?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.32.42
[linux/fpc-iii.git] / net / ipv4 / ip_output.c
blob44b79107866fcfbf997f9850f17184c659109d7e
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The Internet Protocol (IP) output module.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Donald Becker, <becker@super.org>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Richard Underwood
13 * Stefan Becker, <stefanb@yello.ping.de>
14 * Jorge Cwik, <jorge@laser.satlink.net>
15 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
16 * Hirokazu Takahashi, <taka@valinux.co.jp>
17 *
18 * See ip_input.c for original log
19 *
20 * Fixes:
21 * Alan Cox : Missing nonblock feature in ip_build_xmit.
22 * Mike Kilburn : htons() missing in ip_build_xmit.
23 * Bradford Johnson: Fix faulty handling of some frames when
24 * no route is found.
25 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
26 * (in case if packet not accepted by
27 * output firewall rules)
28 * Mike McLagan : Routing by source
29 * Alexey Kuznetsov: use new route cache
30 * Andi Kleen: Fix broken PMTU recovery and remove
31 * some redundant tests.
32 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
33 * Andi Kleen : Replace ip_reply with ip_send_reply.
34 * Andi Kleen : Split fast and slow ip_build_xmit path
35 * for decreased register pressure on x86
36 * and more readibility.
37 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
38 * silently drop skb instead of failing with -EPERM.
39 * Detlev Wengorz : Copy protocol for fragments.
40 * Hirokazu Takahashi: HW checksumming for outgoing UDP
41 * datagrams.
42 * Hirokazu Takahashi: sendfile() on UDP works now.
43 */
44
45 #include <asm/uaccess.h>
46 #include <asm/system.h>
47 #include <linux/module.h>
48 #include <linux/types.h>
49 #include <linux/kernel.h>
50 #include <linux/mm.h>
51 #include <linux/string.h>
52 #include <linux/errno.h>
53 #include <linux/highmem.h>
54
55 #include <linux/socket.h>
56 #include <linux/sockios.h>
57 #include <linux/in.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/etherdevice.h>
61 #include <linux/proc_fs.h>
62 #include <linux/stat.h>
63 #include <linux/init.h>
64
65 #include <net/snmp.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <net/route.h>
69 #include <net/xfrm.h>
70 #include <linux/skbuff.h>
71 #include <net/sock.h>
72 #include <net/arp.h>
73 #include <net/icmp.h>
74 #include <net/checksum.h>
75 #include <net/inetpeer.h>
76 #include <linux/igmp.h>
77 #include <linux/netfilter_ipv4.h>
78 #include <linux/netfilter_bridge.h>
79 #include <linux/mroute.h>
80 #include <linux/netlink.h>
81 #include <linux/tcp.h>
82
83 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84
85 /* Generate a checksum for an outgoing IP datagram. */
86 __inline__ void ip_send_check(struct iphdr *iph)
87 {
88 iph->check = 0;
89 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
90 }
91
92 int __ip_local_out(struct sk_buff *skb)
93 {
94 struct iphdr *iph = ip_hdr(skb);
95
96 iph->tot_len = htons(skb->len);
97 ip_send_check(iph);
98 return nf_hook(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, skb_dst(skb)->dev,
99 dst_output);
100 }
101
102 int ip_local_out(struct sk_buff *skb)
103 {
104 int err;
105
106 err = __ip_local_out(skb);
107 if (likely(err == 1))
108 err = dst_output(skb);
109
110 return err;
111 }
112 EXPORT_SYMBOL_GPL(ip_local_out);
113
114 /* dev_loopback_xmit for use with netfilter. */
115 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
116 {
117 skb_reset_mac_header(newskb);
118 __skb_pull(newskb, skb_network_offset(newskb));
119 newskb->pkt_type = PACKET_LOOPBACK;
120 newskb->ip_summed = CHECKSUM_UNNECESSARY;
121 WARN_ON(!skb_dst(newskb));
122 netif_rx(newskb);
123 return 0;
124 }
125
126 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
127 {
128 int ttl = inet->uc_ttl;
129
130 if (ttl < 0)
131 ttl = dst_metric(dst, RTAX_HOPLIMIT);
132 return ttl;
133 }
134
135 /*
136 * Add an ip header to a skbuff and send it out.
