?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
NETFILTER: remove unnecessary goto statement for error recovery
[tomato.git] / release / src-rt / linux / linux-2.6 / net / ipv4 / ip_output.c
blob47464b76730ed3f357e511aa814d132d1d4b27d6
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 * Version: $Id: ip_output.c,v 1.100 2002/02/01 22:01:03 davem Exp $
9 *
10 * Authors: Ross Biro
11 * 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/mm.h>
53 #include <linux/string.h>
54 #include <linux/errno.h>
55 #include <linux/highmem.h>
56
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>
71 #include <net/xfrm.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <net/arp.h>
75 #include <net/icmp.h>
76 #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>
84 #include <linux/tcp.h>
85
86 int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
87
88 /* Generate a checksum for an outgoing IP datagram. */
89 __inline__ void ip_send_check(struct iphdr *iph)
90 {
91 iph->check = 0;
92 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
93 }
94
95 int __ip_local_out(struct sk_buff *skb)
96 {
97 struct iphdr *iph = ip_hdr(skb);
98
99 iph->tot_len = htons(skb->len);
100 ip_send_check(iph);
101 return nf_hook(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, skb->dst->dev,
102 dst_output);
103 }
104
105 int ip_local_out(struct sk_buff *skb)
106 {
107 int err;
108
109 err = __ip_local_out(skb);
110 if (likely(err == 1))
111 err = dst_output(skb);
112
113 return err;
114 }
115 EXPORT_SYMBOL_GPL(ip_local_out);
116
117 /* dev_loopback_xmit for use with netfilter. */
118 static int ip_dev_loopback_xmit(struct sk_buff *newskb)
119 {
120 skb_reset_mac_header(newskb);
121 __skb_pull(newskb, skb_network_offset(newskb));
122 newskb->pkt_type = PACKET_LOOPBACK;
123 newskb->ip_summed = CHECKSUM_UNNECESSARY;
124 BUG_TRAP(newskb->dst);
125 netif_rx(newskb);
126 return 0;
127 }
128
129 static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
130 {
131 int ttl = inet->uc_ttl;
132
133 if (ttl < 0)
134 ttl = dst_metric(dst, RTAX_HOPLIMIT);
135 return ttl;
136 }
137
138 /*
139 * Add an ip header to a skbuff and send it out.
140 *
141 */
142 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
143 __be32 saddr, __be32 daddr, struct ip_options *opt)
144 {
145 struct inet_sock *inet = inet_sk(sk);
146 struct rtable *rt = (struct rtable *)skb->dst;
147 struct iphdr *iph;
148
149 /* Build the IP header. */
150 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
151 skb_reset_network_header(skb);
152 iph = ip_hdr(skb);
153 iph->version = 4;
154 iph->ihl = 5;
155 iph->tos = inet->tos;
156 if (ip_dont_fragment(sk, &rt->u.dst))
157 iph->frag_off = htons(IP_DF);
158 else
159 iph->frag_off = 0;
160 iph->ttl = ip_select_ttl(inet, &rt->u.dst);
161 iph->daddr = rt->rt_dst;
162 iph->saddr = rt->rt_src;
163 iph->protocol = sk->sk_protocol;
164 ip_select_ident(iph, &rt->u.dst, sk);
165
166 if (opt && opt->optlen) {
167 iph->ihl += opt->optlen>>2;
168 ip_options_build(skb, opt, daddr, rt, 0);
169 }
170
171 skb->priority = sk->sk_priority;
172
173 /* Send it out. */
174 return ip_local_out(skb);
175 }
176
177 EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
178
179 static inline int ip_finish_output2(struct sk_buff *skb)
180 {
181 struct dst_entry *dst = skb->dst;
182 struct rtable *rt = (struct rtable *)dst;
183 struct net_device *dev = dst->dev;
184 int hh_len = LL_RESERVED_SPACE(dev);
185
186 if (rt->rt_type == RTN_MULTICAST)
187 IP_INC_STATS(IPSTATS_MIB_OUTMCASTPKTS);
188 else if (rt->rt_type == RTN_BROADCAST)
189 IP_INC_STATS(IPSTATS_MIB_OUTBCASTPKTS);
190
191 /* Be paranoid, rather than too clever. */
192 if (unlikely(skb_headroom(skb) < hh_len && dev->hard_header)) {
193 struct sk_buff *skb2;
194
195 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
196 if (skb2 == NULL) {
197 kfree_skb(skb);
198 return -ENOMEM;
199 }
200 if (skb->sk)
201 skb_set_owner_w(skb2, skb->sk);
202 kfree_skb(skb);
203 skb = skb2;
204 }
205
206 if (dst->hh)
207 return neigh_hh_output(dst->hh, skb);
208 else if (dst->neighbour)
209 return dst->neighbour->output(skb);
210
211 if (net_ratelimit())
212 printk(KERN_DEBUG "ip_finish_output2: No header cache and no neighbour!\n");
213 kfree_skb(skb);
214 return -EINVAL;
215 }
216
217 static inline int ip_skb_dst_mtu(struct sk_buff *skb)
218 {
219 struct inet_sock *inet = skb->sk ? inet_sk(skb->sk) : NULL;
220
221 return (inet && inet->pmtudisc == IP_PMTUDISC_PROBE) ?
