __be32 saddr, __be32 daddr, struct ip_options *opt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 struct net_device *dev = dst->dev;
167 int hh_len = LL_RESERVED_SPACE(dev);
168
169 /* Be paranoid, rather than too clever. */
170 if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
171 struct sk_buff *skb2;
172
173 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
174 if (skb2 == NULL) {
175 kfree_skb(skb);
176 return -ENOMEM;
177 }
178 if (skb->sk)
179 skb_set_owner_w(skb2, skb->sk);
180 kfree_skb(skb);
181 skb = skb2;
182 }
183
Stephen Hemminger3644f0c2006-12-07 15:08:17 -0800184 if (dst->hh)
185 return neigh_hh_output(dst->hh, skb);
186 else if (dst->neighbour)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 return dst->neighbour->output(skb);
188
189 if (net_ratelimit())
190 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
191 kfree_skb(skb);
192 return -EINVAL;
193}
194
Thomas Graf33d043d2005-08-20 17:27:34 -0700195static inline int ip_finish_output(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Patrick McHardy5c901da2006-01-06 23:05:36 -0800197#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
198 /* Policy lookup after SNAT yielded a new policy */
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800199 if (skb->dst->xfrm != NULL) {
200 IPCB(skb)->flags |= IPSKB_REROUTED;
201 return dst_output(skb);
202 }
Patrick McHardy5c901da2006-01-06 23:05:36 -0800203#endif
Herbert Xu89114af2006-07-08 13:34:32 -0700204 if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb))
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800205 return ip_fragment(skb, ip_finish_output2);
206 else
207 return ip_finish_output2(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
209
210int ip_mc_output(struct sk_buff *skb)
211{
212 struct sock *sk = skb->sk;
213 struct rtable *rt = (struct rtable*)skb->dst;
214 struct net_device *dev = rt->u.dst.dev;
215
216 /*
217 * If the indicated interface is up and running, send the packet.
218 */
219 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
220
221 skb->dev = dev;
222 skb->protocol = htons(ETH_P_IP);
223
224 /*
225 * Multicasts are looped back for other local users
226 */
227
228 if (rt->rt_flags&RTCF_MULTICAST) {
229 if ((!sk || inet_sk(sk)->mc_loop)
230#ifdef CONFIG_IP_MROUTE
231 /* Small optimization: do not loopback not local frames,
232 which returned after forwarding; they will be dropped
233 by ip_mr_input in any case.
234 Note, that local frames are looped back to be delivered
235 to local recipients.
236
237 This check is duplicated in ip_mr_input at the moment.
238 */
239 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
240#endif
241 ) {
242 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
243 if (newskb)
244 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900245 newskb->dev,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 ip_dev_loopback_xmit);
247 }
248
249 /* Multicasts with ttl 0 must not go beyond the host */
250
251 if (skb->nh.iph->ttl == 0) {
252 kfree_skb(skb);
253 return 0;
254 }
255 }
256
257 if (rt->rt_flags&RTCF_BROADCAST) {
258 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
259 if (newskb)
260 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
261 newskb->dev, ip_dev_loopback_xmit);
262 }
263
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800264 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
265 ip_finish_output,
266 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
269int ip_output(struct sk_buff *skb)
270{
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800271 struct net_device *dev = skb->dst->dev;
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
274
Patrick McHardy1bd9bef2006-01-05 12:20:59 -0800275 skb->dev = dev;
276 skb->protocol = htons(ETH_P_IP);
277
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800278 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900279 ip_finish_output,
Patrick McHardy48d5cad2006-02-15 15:10:22 -0800280 !(IPCB(skb)->flags & IPSKB_REROUTED));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281}
282
David S. Millere89862f2007-01-26 01:04:55 -0800283int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284{
David S. Millere89862f2007-01-26 01:04:55 -0800285 struct sock *sk = skb->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 struct inet_sock *inet = inet_sk(sk);
287 struct ip_options *opt = inet->opt;
288 struct rtable *rt;
289 struct iphdr *iph;
290
291 /* Skip all of this if the packet is already routed,
292 * f.e. by something like SCTP.
293 */
294 rt = (struct rtable *) skb->dst;
295 if (rt != NULL)
296 goto packet_routed;
297
298 /* Make sure we can route this packet. */
299 rt = (struct rtable *)__sk_dst_check(sk, 0);
300 if (rt == NULL) {
Al Viro3ca3c682006-09-27 18:28:07 -0700301 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303 /* Use correct destination address if we have options. */
304 daddr = inet->daddr;
305 if(opt && opt->srr)
306 daddr = opt->faddr;
307
308 {
309 struct flowi fl = { .oif = sk->sk_bound_dev_if,
310 .nl_u = { .ip4_u =
311 { .daddr = daddr,
312 .saddr = inet->saddr,
313 .tos = RT_CONN_FLAGS(sk) } },
314 .proto = sk->sk_protocol,
315 .uli_u = { .ports =
316 { .sport = inet->sport,
317 .dport = inet->dport } } };
318
319 /* If this fails, retransmit mechanism of transport layer will
320 * keep trying until route appears or the connection times
321 * itself out.