137 *
138 */
139 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
140 __be32 saddr, __be32 daddr, struct ip_options *opt)
141 {
142 struct inet_sock *inet = inet_sk(sk);
143 struct rtable *rt = skb_rtable(skb);
144 struct iphdr *iph;
145
146 /* Build the IP header. */
147 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
148 skb_reset_network_header(skb);
149 iph = ip_hdr(skb);
150 iph->version = 4;
151 iph->ihl = 5;
152 iph->tos = inet->tos;
153 if (ip_dont_fragment(sk, &rt->u.dst))
154 iph->frag_off = htons(IP_DF);
155 else
156 iph->frag_off = 0;
157 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
158 iph->daddr = rt->rt_dst;
159 iph->saddr = rt->rt_src;
160 iph->protocol = sk->sk_protocol;
161 ip_select_ident(iph, &rt->u.dst, sk);
162
163 if (opt && opt->optlen) {
164 iph->ihl += opt->optlen>>2;
165 ip_options_build(skb, opt, daddr, rt, 0);
166 }
167
168 skb->priority = sk->sk_priority;
169 skb->mark = sk->sk_mark;
170
171 /* Send it out. */
172 return ip_local_out(skb);
173 }
174
175 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
176
177 static inline int ip_finish_output2(struct sk_buff *skb)
178 {
179 struct dst_entry *dst = skb_dst(skb);
180 struct rtable *rt = (struct rtable *)dst;
181 struct net_device *dev = dst->dev;
182 unsigned int hh_len = LL_RESERVED_SPACE(dev);
183
184 if (rt->rt_type == RTN_MULTICAST) {
185 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
186 } else if (rt->rt_type == RTN_BROADCAST)
187 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
188
189 /* Be paranoid, rather than too clever. */
190 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
191 struct sk_buff *skb2;
192
193 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
194 if (skb2 == NULL) {
195 kfree_skb(skb);
196 return -ENOMEM;
197 }
198 if (skb->sk)
199 skb_set_owner_w(skb2, skb->sk);
200 kfree_skb(skb);
201 skb = skb2;
202 }
203
204 if (dst->hh)
205 return neigh_hh_output(dst->hh, skb);
206 else if (dst->neighbour)
207 return dst->neighbour->output(skb);
208
209 if (net_ratelimit())
210 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
211 kfree_skb(skb);
212 return -EINVAL;
213 }
214
215 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
216 {
217 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
218
219 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
220 skb_dst(skb)->dev->mtu : dst_mtu(skb_dst(skb));
221 }
222
223 static int ip_finish_output(struct sk_buff *skb)
224 {
225 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
226 /* Policy lookup after SNAT yielded a new policy */
227 if (skb_dst(skb)->xfrm != NULL) {
228 IPCB(skb)->flags |= IPSKB_REROUTED;
229 return dst_output(skb);
230 }
231 #endif
232 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
233 return ip_fragment(skb, ip_finish_output2);
234 else
235 return ip_finish_output2(skb);
236 }
237
238 int ip_mc_output(struct sk_buff *skb)
239 {
240 struct sock *sk = skb->sk;
241 struct rtable *rt = skb_rtable(skb);
242 struct net_device *dev = rt->u.dst.dev;
243
244 /*
245 * If the indicated interface is up and running, send the packet.
246 */
247 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
248
249 skb->dev = dev;
250 skb->protocol = htons(ETH_P_IP);
251
252 /*
253 * Multicasts are looped back for other local users
254 */
255
256 if (rt->rt_flags&RTCF_MULTICAST) {
257 if ((!sk || inet_sk(sk)->mc_loop)
258 #ifdef CONFIG_IP_MROUTE
259 /* Small optimization: do not loopback not local frames,
260 which returned after forwarding; they will be dropped
261 by ip_mr_input in any case.
262 Note, that local frames are looped back to be delivered
263 to local recipients.
264
265 This check is duplicated in ip_mr_input at the moment.
266 */
267 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
268 #endif
269 ) {
270 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
271 if (newskb)
272 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb,
273 NULL, newskb->dev,
274 ip_dev_loopback_xmit);
275 }
276
277 /* Multicasts with ttl 0 must not go beyond the host */
278
279 if (ip_hdr(skb)->ttl == 0) {
280 kfree_skb(skb);
281 return 0;
282 }
283 }
284
285 if (rt->rt_flags&RTCF_BROADCAST) {
286 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
287 if (newskb)
288 NF_HOOK(PF_INET, NF_INET_POST_ROUTING, newskb, NULL,
289 newskb->dev, ip_dev_loopback_xmit);
290 }
291
292 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, skb->dev,
293 ip_finish_output,
294 !(IPCB(skb)->flags & IPSKB_REROUTED));
295 }
296
297 int ip_output(struct sk_buff *skb)
298 {
299 struct net_device *dev = skb_dst(skb)->dev;
300
301 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
302
303 skb->dev = dev;
304 skb->protocol = htons(ETH_P_IP);
305
306 return NF_HOOK_COND(PF_INET, NF_INET_POST_ROUTING, skb, NULL, dev,
307 ip_finish_output,
308 !(IPCB(skb)->flags & IPSKB_REROUTED));
309 }
310
311 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
312 {
313 struct sock *sk = skb->sk;
314 struct inet_sock *inet = inet_sk(sk);
315 struct ip_options *opt = inet->opt;
316 struct rtable *rt;
317 struct iphdr *iph;
318
319 /* Skip all of this if the packet is already routed,
320 * f.e. by something like SCTP.
321 */
322 rt = skb_rtable(skb);
323 if (rt != NULL)
324 goto packet_routed;
325
326 /* Make sure we can route this packet. */
327 rt = (struct rtable *)__sk_dst_check(sk, 0);
328 if (rt == NULL) {
329 __be32 daddr;
330
331 /* Use correct destination address if we have options. */
332 daddr = inet->daddr;
333 if(opt && opt->srr)
334 daddr = opt->faddr;
335
336 {
337 struct flowi fl = { .oif = sk->sk_bound_dev_if,
338 .mark = sk->sk_mark,
339 .nl_u = { .ip4_u =
340 { .daddr = daddr,
341 .saddr = inet->saddr,
342 .tos = RT_CONN_FLAGS(sk) } },
343 .proto = sk->sk_protocol,
344 .flags = inet_sk_flowi_flags(sk),
345 .uli_u = { .ports =
346 { .sport = inet->sport,
347 .dport = inet->dport } } };
348
349 /* If this fails, retransmit mechanism of transport layer will
350 * keep trying until route appears or the connection times
351 * itself out.