222 skb->dst->dev->mtu : dst_mtu(skb->dst);
223 }
224
225 static inline int ip_finish_output(struct sk_buff *skb)
226 {
227 #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
228 /* Policy lookup after SNAT yielded a new policy */
229 if (skb->dst->xfrm != NULL) {
230 IPCB(skb)->flags |= IPSKB_REROUTED;
231 return dst_output(skb);
232 }
233 #endif
234 if (skb->len > ip_skb_dst_mtu(skb) && !skb_is_gso(skb))
235 return ip_fragment(skb, ip_finish_output2);
236 else
237 return ip_finish_output2(skb);
238 }
239
240 int ip_mc_output(struct sk_buff *skb)
241 {
242 struct sock *sk = skb->sk;
243 struct rtable *rt = (struct rtable*)skb->dst;
244 struct net_device *dev = rt->u.dst.dev;
245
246 /*
247 * If the indicated interface is up and running, send the packet.
248 */
249 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
250
251 skb->dev = dev;
252 skb->protocol = htons(ETH_P_IP);
253
254 /*
255 * Multicasts are looped back for other local users
256 */
257
258 if (rt->rt_flags&RTCF_MULTICAST) {
259 if ((!sk || inet_sk(sk)->mc_loop)
260 #ifdef CONFIG_IP_MROUTE
261 /* Small optimization: do not loopback not local frames,
262 which returned after forwarding; they will be dropped
263 by ip_mr_input in any case.
264 Note, that local frames are looped back to be delivered
265 to local recipients.
266
267 This check is duplicated in ip_mr_input at the moment.
268 */
269 && ((rt->rt_flags&RTCF_LOCAL) || !(IPCB(skb)->flags&IPSKB_FORWARDED))
270 #endif
271 ) {
272 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
273 if (newskb)
274 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
275 newskb->dev,
276 ip_dev_loopback_xmit);
277 }
278
279 /* Multicasts with ttl 0 must not go beyond the host */
280
281 if (ip_hdr(skb)->ttl == 0) {
282 kfree_skb(skb);
283 return 0;
284 }
285 }
286
287 if (rt->rt_flags&RTCF_BROADCAST) {
288 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
289 if (newskb)
290 NF_HOOK(PF_INET, NF_IP_POST_ROUTING, newskb, NULL,
291 newskb->dev, ip_dev_loopback_xmit);
292 }
293
294 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
295 ip_finish_output,
296 !(IPCB(skb)->flags & IPSKB_REROUTED));
297 }
298
299 int ip_output(struct sk_buff *skb)
300 {
301 struct net_device *dev = skb->dst->dev;
302
303 IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
304
305 skb->dev = dev;
306 skb->protocol = htons(ETH_P_IP);
307
308 return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
309 ip_finish_output,
310 !(IPCB(skb)->flags & IPSKB_REROUTED));
311 }
312
313 int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
314 {
315 struct sock *sk = skb->sk;
316 struct inet_sock *inet = inet_sk(sk);
317 struct ip_options *opt = inet->opt;
318 struct rtable *rt;
319 struct iphdr *iph;
320
321 /* Skip all of this if the packet is already routed,
322 * f.e. by something like SCTP.
323 */
324 rt = (struct rtable *) skb->dst;
325 if (rt != NULL)
326 goto packet_routed;
327
328 /* Make sure we can route this packet. */
329 rt = (struct rtable *)__sk_dst_check(sk, 0);
330 if (rt == NULL) {
331 __be32 daddr;
332
333 /* Use correct destination address if we have options. */
334 daddr = inet->daddr;
335 if(opt && opt->srr)
336 daddr = opt->faddr;
337
338 {
339 struct flowi fl = { .oif = sk->sk_bound_dev_if,
340 .nl_u = { .ip4_u =
341 { .daddr = daddr,
342 .saddr = inet->saddr,
343 .tos = RT_CONN_FLAGS(sk) } },
344 .proto = sk->sk_protocol,
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(&rt, &fl, sk, 0))
355 goto no_route;
356 }
357 sk_setup_caps(sk, &rt->u.dst);
358 }
359 skb->dst = 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
390 return ip_local_out(skb);
391
392 no_route:
393 IP_INC_STATS(IPSTATS_MIB_OUTNOROUTES);
394 kfree_skb(skb);
395 return -EHOSTUNREACH;
396 }
397
398
399 static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
400 {
401 to->pkt_type = from->pkt_type;
402 to->priority = from->priority;
403 to->protocol = from->protocol;
404 dst_release(to->dst);
405 to->dst = dst_clone(from->dst);
406 to->dev = from->dev;
407 to->mark = from->mark;
408
409 /* Copy the flags to each fragment. */
410 IPCB(to)->flags = IPCB(from)->flags;
411
412 #ifdef CONFIG_NET_SCHED
413 to->tc_index = from->tc_index;
414 #endif
415 nf_copy(to, from);
416 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
417 to->ipvs_property = from->ipvs_property;
418 #endif
419 skb_copy_secmark(to, from);
420 }
421
422 /*
423 * This IP datagram is too large to be sent in one piece. Break it up into
424 * smaller pieces (each of size equal to IP header plus
425 * a block of the data of the original IP data part) that will yet fit in a
426 * single device frame, and queue such a frame for sending.