322 */
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -0700323 security_sk_classify_flow(sk, &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (ip_route_output_flow(&rt, &fl, sk, 0))
325 goto no_route;
326 }
Arnaldo Carvalho de Melo6cbb0df2005-08-09 19:49:02 -0700327 sk_setup_caps(sk, &rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 }
329 skb->dst = dst_clone(&rt->u.dst);
330
331packet_routed:
332 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
333 goto no_route;
334
335 /* OK, we know where to send it, allocate and build IP header. */
336 iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
Al Viro714e85b2006-11-14 20:51:49 -0800337 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 iph->tot_len = htons(skb->len);
339 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
340 iph->frag_off = htons(IP_DF);
341 else
342 iph->frag_off = 0;
343 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
344 iph->protocol = sk->sk_protocol;
345 iph->saddr = rt->rt_src;
346 iph->daddr = rt->rt_dst;
347 skb->nh.iph = iph;
348 /* Transport layer set skb->h.foo itself. */
349
350 if (opt && opt->optlen) {
351 iph->ihl += opt->optlen >> 2;
352 ip_options_build(skb, opt, inet->daddr, rt, 0);
353 }
354
Herbert Xu89f5f0a2005-11-08 09:41:56 -0800355 ip_select_ident_more(iph, &rt->u.dst, sk,
Herbert Xu79671682006-06-22 02:40:14 -0700356 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
358 /* Add an IP checksum. */
359 ip_send_check(iph);
360
361 skb->priority = sk->sk_priority;
362
363 return NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
364 dst_output);
365
366no_route:
367 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
368 kfree_skb(skb);
369 return -EHOSTUNREACH;
370}
371
372
373static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
374{
375 to->pkt_type = from->pkt_type;
376 to->priority = from->priority;
377 to->protocol = from->protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 dst_release(to->dst);
379 to->dst = dst_clone(from->dst);
380 to->dev = from->dev;
Thomas Graf82e91ff2006-11-09 15:19:14 -0800381 to->mark = from->mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 /* Copy the flags to each fragment. */
384 IPCB(to)->flags = IPCB(from)->flags;
385
386#ifdef CONFIG_NET_SCHED
387 to->tc_index = from->tc_index;
388#endif
389#ifdef CONFIG_NETFILTER
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 /* Connection association is same as pre-frag packet */
391 nf_conntrack_put(to->nfct);
392 to->nfct = from->nfct;
393 nf_conntrack_get(to->nfct);
394 to->nfctinfo = from->nfctinfo;
Julian Anastasovc98d80e2005-10-22 13:39:21 +0300395#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
396 to->ipvs_property = from->ipvs_property;
397#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398#ifdef CONFIG_BRIDGE_NETFILTER
399 nf_bridge_put(to->nf_bridge);
400 to->nf_bridge = from->nf_bridge;
401 nf_bridge_get(to->nf_bridge);
402#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403#endif
James Morris984bc162006-06-09 00:29:17 -0700404 skb_copy_secmark(to, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}
406
407/*
408 * This IP datagram is too large to be sent in one piece. Break it up into
409 * smaller pieces (each of size equal to IP header plus
410 * a block of the data of the original IP data part) that will yet fit in a
411 * single device frame, and queue such a frame for sending.
412 */
413
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700414int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
416 struct iphdr *iph;
417 int raw = 0;
418 int ptr;
419 struct net_device *dev;
420 struct sk_buff *skb2;
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700421 unsigned int mtu, hlen, left, len, ll_rs, pad;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 int offset;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -0800423 __be16 not_last_frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 struct rtable *rt = (struct rtable*)skb->dst;
425 int err = 0;
426
427 dev = rt->u.dst.dev;
428
429 /*
430 * Point into the IP datagram header.
431 */
432
433 iph = skb->nh.iph;
434
435 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
Wei Dong0668b472006-08-31 15:24:48 -0700436 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
438 htonl(dst_mtu(&rt->u.dst)));
439 kfree_skb(skb);
440 return -EMSGSIZE;
441 }
442
443 /*
444 * Setup starting values.
445 */
446
447 hlen = iph->ihl * 4;
448 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
Herbert Xu89cee8b2005-12-13 23:14:27 -0800449 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 /* When frag_list is given, use it. First, check its validity:
452 * some transformers could create wrong frag_list or break existing
453 * one, it is not prohibited. In this case fall back to copying.
454 *
455 * LATER: this step can be merged to real generation of fragments,
456 * we can switch to copy when see the first bad fragment.