352 */
353 security_sk_classify_flow(sk, &fl);
354 if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0))
355 goto no_route;
356 }
357 sk_setup_caps(sk, &rt->u.dst);
358 }
359 skb_dst_set(skb, dst_clone(&rt->u.dst));
360
361 packet_routed:
362 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
363 goto no_route;
364
365 /* OK, we know where to send it, allocate and build IP header. */
366 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
367 skb_reset_network_header(skb);
368 iph = ip_hdr(skb);
369 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
370 if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
371 iph->frag_off = htons(IP_DF);
372 else
373 iph->frag_off = 0;
374 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
375 iph->protocol = sk->sk_protocol;
376 iph->saddr = rt->rt_src;
377 iph->daddr = rt->rt_dst;
378 /* Transport layer set skb->h.foo itself. */
379
380 if (opt && opt->optlen) {
381 iph->ihl += opt->optlen >> 2;
382 ip_options_build(skb, opt, inet->daddr, rt, 0);
383 }
384
385 ip_select_ident_more(iph, &rt->u.dst, sk,
386 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
387
388 skb->priority = sk->sk_priority;
389 skb->mark = sk->sk_mark;
390
391 return ip_local_out(skb);
392
393 no_route:
394 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
395 kfree_skb(skb);
396 return -EHOSTUNREACH;
397 }
398
399
400 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
401 {
402 to->pkt_type = from->pkt_type;
403 to->priority = from->priority;
404 to->protocol = from->protocol;
405 skb_dst_drop(to);
406 skb_dst_set(to, dst_clone(skb_dst(from)));
407 to->dev = from->dev;
408 to->mark = from->mark;
409
410 /* Copy the flags to each fragment. */
411 IPCB(to)->flags = IPCB(from)->flags;
412
413 #ifdef CONFIG_NET_SCHED
414 to->tc_index = from->tc_index;
415 #endif
416 nf_copy(to, from);
417 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
418 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
419 to->nf_trace = from->nf_trace;
420 #endif
421 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
422 to->ipvs_property = from->ipvs_property;
423 #endif
424 skb_copy_secmark(to, from);
425 }
426
427 /*
428 * This IP datagram is too large to be sent in one piece. Break it up into
429 * smaller pieces (each of size equal to IP header plus
430 * a block of the data of the original IP data part) that will yet fit in a
431 * single device frame, and queue such a frame for sending.
432 */
433
434 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
435 {
436 struct iphdr *iph;
437 int raw = 0;
438 int ptr;
439 struct net_device *dev;
440 struct sk_buff *skb2;
441 unsigned int mtu, hlen, left, len, ll_rs, pad;
442 int offset;
443 __be16 not_last_frag;
444 struct rtable *rt = skb_rtable(skb);
445 int err = 0;
446
447 dev = rt->u.dst.dev;
448
449 /*
450 * Point into the IP datagram header.
451 */
452
453 iph = ip_hdr(skb);
454
455 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
456 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
457 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
458 htonl(ip_skb_dst_mtu(skb)));
459 kfree_skb(skb);
460 return -EMSGSIZE;
461 }
462
463 /*
464 * Setup starting values.
465 */
466
467 hlen = iph->ihl * 4;
468 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
469 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
470
471 /* When frag_list is given, use it. First, check its validity:
472 * some transformers could create wrong frag_list or break existing
473 * one, it is not prohibited. In this case fall back to copying.
474 *
475 * LATER: this step can be merged to real generation of fragments,
476 * we can switch to copy when see the first bad fragment.
477 */
478 if (skb_has_frags(skb)) {
479 struct sk_buff *frag, *frag2;
480 int first_len = skb_pagelen(skb);
481
482 if (first_len - hlen > mtu ||
483 ((first_len - hlen) & 7) ||
484 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
485 skb_cloned(skb))
486 goto slow_path;
487
488 skb_walk_frags(skb, frag) {
489 /* Correct geometry. */
490 if (frag->len > mtu ||
491 ((frag->len & 7) && frag->next) ||
492 skb_headroom(frag) < hlen)
493 goto slow_path_clean;
494
495 /* Partially cloned skb? */
496 if (skb_shared(frag))
497 goto slow_path_clean;
498
499 BUG_ON(frag->sk);
500 if (skb->sk) {
501 frag->sk = skb->sk;
502 frag->destructor = sock_wfree;
503 }
504 skb->truesize -= frag->truesize;
505 }
506
507 /* Everything is OK. Generate! */
508
509 err = 0;
510 offset = 0;
511 frag = skb_shinfo(skb)->frag_list;
512 skb_frag_list_init(skb);
513 skb->data_len = first_len - skb_headlen(skb);
514 skb->len = first_len;
515 iph->tot_len = htons(first_len);
516 iph->frag_off = htons(IP_MF);
517 ip_send_check(iph);
518
519 for (;;) {
520 /* Prepare header of the next frame,
521 * before previous one went down. */
522 if (frag) {
523 frag->ip_summed = CHECKSUM_NONE;
524 skb_reset_transport_header(frag);
525 __skb_push(frag, hlen);
526 skb_reset_network_header(frag);
527 memcpy(skb_network_header(frag), iph, hlen);
528 iph = ip_hdr(frag);
529 iph->tot_len = htons(frag->len);
530 ip_copy_metadata(frag, skb);
531 if (offset == 0)
532 ip_options_fragment(frag);
533 offset += skb->len - hlen;
534 iph->frag_off = htons(offset>>3);
535 if (frag->next != NULL)
536 iph->frag_off |= htons(IP_MF);
537 /* Ready, complete checksum */
538 ip_send_check(iph);
539 }
540
541 err = output(skb);
542
543 if (!err)
544 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
545 if (err || !frag)
546 break;
547
548 skb = frag;
549 frag = skb->next;
550 skb->next = NULL;
551 }
552
553 if (err == 0) {
554 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
555 return 0;
556 }
557
558 while (frag) {
559 skb = frag->next;
560 kfree_skb(frag);
561 frag = skb;
562 }
563 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
564 return err;
565
566 slow_path_clean:
567 skb_walk_frags(skb, frag2) {
568 if (frag2 == frag)
569 break;
570 frag2->sk = NULL;
571 frag2->destructor = NULL;
572 skb->truesize += frag2->truesize;
573 }
574 }
575
576 slow_path:
577 left = skb->len - hlen; /* Space per frame */
578 ptr = raw + hlen; /* Where to start from */
579
580 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
581 * we need to make room for the encapsulating header
582 */
583 pad = nf_bridge_pad(skb);
584 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
585 mtu -= pad;
586
587 /*
588 * Fragment the datagram.