427 */
428
429 int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
430 {
431 struct iphdr *iph;
432 int raw = 0;
433 int ptr;
434 struct net_device *dev;
435 struct sk_buff *skb2;
436 unsigned int mtu, hlen, left, len, ll_rs, pad;
437 int offset;
438 __be16 not_last_frag;
439 struct rtable *rt = (struct rtable*)skb->dst;
440 int err = 0;
441
442 dev = rt->u.dst.dev;
443
444 /*
445 * Point into the IP datagram header.
446 */
447
448 iph = ip_hdr(skb);
449
450 if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
451 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
452 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
453 htonl(ip_skb_dst_mtu(skb)));
454 kfree_skb(skb);
455 return -EMSGSIZE;
456 }
457
458 /*
459 * Setup starting values.
460 */
461
462 hlen = iph->ihl * 4;
463 mtu = dst_mtu(&rt->u.dst) - hlen; /* Size of data space */
464 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
465
466 /* When frag_list is given, use it. First, check its validity:
467 * some transformers could create wrong frag_list or break existing
468 * one, it is not prohibited. In this case fall back to copying.
469 *
470 * LATER: this step can be merged to real generation of fragments,
471 * we can switch to copy when see the first bad fragment.
472 */
473 if (skb_shinfo(skb)->frag_list) {
474 struct sk_buff *frag, *frag2;
475 int first_len = skb_pagelen(skb);
476
477 if (first_len - hlen > mtu ||
478 ((first_len - hlen) & 7) ||
479 (iph->frag_off & htons(IP_MF|IP_OFFSET)) ||
480 skb_cloned(skb))
481 goto slow_path;
482
483 for (frag = skb_shinfo(skb)->frag_list; frag; frag = frag->next) {
484 /* Correct geometry. */
485 if (frag->len > mtu ||
486 ((frag->len & 7) && frag->next) ||
487 skb_headroom(frag) < hlen)
488 goto slow_path_clean;
489
490 /* Partially cloned skb? */
491 if (skb_shared(frag))
492 goto slow_path_clean;
493
494 BUG_ON(frag->sk);
495 if (skb->sk) {
496 frag->sk = skb->sk;
497 frag->destructor = sock_wfree;
498 }
499 skb->truesize -= frag->truesize;
500 }
501
502 /* Everything is OK. Generate! */
503
504 err = 0;
505 offset = 0;
506 frag = skb_shinfo(skb)->frag_list;
507 skb_shinfo(skb)->frag_list = NULL;
508 skb->data_len = first_len - skb_headlen(skb);
509 skb->len = first_len;
510 iph->tot_len = htons(first_len);
511 iph->frag_off = htons(IP_MF);
512 ip_send_check(iph);
513
514 for (;;) {
515 /* Prepare header of the next frame,
516 * before previous one went down. */
517 if (frag) {
518 frag->ip_summed = CHECKSUM_NONE;
519 skb_reset_transport_header(frag);
520 __skb_push(frag, hlen);
521 skb_reset_network_header(frag);
522 memcpy(skb_network_header(frag), iph, hlen);
523 iph = ip_hdr(frag);
524 iph->tot_len = htons(frag->len);
525 ip_copy_metadata(frag, skb);
526 if (offset == 0)
527 ip_options_fragment(frag);
528 offset += skb->len - hlen;
529 iph->frag_off = htons(offset>>3);
530 if (frag->next != NULL)
531 iph->frag_off |= htons(IP_MF);
532 /* Ready, complete checksum */
533 ip_send_check(iph);
534 }
535
536 err = output(skb);
537
538 if (!err)
539 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
540 if (err || !frag)
541 break;
542
543 skb = frag;
544 frag = skb->next;
545 skb->next = NULL;
546 }
547
548 if (err == 0) {
549 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
550 return 0;
551 }
552
553 while (frag) {
554 skb = frag->next;
555 kfree_skb(frag);
556 frag = skb;
557 }
558 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
559 return err;
560
561 slow_path_clean:
562 for (frag2 = skb_shinfo(skb)->frag_list; frag2; frag2 = frag2->next) {
563 if (frag2 == frag)
564 break;
565 frag2->sk = NULL;
566 frag2->destructor = NULL;
567 skb->truesize += frag2->truesize;
568 }
569 }
570
571 slow_path:
572 left = skb->len - hlen; /* Space per frame */
573 ptr = raw + hlen; /* Where to start from */
574
575 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
576 * we need to make room for the encapsulating header
577 */
578 pad = nf_bridge_pad(skb);
579 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
580 mtu -= pad;
581
582 /*
583 * Fragment the datagram.