457 */
458 if (skb_shinfo(skb)->frag_list) {
459 struct sk_buff *frag;
460 int first_len = skb_pagelen(skb);
461
462 if (first_len - hlen > mtu ||
463 ((first_len - hlen) & 7) ||
464 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
465 skb_cloned(skb))
466 goto slow_path;
467
468 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
469 /* Correct geometry. */
470 if (frag->len > mtu ||
471 ((frag->len & 7) && frag->next) ||
472 skb_headroom(frag) < hlen)
473 goto slow_path;
474
475 /* Partially cloned skb? */
476 if (skb_shared(frag))
477 goto slow_path;
Herbert Xu2fdba6b2005-05-18 22:52:33 -0700478
479 BUG_ON(frag->sk);
480 if (skb->sk) {
481 sock_hold(skb->sk);
482 frag->sk = skb->sk;
483 frag->destructor = sock_wfree;
484 skb->truesize -= frag->truesize;
485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 }
487
488 /* Everything is OK. Generate! */
489
490 err = 0;
491 offset = 0;
492 frag = skb_shinfo(skb)->frag_list;
493 skb_shinfo(skb)->frag_list = NULL;
494 skb->data_len = first_len - skb_headlen(skb);
495 skb->len = first_len;
496 iph->tot_len = htons(first_len);
497 iph->frag_off = htons(IP_MF);
498 ip_send_check(iph);
499
500 for (;;) {
501 /* Prepare header of the next frame,
502 * before previous one went down. */
503 if (frag) {
504 frag->ip_summed = CHECKSUM_NONE;
505 frag->h.raw = frag->data;
Arnaldo Carvalho de Meloe2d1bca2007-04-10 20:46:21 -0700506 __skb_push(frag, hlen);
507 skb_reset_network_header(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 memcpy(frag->nh.raw, iph, hlen);
509 iph = frag->nh.iph;
510 iph->tot_len = htons(frag->len);
511 ip_copy_metadata(frag, skb);
512 if (offset == 0)
513 ip_options_fragment(frag);
514 offset += skb->len - hlen;
515 iph->frag_off = htons(offset>>3);
516 if (frag->next != NULL)
517 iph->frag_off |= htons(IP_MF);
518 /* Ready, complete checksum */
519 ip_send_check(iph);
520 }
521
522 err = output(skb);
523
Wei Dongdafee492006-08-02 13:41:21 -0700524 if (!err)
525 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (err || !frag)
527 break;
528
529 skb = frag;
530 frag = skb->next;
531 skb->next = NULL;
532 }
533
534 if (err == 0) {
535 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
536 return 0;
537 }
538
539 while (frag) {
540 skb = frag->next;
541 kfree_skb(frag);
542 frag = skb;
543 }
544 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
545 return err;
546 }
547
548slow_path:
549 left = skb->len - hlen; /* Space per frame */
550 ptr = raw + hlen; /* Where to start from */
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
Stephen Hemminger9bcfcaf2006-08-29 17:48:57 -0700553 * we need to make room for the encapsulating header
554 */
555 pad = nf_bridge_pad(skb);
556 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
557 mtu -= pad;
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 /*
560 * Fragment the datagram.
561 */
562
563 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
564 not_last_frag = iph->frag_off & htons(IP_MF);
565
566 /*
567 * Keep copying data until we run out.
568 */
569
Stephen Hemminger132adf52007-03-08 20:44:43 -0800570 while (left > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 len = left;
572 /* IF: it doesn't fit, use 'mtu' - the data space left */
573 if (len > mtu)
574 len = mtu;
575 /* IF: we are not sending upto and including the packet end
576 then align the next start on an eight byte boundary */
577 if (len < left) {
578 len &= ~7;
579 }
580 /*
581 * Allocate buffer.
582 */
583
584 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
Patrick McHardy64ce2072005-08-09 20:50:53 -0700585 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 err = -ENOMEM;
587 goto fail;
588 }
589
590 /*
591 * Set up data on packet
592 */
593
594 ip_copy_metadata(skb2, skb);
595 skb_reserve(skb2, ll_rs);
596 skb_put(skb2, len + hlen);
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700597 skb_reset_network_header(skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 skb2->h.raw = skb2->data + hlen;
599
600 /*
601 * Charge the memory for the fragment to any owner
602 * it might possess
603 */
604
605 if (skb->sk)
606 skb_set_owner_w(skb2, skb->sk);
607
608 /*
609 * Copy the packet header into the new buffer.
610 */
611
612 memcpy(skb2->nh.raw, skb->data, hlen);
613
614 /*
615 * Copy a block of the IP datagram.
616 */
617 if (skb_copy_bits(skb, ptr, skb2->h.raw, len))
618 BUG();
619 left -= len;
620
621 /*
622 * Fill in the new header fields.
623 */
624 iph = skb2->nh.iph;
625 iph->frag_off = htons((offset >> 3));
626
627 /* ANK: dirty, but effective trick. Upgrade options only if
628 * the segment to be fragmented was THE FIRST (otherwise,
629 * options are already fixed) and make it ONCE
630 * on the initial skb, so that all the following fragments
631 * will inherit fixed options.