589 */
590
591 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
592 not_last_frag = iph->frag_off & htons(IP_MF);
593
594 /*
595 * Keep copying data until we run out.
596 */
597
598 while (left > 0) {
599 len = left;
600 /* IF: it doesn't fit, use 'mtu' - the data space left */
601 if (len > mtu)
602 len = mtu;
603 /* IF: we are not sending upto and including the packet end
604 then align the next start on an eight byte boundary */
605 if (len < left) {
606 len &= ~7;
607 }
608 /*
609 * Allocate buffer.
610 */
611
612 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
613 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
614 err = -ENOMEM;
615 goto fail;
616 }
617
618 /*
619 * Set up data on packet
620 */
621
622 ip_copy_metadata(skb2, skb);
623 skb_reserve(skb2, ll_rs);
624 skb_put(skb2, len + hlen);
625 skb_reset_network_header(skb2);
626 skb2->transport_header = skb2->network_header + hlen;
627
628 /*
629 * Charge the memory for the fragment to any owner
630 * it might possess
631 */
632
633 if (skb->sk)
634 skb_set_owner_w(skb2, skb->sk);
635
636 /*
637 * Copy the packet header into the new buffer.
638 */
639
640 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
641
642 /*
643 * Copy a block of the IP datagram.
644 */
645 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
646 BUG();
647 left -= len;
648
649 /*
650 * Fill in the new header fields.
651 */
652 iph = ip_hdr(skb2);
653 iph->frag_off = htons((offset >> 3));
654
655 /* ANK: dirty, but effective trick. Upgrade options only if
656 * the segment to be fragmented was THE FIRST (otherwise,
657 * options are already fixed) and make it ONCE
658 * on the initial skb, so that all the following fragments
659 * will inherit fixed options.
660 */
661 if (offset == 0)
662 ip_options_fragment(skb);
663
664 /*
665 * Added AC : If we are fragmenting a fragment that's not the
666 * last fragment then keep MF on each bit
667 */
668 if (left > 0 || not_last_frag)
669 iph->frag_off |= htons(IP_MF);
670 ptr += len;
671 offset += len;
672
673 /*
674 * Put this fragment into the sending queue.
675 */
676 iph->tot_len = htons(len + hlen);
677
678 ip_send_check(iph);
679
680 err = output(skb2);
681 if (err)
682 goto fail;
683
684 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
685 }
686 kfree_skb(skb);
687 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
688 return err;
689
690 fail:
691 kfree_skb(skb);
692 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
693 return err;
694 }
695
696 EXPORT_SYMBOL(ip_fragment);
697
698 int
699 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
700 {
701 struct iovec *iov = from;
702
703 if (skb->ip_summed == CHECKSUM_PARTIAL) {
704 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
705 return -EFAULT;
706 } else {
707 __wsum csum = 0;
708 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
709 return -EFAULT;
710 skb->csum = csum_block_add(skb->csum, csum, odd);
711 }
712 return 0;
713 }
714
715 static inline __wsum
716 csum_page(struct page *page, int offset, int copy)
717 {
718 char *kaddr;
719 __wsum csum;
720 kaddr = kmap(page);
721 csum = csum_partial(kaddr + offset, copy, 0);
722 kunmap(page);
723 return csum;
724 }
725
726 static inline int ip_ufo_append_data(struct sock *sk,
727 int getfrag(void *from, char *to, int offset, int len,
728 int odd, struct sk_buff *skb),
729 void *from, int length, int hh_len, int fragheaderlen,
730 int transhdrlen, int mtu, unsigned int flags)
731 {
732 struct sk_buff *skb;
733 int err;
734
735 /* There is support for UDP fragmentation offload by network
736 * device, so create one single skb packet containing complete
737 * udp datagram
738 */
739 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
740 skb = sock_alloc_send_skb(sk,
741 hh_len + fragheaderlen + transhdrlen + 20,
742 (flags & MSG_DONTWAIT), &err);
743
744 if (skb == NULL)
745 return err;
746
747 /* reserve space for Hardware header */
748 skb_reserve(skb, hh_len);
749
750 /* create space for UDP/IP header */
751 skb_put(skb, fragheaderlen + transhdrlen);
752
753 /* initialize network header pointer */
754 skb_reset_network_header(skb);
755
756 /* initialize protocol header pointer */
757 skb->transport_header = skb->network_header + fragheaderlen;
758
759 skb->ip_summed = CHECKSUM_PARTIAL;
760 skb->csum = 0;
761 sk->sk_sndmsg_off = 0;
762
763 /* specify the length of each IP datagram fragment */
764 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
765 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
766 __skb_queue_tail(&sk->sk_write_queue, skb);
767 }
768
769 return skb_append_datato_frags(sk, skb, getfrag, from,
770 (length - transhdrlen));
771 }
772
773 /*
774 * ip_append_data() and ip_append_page() can make one large IP datagram
775 * from many pieces of data. Each pieces will be holded on the socket
776 * until ip_push_pending_frames() is called. Each piece can be a page
777 * or non-page data.