584 */
585
586 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
587 not_last_frag = iph->frag_off & htons(IP_MF);
588
589 /*
590 * Keep copying data until we run out.
591 */
592
593 while (left > 0) {
594 len = left;
595 /* IF: it doesn't fit, use 'mtu' - the data space left */
596 if (len > mtu)
597 len = mtu;
598 /* IF: we are not sending upto and including the packet end
599 then align the next start on an eight byte boundary */
600 if (len < left) {
601 len &= ~7;
602 }
603 /*
604 * Allocate buffer.
605 */
606
607 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
608 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
609 err = -ENOMEM;
610 goto fail;
611 }
612
613 /*
614 * Set up data on packet
615 */
616
617 ip_copy_metadata(skb2, skb);
618 skb_reserve(skb2, ll_rs);
619 skb_put(skb2, len + hlen);
620 skb_reset_network_header(skb2);
621 skb2->transport_header = skb2->network_header + hlen;
622
623 /*
624 * Charge the memory for the fragment to any owner
625 * it might possess
626 */
627
628 if (skb->sk)
629 skb_set_owner_w(skb2, skb->sk);
630
631 /*
632 * Copy the packet header into the new buffer.
633 */
634
635 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
636
637 /*
638 * Copy a block of the IP datagram.
639 */
640 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
641 BUG();
642 left -= len;
643
644 /*
645 * Fill in the new header fields.
646 */
647 iph = ip_hdr(skb2);
648 iph->frag_off = htons((offset >> 3));
649
650 /* ANK: dirty, but effective trick. Upgrade options only if
651 * the segment to be fragmented was THE FIRST (otherwise,
652 * options are already fixed) and make it ONCE
653 * on the initial skb, so that all the following fragments
654 * will inherit fixed options.
655 */
656 if (offset == 0)
657 ip_options_fragment(skb);
658
659 /*
660 * Added AC : If we are fragmenting a fragment that's not the
661 * last fragment then keep MF on each bit
662 */
663 if (left > 0 || not_last_frag)
664 iph->frag_off |= htons(IP_MF);
665 ptr += len;
666 offset += len;
667
668 /*
669 * Put this fragment into the sending queue.
670 */
671 iph->tot_len = htons(len + hlen);
672
673 ip_send_check(iph);
674
675 err = output(skb2);
676 if (err)
677 goto fail;
678
679 IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
680 }
681 kfree_skb(skb);
682 IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
683 return err;
684
685 fail:
686 kfree_skb(skb);
687 IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
688 return err;
689 }
690
691 EXPORT_SYMBOL(ip_fragment);
692
693 int
694 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
695 {
696 struct iovec *iov = from;
697
698 if (skb->ip_summed == CHECKSUM_PARTIAL) {
699 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
700 return -EFAULT;
701 } else {
702 __wsum csum = 0;
703 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
704 return -EFAULT;
705 skb->csum = csum_block_add(skb->csum, csum, odd);
706 }
707 return 0;
708 }
709
710 static inline __wsum
711 csum_page(struct page *page, int offset, int copy)
712 {
713 char *kaddr;
714 __wsum csum;
715 kaddr = kmap(page);
716 csum = csum_partial(kaddr + offset, copy, 0);
717 kunmap(page);
718 return csum;
719 }
720
721 static inline int ip_ufo_append_data(struct sock *sk,
722 int getfrag(void *from, char *to, int offset, int len,
723 int odd, struct sk_buff *skb),
724 void *from, int length, int hh_len, int fragheaderlen,
725 int transhdrlen, int mtu,unsigned int flags)
726 {
727 struct sk_buff *skb;
728 int err;
729
730 /* There is support for UDP fragmentation offload by network
731 * device, so create one single skb packet containing complete
732 * udp datagram
733 */
734 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
735 skb = sock_alloc_send_skb(sk,
736 hh_len + fragheaderlen + transhdrlen + 20,
737 (flags & MSG_DONTWAIT), &err);
738
739 if (skb == NULL)
740 return err;
741
742 /* reserve space for Hardware header */
743 skb_reserve(skb, hh_len);
744
745 /* create space for UDP/IP header */
746 skb_put(skb,fragheaderlen + transhdrlen);
747
748 /* initialize network header pointer */
749 skb_reset_network_header(skb);
750
751 /* initialize protocol header pointer */
752 skb->transport_header = skb->network_header + fragheaderlen;
753
754 skb->ip_summed = CHECKSUM_PARTIAL;
755 skb->csum = 0;
756 sk->sk_sndmsg_off = 0;
757
758 /* specify the length of each IP datagram fragment */
759 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
760 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
761 __skb_queue_tail(&sk->sk_write_queue, skb);
762 }
763
764 return skb_append_datato_frags(sk, skb, getfrag, from,
765 (length - transhdrlen));
766 }
767
768 /*
769 * ip_append_data() and ip_append_page() can make one large IP datagram
770 * from many pieces of data. Each pieces will be holded on the socket
771 * until ip_push_pending_frames() is called. Each piece can be a page
772 * or non-page data.