632 */
633 if (offset == 0)
634 ip_options_fragment(skb);
635
636 /*
637 * Added AC : If we are fragmenting a fragment that's not the
638 * last fragment then keep MF on each bit
639 */
640 if (left > 0 || not_last_frag)
641 iph->frag_off |= htons(IP_MF);
642 ptr += len;
643 offset += len;
644
645 /*
646 * Put this fragment into the sending queue.
647 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 iph->tot_len = htons(len + hlen);
649
650 ip_send_check(iph);
651
652 err = output(skb2);
653 if (err)
654 goto fail;
Wei Dongdafee492006-08-02 13:41:21 -0700655
656 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 }
658 kfree_skb(skb);
659 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
660 return err;
661
662fail:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900663 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
665 return err;
666}
667
Patrick McHardy2e2f7ae2006-04-04 13:42:35 -0700668EXPORT_SYMBOL(ip_fragment);
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670int
671ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
672{
673 struct iovec *iov = from;
674
Patrick McHardy84fa7932006-08-29 16:44:56 -0700675 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
677 return -EFAULT;
678 } else {
Al Viro44bb9362006-11-14 21:36:14 -0800679 __wsum csum = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
681 return -EFAULT;
682 skb->csum = csum_block_add(skb->csum, csum, odd);
683 }
684 return 0;
685}
686
Al Viro44bb9362006-11-14 21:36:14 -0800687static inline __wsum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688csum_page(struct page *page, int offset, int copy)
689{
690 char *kaddr;
Al Viro44bb9362006-11-14 21:36:14 -0800691 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 kaddr = kmap(page);
693 csum = csum_partial(kaddr + offset, copy, 0);
694 kunmap(page);
695 return csum;
696}
697
Adrian Bunk4b30b1c2005-11-29 16:27:20 -0800698static inline int ip_ufo_append_data(struct sock *sk,
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700699 int getfrag(void *from, char *to, int offset, int len,
700 int odd, struct sk_buff *skb),
701 void *from, int length, int hh_len, int fragheaderlen,
702 int transhdrlen, int mtu,unsigned int flags)
703{
704 struct sk_buff *skb;
705 int err;
706
707 /* There is support for UDP fragmentation offload by network
708 * device, so create one single skb packet containing complete
709 * udp datagram
710 */
711 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
712 skb = sock_alloc_send_skb(sk,
713 hh_len + fragheaderlen + transhdrlen + 20,
714 (flags & MSG_DONTWAIT), &err);
715
716 if (skb == NULL)
717 return err;
718
719 /* reserve space for Hardware header */
720 skb_reserve(skb, hh_len);
721
722 /* create space for UDP/IP header */
723 skb_put(skb,fragheaderlen + transhdrlen);
724
725 /* initialize network header pointer */
Arnaldo Carvalho de Meloc1d2bbe2007-04-10 20:45:18 -0700726 skb_reset_network_header(skb);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700727
728 /* initialize protocol header pointer */
729 skb->h.raw = skb->data + fragheaderlen;
730
Patrick McHardy84fa7932006-08-29 16:44:56 -0700731 skb->ip_summed = CHECKSUM_PARTIAL;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700732 skb->csum = 0;
733 sk->sk_sndmsg_off = 0;
734 }
735
736 err = skb_append_datato_frags(sk,skb, getfrag, from,
737 (length - transhdrlen));
738 if (!err) {
739 /* specify the length of each IP datagram fragment*/
Herbert Xu79671682006-06-22 02:40:14 -0700740 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700741 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700742 __skb_queue_tail(&sk->sk_write_queue, skb);
743
744 return 0;
745 }
746 /* There is not enough support do UFO ,
747 * so follow normal path
748 */
749 kfree_skb(skb);
750 return err;
751}
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753/*
754 * ip_append_data() and ip_append_page() can make one large IP datagram
755 * from many pieces of data. Each pieces will be holded on the socket
756 * until ip_push_pending_frames() is called. Each piece can be a page
757 * or non-page data.
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900758 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 * Not only UDP, other transport protocols - e.g. raw sockets - can use
760 * this interface potentially.
761 *
762 * LATER: length must be adjusted by pad at tail, when it is required.
763 */
764int ip_append_data(struct sock *sk,
765 int getfrag(void *from, char *to, int offset, int len,
766 int odd, struct sk_buff *skb),
767 void *from, int length, int transhdrlen,
768 struct ipcm_cookie *ipc, struct rtable *rt,
769 unsigned int flags)
770{
771 struct inet_sock *inet = inet_sk(sk);
772 struct sk_buff *skb;
773
774 struct ip_options *opt = NULL;
775 int hh_len;
776 int exthdrlen;
777 int mtu;
778 int copy;
779 int err;
780 int offset = 0;
781 unsigned int maxfraglen, fragheaderlen;
782 int csummode = CHECKSUM_NONE;
783
784 if (flags&MSG_PROBE)
785 return 0;
786
787 if (skb_queue_empty(&sk->sk_write_queue)) {
788 /*
789 * setup for corking.