778 *
779 * Not only UDP, other transport protocols - e.g. raw sockets - can use
780 * this interface potentially.
781 *
782 * LATER: length must be adjusted by pad at tail, when it is required.
783 */
784 int ip_append_data(struct sock *sk,
785 int getfrag(void *from, char *to, int offset, int len,
786 int odd, struct sk_buff *skb),
787 void *from, int length, int transhdrlen,
788 struct ipcm_cookie *ipc, struct rtable **rtp,
789 unsigned int flags)
790 {
791 struct inet_sock *inet = inet_sk(sk);
792 struct sk_buff *skb;
793
794 struct ip_options *opt = NULL;
795 int hh_len;
796 int exthdrlen;
797 int mtu;
798 int copy;
799 int err;
800 int offset = 0;
801 unsigned int maxfraglen, fragheaderlen;
802 int csummode = CHECKSUM_NONE;
803 struct rtable *rt;
804
805 if (flags&MSG_PROBE)
806 return 0;
807
808 if (skb_queue_empty(&sk->sk_write_queue)) {
809 /*
810 * setup for corking.
811 */
812 opt = ipc->opt;
813 if (opt) {
814 if (inet->cork.opt == NULL) {
815 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
816 if (unlikely(inet->cork.opt == NULL))
817 return -ENOBUFS;
818 }
819 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
820 inet->cork.flags |= IPCORK_OPT;
821 inet->cork.addr = ipc->addr;
822 }
823 rt = *rtp;
824 if (unlikely(!rt))
825 return -EFAULT;
826 /*
827 * We steal reference to this route, caller should not release it
828 */
829 *rtp = NULL;
830 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
831 rt->u.dst.dev->mtu :
832 dst_mtu(rt->u.dst.path);
833 inet->cork.dst = &rt->u.dst;
834 inet->cork.length = 0;
835 sk->sk_sndmsg_page = NULL;
836 sk->sk_sndmsg_off = 0;
837 if ((exthdrlen = rt->u.dst.header_len) != 0) {
838 length += exthdrlen;
839 transhdrlen += exthdrlen;
840 }
841 } else {
842 rt = (struct rtable *)inet->cork.dst;
843 if (inet->cork.flags & IPCORK_OPT)
844 opt = inet->cork.opt;
845
846 transhdrlen = 0;
847 exthdrlen = 0;
848 mtu = inet->cork.fragsize;
849 }
850 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
851
852 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
853 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
854
855 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
856 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
857 return -EMSGSIZE;
858 }
859
860 /*
861 * transhdrlen > 0 means that this is the first fragment and we wish
862 * it won't be fragmented in the future.
863 */
864 if (transhdrlen &&
865 length + fragheaderlen <= mtu &&
866 rt->u.dst.dev->features & NETIF_F_V4_CSUM &&
867 !exthdrlen)
868 csummode = CHECKSUM_PARTIAL;
869
870 skb = skb_peek_tail(&sk->sk_write_queue);
871
872 inet->cork.length += length;
873 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
874 (sk->sk_protocol == IPPROTO_UDP) &&
875 (rt->u.dst.dev->features & NETIF_F_UFO)) {
876 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
877 fragheaderlen, transhdrlen, mtu,
878 flags);
879 if (err)
880 goto error;
881 return 0;
882 }
883
884 /* So, what's going on in the loop below?
885 *
886 * We use calculated fragment length to generate chained skb,
887 * each of segments is IP fragment ready for sending to network after
888 * adding appropriate IP header.
889 */
890
891 if (!skb)
892 goto alloc_new_skb;
893
894 while (length > 0) {
895 /* Check if the remaining data fits into current packet. */
896 copy = mtu - skb->len;
897 if (copy < length)
898 copy = maxfraglen - skb->len;
899 if (copy <= 0) {
900 char *data;
901 unsigned int datalen;
902 unsigned int fraglen;
903 unsigned int fraggap;
904 unsigned int alloclen;
905 struct sk_buff *skb_prev;
906 alloc_new_skb:
907 skb_prev = skb;
908 if (skb_prev)
909 fraggap = skb_prev->len - maxfraglen;
910 else
911 fraggap = 0;
912
913 /*
914 * If remaining data exceeds the mtu,
915 * we know we need more fragment(s).
916 */
917 datalen = length + fraggap;
918 if (datalen > mtu - fragheaderlen)
919 datalen = maxfraglen - fragheaderlen;
920 fraglen = datalen + fragheaderlen;
921
922 if ((flags & MSG_MORE) &&
923 !(rt->u.dst.dev->features&NETIF_F_SG))
924 alloclen = mtu;
925 else
926 alloclen = datalen + fragheaderlen;
927
928 /* The last fragment gets additional space at tail.
929 * Note, with MSG_MORE we overallocate on fragments,
930 * because we have no idea what fragment will be
931 * the last.