773 *
774 * Not only UDP, other transport protocols - e.g. raw sockets - can use
775 * this interface potentially.
776 *
777 * LATER: length must be adjusted by pad at tail, when it is required.
778 */
779 int ip_append_data(struct sock *sk,
780 int getfrag(void *from, char *to, int offset, int len,
781 int odd, struct sk_buff *skb),
782 void *from, int length, int transhdrlen,
783 struct ipcm_cookie *ipc, struct rtable *rt,
784 unsigned int flags)
785 {
786 struct inet_sock *inet = inet_sk(sk);
787 struct sk_buff *skb;
788
789 struct ip_options *opt = NULL;
790 int hh_len;
791 int exthdrlen;
792 int mtu;
793 int copy;
794 int err;
795 int offset = 0;
796 unsigned int maxfraglen, fragheaderlen;
797 int csummode = CHECKSUM_NONE;
798
799 if (flags&MSG_PROBE)
800 return 0;
801
802 if (skb_queue_empty(&sk->sk_write_queue)) {
803 /*
804 * setup for corking.
805 */
806 opt = ipc->opt;
807 if (opt) {
808 if (inet->cork.opt == NULL) {
809 inet->cork.opt = kmalloc(sizeof(struct ip_options) + 40, sk->sk_allocation);
810 if (unlikely(inet->cork.opt == NULL))
811 return -ENOBUFS;
812 }
813 memcpy(inet->cork.opt, opt, sizeof(struct ip_options)+opt->optlen);
814 inet->cork.flags |= IPCORK_OPT;
815 inet->cork.addr = ipc->addr;
816 }
817 dst_hold(&rt->u.dst);
818 inet->cork.fragsize = mtu = inet->pmtudisc == IP_PMTUDISC_PROBE ?
819 rt->u.dst.dev->mtu :
820 dst_mtu(rt->u.dst.path);
821 inet->cork.rt = rt;
822 inet->cork.length = 0;
823 sk->sk_sndmsg_page = NULL;
824 sk->sk_sndmsg_off = 0;
825 if ((exthdrlen = rt->u.dst.header_len) != 0) {
826 length += exthdrlen;
827 transhdrlen += exthdrlen;
828 }
829 } else {
830 rt = inet->cork.rt;
831 if (inet->cork.flags & IPCORK_OPT)
832 opt = inet->cork.opt;
833
834 transhdrlen = 0;
835 exthdrlen = 0;
836 mtu = inet->cork.fragsize;
837 }
838 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
839
840 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
841 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
842
843 if (inet->cork.length + length > 0xFFFF - fragheaderlen) {
844 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu-exthdrlen);
845 return -EMSGSIZE;
846 }
847
848 /*
849 * transhdrlen > 0 means that this is the first fragment and we wish
850 * it won't be fragmented in the future.
851 */
852 if (transhdrlen &&
853 length + fragheaderlen <= mtu &&
854 rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
855 !exthdrlen)
856 csummode = CHECKSUM_PARTIAL;
857
858 inet->cork.length += length;
859 if (((length> mtu) || !skb_queue_empty(&sk->sk_write_queue)) &&
860 (sk->sk_protocol == IPPROTO_UDP) &&
861 (rt->u.dst.dev->features & NETIF_F_UFO)) {
862 err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
863 fragheaderlen, transhdrlen, mtu,
864 flags);
865 if (err)
866 goto error;
867 return 0;
868 }
869
870 /* So, what's going on in the loop below?
871 *
872 * We use calculated fragment length to generate chained skb,
873 * each of segments is IP fragment ready for sending to network after
874 * adding appropriate IP header.
875 */
876
877 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
878 goto alloc_new_skb;
879
880 while (length > 0) {
881 /* Check if the remaining data fits into current packet. */
882 copy = mtu - skb->len;
883 if (copy < length)
884 copy = maxfraglen - skb->len;
885 if (copy <= 0) {
886 char *data;
887 unsigned int datalen;
888 unsigned int fraglen;
889 unsigned int fraggap;
890 unsigned int alloclen;
891 struct sk_buff *skb_prev;
892 alloc_new_skb:
893 skb_prev = skb;
894 if (skb_prev)
895 fraggap = skb_prev->len - maxfraglen;
896 else
897 fraggap = 0;
898
899 /*
900 * If remaining data exceeds the mtu,
901 * we know we need more fragment(s).