790 */
791 opt = ipc->opt;
792 if (opt) {
793 if (inet->cork.opt == NULL) {
794 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
795 if (unlikely(inet->cork.opt == NULL))
796 return -ENOBUFS;
797 }
798 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
799 inet->cork.flags |= IPCORK_OPT;
800 inet->cork.addr = ipc->addr;
801 }
802 dst_hold(&rt->u.dst);
803 inet->cork.fragsize = mtu = dst_mtu(rt->u.dst.path);
804 inet->cork.rt = rt;
805 inet->cork.length = 0;
806 sk->sk_sndmsg_page = NULL;
807 sk->sk_sndmsg_off = 0;
808 if ((exthdrlen = rt->u.dst.header_len) != 0) {
809 length += exthdrlen;
810 transhdrlen += exthdrlen;
811 }
812 } else {
813 rt = inet->cork.rt;
814 if (inet->cork.flags & IPCORK_OPT)
815 opt = inet->cork.opt;
816
817 transhdrlen = 0;
818 exthdrlen = 0;
819 mtu = inet->cork.fragsize;
820 }
821 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
822
823 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
824 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
825
826 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
827 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
828 return -EMSGSIZE;
829 }
830
831 /*
832 * transhdrlen > 0 means that this is the first fragment and we wish
833 * it won't be fragmented in the future.
834 */
835 if (transhdrlen &&
836 length + fragheaderlen <= mtu &&
Herbert Xu8648b302006-06-17 22:06:05 -0700837 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 !exthdrlen)
Patrick McHardy84fa7932006-08-29 16:44:56 -0700839 csummode = CHECKSUM_PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 inet->cork.length += length;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700842 if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
843 (rt->u.dst.dev->features & NETIF_F_UFO)) {
844
Patrick McHardybaa829d2006-03-12 20:35:12 -0800845 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
846 fragheaderlen, transhdrlen, mtu,
847 flags);
848 if (err)
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700849 goto error;
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700850 return 0;
851 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
853 /* So, what's going on in the loop below?
854 *
855 * We use calculated fragment length to generate chained skb,
856 * each of segments is IP fragment ready for sending to network after
857 * adding appropriate IP header.
858 */
859
860 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
861 goto alloc_new_skb;
862
863 while (length > 0) {
864 /* Check if the remaining data fits into current packet. */
865 copy = mtu - skb->len;
866 if (copy < length)
867 copy = maxfraglen - skb->len;
868 if (copy <= 0) {
869 char *data;
870 unsigned int datalen;
871 unsigned int fraglen;
872 unsigned int fraggap;
873 unsigned int alloclen;
874 struct sk_buff *skb_prev;
875alloc_new_skb:
876 skb_prev = skb;
877 if (skb_prev)
878 fraggap = skb_prev->len - maxfraglen;
879 else
880 fraggap = 0;
881
882 /*
883 * If remaining data exceeds the mtu,
884 * we know we need more fragment(s).
885 */
886 datalen = length + fraggap;
887 if (datalen > mtu - fragheaderlen)
888 datalen = maxfraglen - fragheaderlen;
889 fraglen = datalen + fragheaderlen;
890
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900891 if ((flags & MSG_MORE) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 !(rt->u.dst.dev->features&NETIF_F_SG))
893 alloclen = mtu;
894 else
895 alloclen = datalen + fragheaderlen;
896
897 /* The last fragment gets additional space at tail.
898 * Note, with MSG_MORE we overallocate on fragments,
899 * because we have no idea what fragment will be
900 * the last.
901 */
Zach Brown3d9dd752006-04-14 16:04:18 -0700902 if (datalen == length + fraggap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 alloclen += rt->u.dst.trailer_len;
904
905 if (transhdrlen) {
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900906 skb = sock_alloc_send_skb(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 alloclen + hh_len + 15,
908 (flags & MSG_DONTWAIT), &err);
909 } else {
910 skb = NULL;
911 if (atomic_read(&sk->sk_wmem_alloc) <=
912 2 * sk->sk_sndbuf)
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900913 skb = sock_wmalloc(sk,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 alloclen + hh_len + 15, 1,
915 sk->sk_allocation);
916 if (unlikely(skb == NULL))
917 err = -ENOBUFS;
918 }
919 if (skb == NULL)
920 goto error;
921
922 /*
923 * Fill in the control structures
924 */
925 skb->ip_summed = csummode;
926 skb->csum = 0;
927 skb_reserve(skb, hh_len);
928
929 /*
930 * Find where to start putting bytes.