932 */
933 if (datalen == length + fraggap)
934 alloclen += rt->u.dst.trailer_len;
935
936 if (transhdrlen) {
937 skb = sock_alloc_send_skb(sk,
938 alloclen + hh_len + 15,
939 (flags & MSG_DONTWAIT), &err);
940 } else {
941 skb = NULL;
942 if (atomic_read(&sk->sk_wmem_alloc) <=
943 2 * sk->sk_sndbuf)
944 skb = sock_wmalloc(sk,
945 alloclen + hh_len + 15, 1,
946 sk->sk_allocation);
947 if (unlikely(skb == NULL))
948 err = -ENOBUFS;
949 else
950 /* only the initial fragment is
951 time stamped */
952 ipc->shtx.flags = 0;
953 }
954 if (skb == NULL)
955 goto error;
956
957 /*
958 * Fill in the control structures
959 */
960 skb->ip_summed = csummode;
961 skb->csum = 0;
962 skb_reserve(skb, hh_len);
963 *skb_tx(skb) = ipc->shtx;
964
965 /*
966 * Find where to start putting bytes.
967 */
968 data = skb_put(skb, fraglen);
969 skb_set_network_header(skb, exthdrlen);
970 skb->transport_header = (skb->network_header +
971 fragheaderlen);
972 data += fragheaderlen;
973
974 if (fraggap) {
975 skb->csum = skb_copy_and_csum_bits(
976 skb_prev, maxfraglen,
977 data + transhdrlen, fraggap, 0);
978 skb_prev->csum = csum_sub(skb_prev->csum,
979 skb->csum);
980 data += fraggap;
981 pskb_trim_unique(skb_prev, maxfraglen);
982 }
983
984 copy = datalen - transhdrlen - fraggap;
985 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
986 err = -EFAULT;
987 kfree_skb(skb);
988 goto error;
989 }
990
991 offset += copy;
992 length -= datalen - fraggap;
993 transhdrlen = 0;
994 exthdrlen = 0;
995 csummode = CHECKSUM_NONE;
996
997 /*
998 * Put the packet on the pending queue.
999 */
1000 __skb_queue_tail(&sk->sk_write_queue, skb);
1001 continue;
1002 }
1003
1004 if (copy > length)
1005 copy = length;
1006
1007 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
1008 unsigned int off;
1009
1010 off = skb->len;
1011 if (getfrag(from, skb_put(skb, copy),
1012 offset, copy, off, skb) < 0) {
1013 __skb_trim(skb, off);
1014 err = -EFAULT;
1015 goto error;
1016 }
1017 } else {
1018 int i = skb_shinfo(skb)->nr_frags;
1019 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1020 struct page *page = sk->sk_sndmsg_page;
1021 int off = sk->sk_sndmsg_off;
1022 unsigned int left;
1023
1024 if (page && (left = PAGE_SIZE - off) > 0) {
1025 if (copy >= left)
1026 copy = left;
1027 if (page != frag->page) {
1028 if (i == MAX_SKB_FRAGS) {
1029 err = -EMSGSIZE;
1030 goto error;
1031 }
1032 get_page(page);
1033 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1034 frag = &skb_shinfo(skb)->frags[i];
1035 }
1036 } else if (i < MAX_SKB_FRAGS) {
1037 if (copy > PAGE_SIZE)
1038 copy = PAGE_SIZE;
1039 page = alloc_pages(sk->sk_allocation, 0);
1040 if (page == NULL) {
1041 err = -ENOMEM;
1042 goto error;
1043 }
1044 sk->sk_sndmsg_page = page;
1045 sk->sk_sndmsg_off = 0;
1046
1047 skb_fill_page_desc(skb, i, page, 0, 0);
1048 frag = &skb_shinfo(skb)->frags[i];
1049 } else {
1050 err = -EMSGSIZE;
1051 goto error;
1052 }
1053 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1054 err = -EFAULT;
1055 goto error;
1056 }
1057 sk->sk_sndmsg_off += copy;
1058 frag->size += copy;
1059 skb->len += copy;
1060 skb->data_len += copy;
1061 skb->truesize += copy;
1062 atomic_add(copy, &sk->sk_wmem_alloc);
1063 }
1064 offset += copy;
1065 length -= copy;
1066 }
1067
1068 return 0;
1069
1070 error:
1071 inet->cork.length -= length;
1072 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1073 return err;
1074 }
1075
1076 ssize_t ip_append_page(struct sock *sk, struct page *page,
1077 int offset, size_t size, int flags)
1078 {
1079 struct inet_sock *inet = inet_sk(sk);
1080 struct sk_buff *skb;
1081 struct rtable *rt;
1082 struct ip_options *opt = NULL;
1083 int hh_len;
1084 int mtu;
1085 int len;
1086 int err;
1087 unsigned int maxfraglen, fragheaderlen, fraggap;
1088
1089 if (inet->hdrincl)
1090 return -EPERM;
1091
1092 if (flags&MSG_PROBE)
1093 return 0;
1094
1095 if (skb_queue_empty(&sk->sk_write_queue))
1096 return -EINVAL;
1097
1098 rt = (struct rtable *)inet->cork.dst;
1099 if (inet->cork.flags & IPCORK_OPT)
1100 opt = inet->cork.opt;
1101
1102 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1103 return -EOPNOTSUPP;
1104
1105 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1106 mtu = inet->cork.fragsize;
1107
1108 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1109 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1110
1111 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1112 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1113 return -EMSGSIZE;
1114 }
1115
1116 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1117 return -EINVAL;
1118
1119 inet->cork.length += size;
1120 if ((size + skb->len > mtu) &&
1121 (sk->sk_protocol == IPPROTO_UDP) &&
1122 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1123 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1124 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1125 }
1126
1127
1128 while (size > 0) {
1129 int i;
1130
1131 if (skb_is_gso(skb))
1132 len = size;
1133 else {
1134
1135 /* Check if the remaining data fits into current packet. */
1136 len = mtu - skb->len;
1137 if (len < size)
1138 len = maxfraglen - skb->len;
1139 }
1140 if (len <= 0) {
1141 struct sk_buff *skb_prev;
1142 int alloclen;
1143
1144 skb_prev = skb;
1145 fraggap = skb_prev->len - maxfraglen;
1146
1147 alloclen = fragheaderlen + hh_len + fraggap + 15;
1148 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1149 if (unlikely(!skb)) {
1150 err = -ENOBUFS;
1151 goto error;
1152 }
1153
1154 /*
1155 * Fill in the control structures
1156 */
1157 skb->ip_summed = CHECKSUM_NONE;
1158 skb->csum = 0;
1159 skb_reserve(skb, hh_len);
1160
1161 /*
1162 * Find where to start putting bytes.