902 */
903 datalen = length + fraggap;
904 if (datalen > mtu - fragheaderlen)
905 datalen = maxfraglen - fragheaderlen;
906 fraglen = datalen + fragheaderlen;
907
908 if ((flags & MSG_MORE) &&
909 !(rt->u.dst.dev->features&NETIF_F_SG))
910 alloclen = mtu;
911 else
912 alloclen = datalen + fragheaderlen;
913
914 /* The last fragment gets additional space at tail.
915 * Note, with MSG_MORE we overallocate on fragments,
916 * because we have no idea what fragment will be
917 * the last.
918 */
919 if (datalen == length + fraggap)
920 alloclen += rt->u.dst.trailer_len;
921
922 if (transhdrlen) {
923 skb = sock_alloc_send_skb(sk,
924 alloclen + hh_len + 15,
925 (flags & MSG_DONTWAIT), &err);
926 } else {
927 skb = NULL;
928 if (atomic_read(&sk->sk_wmem_alloc) <=
929 2 * sk->sk_sndbuf)
930 skb = sock_wmalloc(sk,
931 alloclen + hh_len + 15, 1,
932 sk->sk_allocation);
933 if (unlikely(skb == NULL))
934 err = -ENOBUFS;
935 }
936 if (skb == NULL)
937 goto error;
938
939 /*
940 * Fill in the control structures
941 */
942 skb->ip_summed = csummode;
943 skb->csum = 0;
944 skb_reserve(skb, hh_len);
945
946 /*
947 * Find where to start putting bytes.
948 */
949 data = skb_put(skb, fraglen);
950 skb_set_network_header(skb, exthdrlen);
951 skb->transport_header = (skb->network_header +
952 fragheaderlen);
953 data += fragheaderlen;
954
955 if (fraggap) {
956 skb->csum = skb_copy_and_csum_bits(
957 skb_prev, maxfraglen,
958 data + transhdrlen, fraggap, 0);
959 skb_prev->csum = csum_sub(skb_prev->csum,
960 skb->csum);
961 data += fraggap;
962 pskb_trim_unique(skb_prev, maxfraglen);
963 }
964
965 copy = datalen - transhdrlen - fraggap;
966 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
967 err = -EFAULT;
968 kfree_skb(skb);
969 goto error;
970 }
971
972 offset += copy;
973 length -= datalen - fraggap;
974 transhdrlen = 0;
975 exthdrlen = 0;
976 csummode = CHECKSUM_NONE;
977
978 /*
979 * Put the packet on the pending queue.
980 */
981 __skb_queue_tail(&sk->sk_write_queue, skb);
982 continue;
983 }
984
985 if (copy > length)
986 copy = length;
987
988 if (!(rt->u.dst.dev->features&NETIF_F_SG)) {
989 unsigned int off;
990
991 off = skb->len;
992 if (getfrag(from, skb_put(skb, copy),
993 offset, copy, off, skb) < 0) {
994 __skb_trim(skb, off);
995 err = -EFAULT;
996 goto error;
997 }
998 } else {
999 int i = skb_shinfo(skb)->nr_frags;
1000 skb_frag_t *frag = &skb_shinfo(skb)->frags[i-1];
1001 struct page *page = sk->sk_sndmsg_page;
1002 int off = sk->sk_sndmsg_off;
1003 unsigned int left;
1004
1005 if (page && (left = PAGE_SIZE - off) > 0) {
1006 if (copy >= left)
1007 copy = left;
1008 if (page != frag->page) {
1009 if (i == MAX_SKB_FRAGS) {
1010 err = -EMSGSIZE;
1011 goto error;
1012 }
1013 get_page(page);
1014 skb_fill_page_desc(skb, i, page, sk->sk_sndmsg_off, 0);
1015 frag = &skb_shinfo(skb)->frags[i];
1016 }
1017 } else if (i < MAX_SKB_FRAGS) {
1018 if (copy > PAGE_SIZE)
1019 copy = PAGE_SIZE;
1020 page = alloc_pages(sk->sk_allocation, 0);
1021 if (page == NULL) {
1022 err = -ENOMEM;
1023 goto error;
1024 }
1025 sk->sk_sndmsg_page = page;
1026 sk->sk_sndmsg_off = 0;
1027
1028 skb_fill_page_desc(skb, i, page, 0, 0);
1029 frag = &skb_shinfo(skb)->frags[i];
1030 } else {
1031 err = -EMSGSIZE;
1032 goto error;
1033 }
1034 if (getfrag(from, page_address(frag->page)+frag->page_offset+frag->size, offset, copy, skb->len, skb) < 0) {
1035 err = -EFAULT;
1036 goto error;
1037 }
1038 sk->sk_sndmsg_off += copy;
1039 frag->size += copy;
1040 skb->len += copy;
1041 skb->data_len += copy;
1042 skb->truesize += copy;
1043 atomic_add(copy, &sk->sk_wmem_alloc);
1044 }
1045 offset += copy;
1046 length -= copy;
1047 }
1048
1049 return 0;
1050
1051 error:
1052 inet->cork.length -= length;