931 */
932 data = skb_put(skb, fraglen);
933 skb->nh.raw = data + exthdrlen;
934 data += fragheaderlen;
935 skb->h.raw = data + exthdrlen;
936
937 if (fraggap) {
938 skb->csum = skb_copy_and_csum_bits(
939 skb_prev, maxfraglen,
940 data + transhdrlen, fraggap, 0);
941 skb_prev->csum = csum_sub(skb_prev->csum,
942 skb->csum);
943 data += fraggap;
Herbert Xue9fa4f72006-08-13 20:12:58 -0700944 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 }
946
947 copy = datalen - transhdrlen - fraggap;
948 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
949 err = -EFAULT;
950 kfree_skb(skb);
951 goto error;
952 }
953
954 offset += copy;
955 length -= datalen - fraggap;
956 transhdrlen = 0;
957 exthdrlen = 0;
958 csummode = CHECKSUM_NONE;
959
960 /*
961 * Put the packet on the pending queue.
962 */
963 __skb_queue_tail(&sk->sk_write_queue, skb);
964 continue;
965 }
966
967 if (copy > length)
968 copy = length;
969
970 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
971 unsigned int off;
972
973 off = skb->len;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900974 if (getfrag(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 offset, copy, off, skb) < 0) {
976 __skb_trim(skb, off);
977 err = -EFAULT;
978 goto error;
979 }
980 } else {
981 int i = skb_shinfo(skb)->nr_frags;
982 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
983 struct page *page = sk->sk_sndmsg_page;
984 int off = sk->sk_sndmsg_off;
985 unsigned int left;
986
987 if (page && (left = PAGE_SIZE - off) > 0) {
988 if (copy >= left)
989 copy = left;
990 if (page != frag->page) {
991 if (i == MAX_SKB_FRAGS) {
992 err = -EMSGSIZE;
993 goto error;
994 }
995 get_page(page);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900996 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 frag = &skb_shinfo(skb)->frags[i];
998 }
999 } else if (i < MAX_SKB_FRAGS) {
1000 if (copy > PAGE_SIZE)
1001 copy = PAGE_SIZE;
1002 page = alloc_pages(sk->sk_allocation, 0);
1003 if (page == NULL) {
1004 err = -ENOMEM;
1005 goto error;
1006 }
1007 sk->sk_sndmsg_page = page;
1008 sk->sk_sndmsg_off = 0;
1009
1010 skb_fill_page_desc(skb, i, page, 0, 0);
1011 frag = &skb_shinfo(skb)->frags[i];
1012 skb->truesize += PAGE_SIZE;
1013 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1014 } else {
1015 err = -EMSGSIZE;
1016 goto error;
1017 }
1018 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1019 err = -EFAULT;
1020 goto error;
1021 }
1022 sk->sk_sndmsg_off += copy;
1023 frag->size += copy;
1024 skb->len += copy;
1025 skb->data_len += copy;
1026 }
1027 offset += copy;
1028 length -= copy;
1029 }
1030
1031 return 0;
1032
1033error:
1034 inet->cork.length -= length;
1035 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001036 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037}
1038
1039ssize_t ip_append_page(struct sock *sk, struct page *page,
1040 int offset, size_t size, int flags)
1041{
1042 struct inet_sock *inet = inet_sk(sk);
1043 struct sk_buff *skb;
1044 struct rtable *rt;
1045 struct ip_options *opt = NULL;
1046 int hh_len;
1047 int mtu;
1048 int len;
1049 int err;
1050 unsigned int maxfraglen, fragheaderlen, fraggap;
1051
1052 if (inet->hdrincl)
1053 return -EPERM;
1054
1055 if (flags&MSG_PROBE)
1056 return 0;
1057
1058 if (skb_queue_empty(&sk->sk_write_queue))
1059 return -EINVAL;
1060
1061 rt = inet->cork.rt;
1062 if (inet->cork.flags & IPCORK_OPT)
1063 opt = inet->cork.opt;
1064
1065 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1066 return -EOPNOTSUPP;
1067
1068 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1069 mtu = inet->cork.fragsize;
1070
1071 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1072 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1073
1074 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1075 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1076 return -EMSGSIZE;
1077 }
1078
1079 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1080 return -EINVAL;
1081
1082 inet->cork.length += size;
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001083 if ((sk->sk_protocol == IPPROTO_UDP) &&
Herbert Xu79671682006-06-22 02:40:14 -07001084 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1085 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
Herbert Xuf83ef8c2006-06-30 13:37:03 -07001086 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Herbert Xu79671682006-06-22 02:40:14 -07001087 }
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
1090 while (size > 0) {
1091 int i;
1092
Herbert Xu89114af2006-07-08 13:34:32 -07001093 if (skb_is_gso(skb))
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001094 len = size;
1095 else {
1096
1097 /* Check if the remaining data fits into current packet. */
1098 len = mtu - skb->len;
1099 if (len < size)
1100 len = maxfraglen - skb->len;
1101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 if (len <= 0) {
1103 struct sk_buff *skb_prev;
1104 char *data;
1105 struct iphdr *iph;
1106 int alloclen;
1107
1108 skb_prev = skb;
Jayachandran C0d0d2bb2005-10-13 11:43:02 -07001109 fraggap = skb_prev->len - maxfraglen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
1111 alloclen = fragheaderlen + hh_len + fraggap + 15;
1112 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1113 if (unlikely(!skb)) {
1114 err = -ENOBUFS;
1115 goto error;
1116 }
1117
1118 /*
1119 * Fill in the control structures
1120 */
1121 skb->ip_summed = CHECKSUM_NONE;
1122 skb->csum = 0;
1123 skb_reserve(skb, hh_len);
1124
1125 /*
1126 * Find where to start putting bytes.