1163 */
1164 skb_put(skb, fragheaderlen + fraggap);
1165 skb_reset_network_header(skb);
1166 skb->transport_header = (skb->network_header +
1167 fragheaderlen);
1168 if (fraggap) {
1169 skb->csum = skb_copy_and_csum_bits(skb_prev,
1170 maxfraglen,
1171 skb_transport_header(skb),
1172 fraggap, 0);
1173 skb_prev->csum = csum_sub(skb_prev->csum,
1174 skb->csum);
1175 pskb_trim_unique(skb_prev, maxfraglen);
1176 }
1177
1178 /*
1179 * Put the packet on the pending queue.
1180 */
1181 __skb_queue_tail(&sk->sk_write_queue, skb);
1182 continue;
1183 }
1184
1185 i = skb_shinfo(skb)->nr_frags;
1186 if (len > size)
1187 len = size;
1188 if (skb_can_coalesce(skb, i, page, offset)) {
1189 skb_shinfo(skb)->frags[i-1].size += len;
1190 } else if (i < MAX_SKB_FRAGS) {
1191 get_page(page);
1192 skb_fill_page_desc(skb, i, page, offset, len);
1193 } else {
1194 err = -EMSGSIZE;
1195 goto error;
1196 }
1197
1198 if (skb->ip_summed == CHECKSUM_NONE) {
1199 __wsum csum;
1200 csum = csum_page(page, offset, len);
1201 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1202 }
1203
1204 skb->len += len;
1205 skb->data_len += len;
1206 skb->truesize += len;
1207 atomic_add(len, &sk->sk_wmem_alloc);
1208 offset += len;
1209 size -= len;
1210 }
1211 return 0;
1212
1213 error:
1214 inet->cork.length -= size;
1215 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1216 return err;
1217 }
1218
1219 static void ip_cork_release(struct inet_sock *inet)
1220 {
1221 inet->cork.flags &= ~IPCORK_OPT;
1222 kfree(inet->cork.opt);
1223 inet->cork.opt = NULL;
1224 dst_release(inet->cork.dst);
1225 inet->cork.dst = NULL;
1226 }
1227
1228 /*
1229 * Combined all pending IP fragments on the socket as one IP datagram
1230 * and push them out.
1231 */
1232 int ip_push_pending_frames(struct sock *sk)
1233 {
1234 struct sk_buff *skb, *tmp_skb;
1235 struct sk_buff **tail_skb;
1236 struct inet_sock *inet = inet_sk(sk);
1237 struct net *net = sock_net(sk);
1238 struct ip_options *opt = NULL;
1239 struct rtable *rt = (struct rtable *)inet->cork.dst;
1240 struct iphdr *iph;
1241 __be16 df = 0;
1242 __u8 ttl;
1243 int err = 0;
1244
1245 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1246 goto out;
1247 tail_skb = &(skb_shinfo(skb)->frag_list);
1248
1249 /* move skb->data to ip header from ext header */
1250 if (skb->data < skb_network_header(skb))
1251 __skb_pull(skb, skb_network_offset(skb));
1252 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1253 __skb_pull(tmp_skb, skb_network_header_len(skb));
1254 *tail_skb = tmp_skb;
1255 tail_skb = &(tmp_skb->next);
1256 skb->len += tmp_skb->len;
1257 skb->data_len += tmp_skb->len;
1258 skb->truesize += tmp_skb->truesize;
1259 tmp_skb->destructor = NULL;
1260 tmp_skb->sk = NULL;
1261 }
1262
1263 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1264 * to fragment the frame generated here. No matter, what transforms
1265 * how transforms change size of the packet, it will come out.
1266 */
1267 if (inet->pmtudisc < IP_PMTUDISC_DO)
1268 skb->local_df = 1;
1269
1270 /* DF bit is set when we want to see DF on outgoing frames.