1053 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1054 return err;
1055 }
1056
1057 ssize_t ip_append_page(struct sock *sk, struct page *page,
1058 int offset, size_t size, int flags)
1059 {
1060 struct inet_sock *inet = inet_sk(sk);
1061 struct sk_buff *skb;
1062 struct rtable *rt;
1063 struct ip_options *opt = NULL;
1064 int hh_len;
1065 int mtu;
1066 int len;
1067 int err;
1068 unsigned int maxfraglen, fragheaderlen, fraggap;
1069
1070 if (inet->hdrincl)
1071 return -EPERM;
1072
1073 if (flags&MSG_PROBE)
1074 return 0;
1075
1076 if (skb_queue_empty(&sk->sk_write_queue))
1077 return -EINVAL;
1078
1079 rt = inet->cork.rt;
1080 if (inet->cork.flags & IPCORK_OPT)
1081 opt = inet->cork.opt;
1082
1083 if (!(rt->u.dst.dev->features&NETIF_F_SG))
1084 return -EOPNOTSUPP;
1085
1086 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
1087 mtu = inet->cork.fragsize;
1088
1089 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1090 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1091
1092 if (inet->cork.length + size > 0xFFFF - fragheaderlen) {
1093 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport, mtu);
1094 return -EMSGSIZE;
1095 }
1096
1097 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1098 return -EINVAL;
1099
1100 inet->cork.length += size;
1101 if ((sk->sk_protocol == IPPROTO_UDP) &&
1102 (rt->u.dst.dev->features & NETIF_F_UFO)) {
1103 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1104 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1105 }
1106
1107
1108 while (size > 0) {
1109 int i;
1110
1111 if (skb_is_gso(skb))
1112 len = size;
1113 else {
1114
1115 /* Check if the remaining data fits into current packet. */
1116 len = mtu - skb->len;
1117 if (len < size)
1118 len = maxfraglen - skb->len;
1119 }
1120 if (len <= 0) {
1121 struct sk_buff *skb_prev;
1122 int alloclen;
1123
1124 skb_prev = skb;
1125 fraggap = skb_prev->len - maxfraglen;
1126
1127 alloclen = fragheaderlen + hh_len + fraggap + 15;
1128 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1129 if (unlikely(!skb)) {
1130 err = -ENOBUFS;
1131 goto error;
1132 }
1133
1134 /*
1135 * Fill in the control structures
1136 */
1137 skb->ip_summed = CHECKSUM_NONE;
1138 skb->csum = 0;
1139 skb_reserve(skb, hh_len);
1140
1141 /*
1142 * Find where to start putting bytes.
1143 */
1144 skb_put(skb, fragheaderlen + fraggap);
1145 skb_reset_network_header(skb);
1146 skb->transport_header = (skb->network_header +
1147 fragheaderlen);
1148 if (fraggap) {
1149 skb->csum = skb_copy_and_csum_bits(skb_prev,
1150 maxfraglen,
1151 skb_transport_header(skb),
1152 fraggap, 0);
1153 skb_prev->csum = csum_sub(skb_prev->csum,
1154 skb->csum);
1155 pskb_trim_unique(skb_prev, maxfraglen);
1156 }
1157
1158 /*
1159 * Put the packet on the pending queue.
1160 */
1161 __skb_queue_tail(&sk->sk_write_queue, skb);
1162 continue;
1163 }
1164
1165 i = skb_shinfo(skb)->nr_frags;
1166 if (len > size)
1167 len = size;
1168 if (skb_can_coalesce(skb, i, page, offset)) {
1169 skb_shinfo(skb)->frags[i-1].size += len;
1170 } else if (i < MAX_SKB_FRAGS) {
1171 get_page(page);
1172 skb_fill_page_desc(skb, i, page, offset, len);
1173 } else {
1174 err = -EMSGSIZE;
1175 goto error;
1176 }
1177
1178 if (skb->ip_summed == CHECKSUM_NONE) {
1179 __wsum csum;
1180 csum = csum_page(page, offset, len);
1181 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1182 }
1183
1184 skb->len += len;
1185 skb->data_len += len;
1186 skb->truesize += len;
1187 atomic_add(len, &sk->sk_wmem_alloc);
1188 offset += len;
1189 size -= len;
1190 }
1191 return 0;
1192
1193 error:
1194 inet->cork.length -= size;
1195 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1196 return err;
1197 }
1198
1199 /*
1200 * Combined all pending IP fragments on the socket as one IP datagram
1201 * and push them out.