1127 */
1128 data = skb_put(skb, fragheaderlen + fraggap);
1129 skb->nh.iph = iph = (struct iphdr *)data;
1130 data += fragheaderlen;
1131 skb->h.raw = data;
1132
1133 if (fraggap) {
1134 skb->csum = skb_copy_and_csum_bits(
1135 skb_prev, maxfraglen,
1136 data, fraggap, 0);
1137 skb_prev->csum = csum_sub(skb_prev->csum,
1138 skb->csum);
Herbert Xue9fa4f72006-08-13 20:12:58 -07001139 pskb_trim_unique(skb_prev, maxfraglen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 }
1141
1142 /*
1143 * Put the packet on the pending queue.
1144 */
1145 __skb_queue_tail(&sk->sk_write_queue, skb);
1146 continue;
1147 }
1148
1149 i = skb_shinfo(skb)->nr_frags;
1150 if (len > size)
1151 len = size;
1152 if (skb_can_coalesce(skb, i, page, offset)) {
1153 skb_shinfo(skb)->frags[i-1].size += len;
1154 } else if (i < MAX_SKB_FRAGS) {
1155 get_page(page);
1156 skb_fill_page_desc(skb, i, page, offset, len);
1157 } else {
1158 err = -EMSGSIZE;
1159 goto error;
1160 }
1161
1162 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viro44bb9362006-11-14 21:36:14 -08001163 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 csum = csum_page(page, offset, len);
1165 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1166 }
1167
1168 skb->len += len;
1169 skb->data_len += len;
1170 offset += len;
1171 size -= len;
1172 }
1173 return 0;
1174
1175error:
1176 inet->cork.length -= size;
1177 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1178 return err;
1179}
1180
1181/*
1182 * Combined all pending IP fragments on the socket as one IP datagram
1183 * and push them out.
1184 */
1185int ip_push_pending_frames(struct sock *sk)
1186{
1187 struct sk_buff *skb, *tmp_skb;
1188 struct sk_buff **tail_skb;
1189 struct inet_sock *inet = inet_sk(sk);
1190 struct ip_options *opt = NULL;
1191 struct rtable *rt = inet->cork.rt;
1192 struct iphdr *iph;
Alexey Dobriyan76ab6082006-01-06 13:24:29 -08001193 __be16 df = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 __u8 ttl;
1195 int err = 0;
1196
1197 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1198 goto out;
1199 tail_skb = &(skb_shinfo(skb)->frag_list);
1200
1201 /* move skb->data to ip header from ext header */
1202 if (skb->data < skb->nh.raw)
1203 __skb_pull(skb, skb->nh.raw - skb->data);
1204 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1205 __skb_pull(tmp_skb, skb->h.raw - skb->nh.raw);
1206 *tail_skb = tmp_skb;
1207 tail_skb = &(tmp_skb->next);
1208 skb->len += tmp_skb->len;
1209 skb->data_len += tmp_skb->len;
1210 skb->truesize += tmp_skb->truesize;
1211 __sock_put(tmp_skb->sk);
1212 tmp_skb->destructor = NULL;
1213 tmp_skb->sk = NULL;
1214 }
1215
1216 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1217 * to fragment the frame generated here. No matter, what transforms
1218 * how transforms change size of the packet, it will come out.
1219 */
1220 if (inet->pmtudisc != IP_PMTUDISC_DO)
1221 skb->local_df = 1;
1222
1223 /* DF bit is set when we want to see DF on outgoing frames.
1224 * If local_df is set too, we still allow to fragment this frame
1225 * locally. */
1226 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1227 (skb->len <= dst_mtu(&rt->u.dst) &&
1228 ip_dont_fragment(sk, &rt->u.dst)))
1229 df = htons(IP_DF);
1230
1231 if (inet->cork.flags & IPCORK_OPT)
1232 opt = inet->cork.opt;
1233
1234 if (rt->rt_type == RTN_MULTICAST)
1235 ttl = inet->mc_ttl;
1236 else
1237 ttl = ip_select_ttl(inet, &rt->u.dst);
1238
1239 iph = (struct iphdr *)skb->data;
1240 iph->version = 4;
1241 iph->ihl = 5;
1242 if (opt) {
1243 iph->ihl += opt->optlen>>2;
1244 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1245 }
1246 iph->tos = inet->tos;
1247 iph->tot_len = htons(skb->len);
1248 iph->frag_off = df;
Alexey Kuznetsov1a55d572006-03-22 14:27:59 -08001249 ip_select_ident(iph, &rt->u.dst, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 iph->ttl = ttl;
1251 iph->protocol = sk->sk_protocol;
1252 iph->saddr = rt->rt_src;
1253 iph->daddr = rt->rt_dst;
1254 ip_send_check(iph);
1255
1256 skb->priority = sk->sk_priority;
1257 skb->dst = dst_clone(&rt->u.dst);
1258
1259 /* Netfilter gets whole the not fragmented skb. */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001260 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 skb->dst->dev, dst_output);
1262 if (err) {
1263 if (err > 0)
1264 err = inet->recverr ? net_xmit_errno(err) : 0;
1265 if (err)
1266 goto error;
1267 }
1268
1269out:
1270 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001271 kfree(inet->cork.opt);
1272 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (inet->cork.rt) {
1274 ip_rt_put(inet->cork.rt);
1275 inet->cork.rt = NULL;
1276 }
1277 return err;
1278
1279error:
1280 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1281 goto out;
1282}
1283
1284/*
1285 * Throw away all pending data on the socket.