1271 * If local_df is set too, we still allow to fragment this frame
1272 * locally. */
1273 if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1274 (skb->len <= dst_mtu(&rt->u.dst) &&
1275 ip_dont_fragment(sk, &rt->u.dst)))
1276 df = htons(IP_DF);
1277
1278 if (inet->cork.flags & IPCORK_OPT)
1279 opt = inet->cork.opt;
1280
1281 if (rt->rt_type == RTN_MULTICAST)
1282 ttl = inet->mc_ttl;
1283 else
1284 ttl = ip_select_ttl(inet, &rt->u.dst);
1285
1286 iph = (struct iphdr *)skb->data;
1287 iph->version = 4;
1288 iph->ihl = 5;
1289 if (opt) {
1290 iph->ihl += opt->optlen>>2;
1291 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1292 }
1293 iph->tos = inet->tos;
1294 iph->frag_off = df;
1295 ip_select_ident(iph, &rt->u.dst, sk);
1296 iph->ttl = ttl;
1297 iph->protocol = sk->sk_protocol;
1298 iph->saddr = rt->rt_src;
1299 iph->daddr = rt->rt_dst;
1300
1301 skb->priority = sk->sk_priority;
1302 skb->mark = sk->sk_mark;
1303 /*
1304 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1305 * on dst refcount
1306 */
1307 inet->cork.dst = NULL;
1308 skb_dst_set(skb, &rt->u.dst);
1309
1310 if (iph->protocol == IPPROTO_ICMP)
1311 icmp_out_count(net, ((struct icmphdr *)
1312 skb_transport_header(skb))->type);
1313
1314 /* Netfilter gets whole the not fragmented skb. */
1315 err = ip_local_out(skb);
1316 if (err) {
1317 if (err > 0)
1318 err = net_xmit_errno(err);
1319 if (err)
1320 goto error;
1321 }
1322
1323 out:
1324 ip_cork_release(inet);
1325 return err;
1326
1327 error:
1328 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1329 goto out;
1330 }
1331
1332 /*
1333 * Throw away all pending data on the socket.
1334 */
1335 void ip_flush_pending_frames(struct sock *sk)
1336 {
1337 struct sk_buff *skb;
1338
1339 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1340 kfree_skb(skb);
1341
1342 ip_cork_release(inet_sk(sk));
1343 }
1344
1345
1346 /*
1347 * Fetch data from kernel space and fill in checksum if needed.
1348 */
1349 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1350 int len, int odd, struct sk_buff *skb)
1351 {
1352 __wsum csum;
1353
1354 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1355 skb->csum = csum_block_add(skb->csum, csum, odd);
1356 return 0;
1357 }
1358
1359 /*
1360 * Generic function to send a packet as reply to another packet.
1361 * Used to send TCP resets so far. ICMP should use this function too.
1362 *
1363 * Should run single threaded per socket because it uses the sock
1364 * structure to pass arguments.
1365 */
1366 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1367 unsigned int len)
1368 {
1369 struct inet_sock *inet = inet_sk(sk);
1370 struct {
1371 struct ip_options opt;
1372 char data[40];
1373 } replyopts;
1374 struct ipcm_cookie ipc;
1375 __be32 daddr;
1376 struct rtable *rt = skb_rtable(skb);
1377
1378 if (ip_options_echo(&replyopts.opt, skb))
1379 return;
1380
1381 daddr = ipc.addr = rt->rt_src;
1382 ipc.opt = NULL;
1383 ipc.shtx.flags = 0;
1384
1385 if (replyopts.opt.optlen) {
1386 ipc.opt = &replyopts.opt;
1387
1388 if (ipc.opt->srr)
1389 daddr = replyopts.opt.faddr;
1390 }
1391
1392 {
1393 struct flowi fl = { .oif = arg->bound_dev_if,
1394 .nl_u = { .ip4_u =
1395 { .daddr = daddr,
1396 .saddr = rt->rt_spec_dst,
1397 .tos = RT_TOS(ip_hdr(skb)->tos) } },
1398 /* Not quite clean, but right. */
1399 .uli_u = { .ports =
1400 { .sport = tcp_hdr(skb)->dest,
1401 .dport = tcp_hdr(skb)->source } },
1402 .proto = sk->sk_protocol,
1403 .flags = ip_reply_arg_flowi_flags(arg) };
1404 security_skb_classify_flow(skb, &fl);
1405 if (ip_route_output_key(sock_net(sk), &rt, &fl))
1406 return;
1407 }
1408
1409 /* And let IP do all the hard work.
1410
1411 This chunk is not reenterable, hence spinlock.
1412 Note that it uses the fact, that this function is called
1413 with locally disabled BH and that sk cannot be already spinlocked.
1414 */
1415 bh_lock_sock(sk);
1416 inet->tos = ip_hdr(skb)->tos;
1417 sk->sk_priority = skb->priority;
1418 sk->sk_protocol = ip_hdr(skb)->protocol;
1419 sk->sk_bound_dev_if = arg->bound_dev_if;
1420 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1421 &ipc, &rt, MSG_DONTWAIT);
1422 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1423 if (arg->csumoffset >= 0)
1424 *((__sum16 *)skb_transport_header(skb) +
1425 arg->csumoffset) = csum_fold(csum_add(skb->csum,
1426 arg->csum));
1427 skb->ip_summed = CHECKSUM_NONE;
1428 ip_push_pending_frames(sk);
1429 }
1430
1431 bh_unlock_sock(sk);
1432
1433 ip_rt_put(rt);
1434 }
1435
1436 void __init ip_init(void)
1437 {
1438 ip_rt_init();
1439 inet_initpeers();
1440
1441 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1442 igmp_mc_proc_init();
1443 #endif
1444 }
1445
1446 EXPORT_SYMBOL(ip_generic_getfrag);
1447 EXPORT_SYMBOL(ip_queue_xmit);
1448 EXPORT_SYMBOL(ip_send_check);
Linux with added FPC-III support