1286 */
1287void ip_flush_pending_frames(struct sock *sk)
1288{
1289 struct inet_sock *inet = inet_sk(sk);
1290 struct sk_buff *skb;
1291
1292 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1293 kfree_skb(skb);
1294
1295 inet->cork.flags &= ~IPCORK_OPT;
Jesper Juhla51482b2005-11-08 09:41:34 -08001296 kfree(inet->cork.opt);
1297 inet->cork.opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if (inet->cork.rt) {
1299 ip_rt_put(inet->cork.rt);
1300 inet->cork.rt = NULL;
1301 }
1302}
1303
1304
1305/*
1306 * Fetch data from kernel space and fill in checksum if needed.
1307 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001308static int ip_reply_glue_bits(void *dptr, char *to, int offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 int len, int odd, struct sk_buff *skb)
1310{
Al Viro50842052006-11-14 21:36:34 -08001311 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
1313 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1314 skb->csum = csum_block_add(skb->csum, csum, odd);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001315 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
1317
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001318/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 * Generic function to send a packet as reply to another packet.
1320 * Used to send TCP resets so far. ICMP should use this function too.
1321 *
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001322 * Should run single threaded per socket because it uses the sock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 * structure to pass arguments.
1324 *
1325 * LATER: switch from ip_build_xmit to ip_append_*
1326 */
1327void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1328 unsigned int len)
1329{
1330 struct inet_sock *inet = inet_sk(sk);
1331 struct {
1332 struct ip_options opt;
1333 char data[40];
1334 } replyopts;
1335 struct ipcm_cookie ipc;
Al Viro3ca3c682006-09-27 18:28:07 -07001336 __be32 daddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 struct rtable *rt = (struct rtable*)skb->dst;
1338
1339 if (ip_options_echo(&replyopts.opt, skb))
1340 return;
1341
1342 daddr = ipc.addr = rt->rt_src;
1343 ipc.opt = NULL;
1344
1345 if (replyopts.opt.optlen) {
1346 ipc.opt = &replyopts.opt;
1347
1348 if (ipc.opt->srr)
1349 daddr = replyopts.opt.faddr;
1350 }
1351
1352 {
1353 struct flowi fl = { .nl_u = { .ip4_u =
1354 { .daddr = daddr,
1355 .saddr = rt->rt_spec_dst,
1356 .tos = RT_TOS(skb->nh.iph->tos) } },
1357 /* Not quite clean, but right. */
1358 .uli_u = { .ports =
1359 { .sport = skb->h.th->dest,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09001360 .dport = skb->h.th->source } },
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 .proto = sk->sk_protocol };
Venkat Yekkiralabeb8d132006-08-04 23:12:42 -07001362 security_skb_classify_flow(skb, &fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (ip_route_output_key(&rt, &fl))
1364 return;
1365 }
1366
1367 /* And let IP do all the hard work.
1368
1369 This chunk is not reenterable, hence spinlock.
1370 Note that it uses the fact, that this function is called
1371 with locally disabled BH and that sk cannot be already spinlocked.
1372 */
1373 bh_lock_sock(sk);
1374 inet->tos = skb->nh.iph->tos;
1375 sk->sk_priority = skb->priority;
1376 sk->sk_protocol = skb->nh.iph->protocol;
1377 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1378 &ipc, rt, MSG_DONTWAIT);
1379 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1380 if (arg->csumoffset >= 0)
Al Virod3bc23e2006-11-14 21:24:49 -08001381 *((__sum16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 skb->ip_summed = CHECKSUM_NONE;
1383 ip_push_pending_frames(sk);
1384 }
1385
1386 bh_unlock_sock(sk);
1387
1388 ip_rt_put(rt);
1389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391void __init ip_init(void)
1392{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 ip_rt_init();
1394 inet_initpeers();
1395
1396#if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1397 igmp_mc_proc_init();
1398#endif
1399}
1400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401EXPORT_SYMBOL(ip_generic_getfrag);
1402EXPORT_SYMBOL(ip_queue_xmit);
1403EXPORT_SYMBOL(ip_send_check);