1202 */
1203 int ip_push_pending_frames(struct sock *sk)
1204 {
1205 struct sk_buff *skb, *tmp_skb;
1206 struct sk_buff **tail_skb;
1207 struct inet_sock *inet = inet_sk(sk);
1208 struct ip_options *opt = NULL;
1209 struct rtable *rt = inet->cork.rt;
1210 struct iphdr *iph;
1211 __be16 df = 0;
1212 __u8 ttl;
1213 int err = 0;
1214
1215 if ((skb = __skb_dequeue(&sk->sk_write_queue)) == NULL)
1216 goto out;
1217 tail_skb = &(skb_shinfo(skb)->frag_list);
1218
1219 /* move skb->data to ip header from ext header */
1220 if (skb->data < skb_network_header(skb))
1221 __skb_pull(skb, skb_network_offset(skb));
1222 while ((tmp_skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) {
1223 __skb_pull(tmp_skb, skb_network_header_len(skb));
1224 *tail_skb = tmp_skb;
1225 tail_skb = &(tmp_skb->next);
1226 skb->len += tmp_skb->len;
1227 skb->data_len += tmp_skb->len;
1228 skb->truesize += tmp_skb->truesize;
1229 tmp_skb->destructor = NULL;
1230 tmp_skb->sk = NULL;
1231 }
1232
1233 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1234 * to fragment the frame generated here. No matter, what transforms
1235 * how transforms change size of the packet, it will come out.
1236 */
1237 if (inet->pmtudisc < IP_PMTUDISC_DO)
1238 skb->local_df = 1;
1239
1240 /* DF bit is set when we want to see DF on outgoing frames.
1241 * If local_df is set too, we still allow to fragment this frame
1242 * locally. */
1243 if (inet->pmtudisc >= IP_PMTUDISC_DO ||
1244 (skb->len <= dst_mtu(&rt->u.dst) &&
1245 ip_dont_fragment(sk, &rt->u.dst)))
1246 df = htons(IP_DF);
1247
1248 if (inet->cork.flags & IPCORK_OPT)
1249 opt = inet->cork.opt;
1250
1251 if (rt->rt_type == RTN_MULTICAST)
1252 ttl = inet->mc_ttl;
1253 else
1254 ttl = ip_select_ttl(inet, &rt->u.dst);
1255
1256 iph = (struct iphdr *)skb->data;
1257 iph->version = 4;
1258 iph->ihl = 5;
1259 if (opt) {
1260 iph->ihl += opt->optlen>>2;
1261 ip_options_build(skb, opt, inet->cork.addr, rt, 0);
1262 }
1263 iph->tos = inet->tos;
1264 iph->frag_off = df;
1265 ip_select_ident(iph, &rt->u.dst, sk);
1266 iph->ttl = ttl;
1267 iph->protocol = sk->sk_protocol;
1268 iph->saddr = rt->rt_src;
1269 iph->daddr = rt->rt_dst;
1270
1271 skb->priority = sk->sk_priority;
1272 skb->dst = dst_clone(&rt->u.dst);
1273
1274 /* Netfilter gets whole the not fragmented skb. */
1275 err = ip_local_out(skb);
1276 if (err) {
1277 if (err > 0)
1278 err = inet->recverr ? net_xmit_errno(err) : 0;
1279 if (err)
1280 goto error;
1281 }
1282
1283 out:
1284 inet->cork.flags &= ~IPCORK_OPT;
1285 kfree(inet->cork.opt);
1286 inet->cork.opt = NULL;
1287 if (inet->cork.rt) {
1288 ip_rt_put(inet->cork.rt);
1289 inet->cork.rt = NULL;
1290 }
1291 return err;
1292
1293 error:
1294 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
1295 goto out;
1296 }
1297
1298 /*
1299 * Throw away all pending data on the socket.
1300 */
1301 void ip_flush_pending_frames(struct sock *sk)
1302 {
1303 struct inet_sock *inet = inet_sk(sk);
1304 struct sk_buff *skb;
1305
1306 while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
1307 kfree_skb(skb);
1308
1309 inet->cork.flags &= ~IPCORK_OPT;
1310 kfree(inet->cork.opt);
1311 inet->cork.opt = NULL;
1312 if (inet->cork.rt) {
1313 ip_rt_put(inet->cork.rt);
1314 inet->cork.rt = NULL;
1315 }
1316 }
1317
1318
1319 /*
1320 * Fetch data from kernel space and fill in checksum if needed.
1321 */
1322 static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1323 int len, int odd, struct sk_buff *skb)
1324 {
1325 __wsum csum;
1326
1327 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1328 skb->csum = csum_block_add(skb->csum, csum, odd);
1329 return 0;
1330 }
1331
1332 /*
1333 * Generic function to send a packet as reply to another packet.
1334 * Used to send TCP resets so far. ICMP should use this function too.
1335 *
1336 * Should run single threaded per socket because it uses the sock
1337 * structure to pass arguments.
1338 *
1339 * LATER: switch from ip_build_xmit to ip_append_*
1340 */
1341 void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *arg,
1342 unsigned int len)
1343 {
1344 struct inet_sock *inet = inet_sk(sk);
1345 struct {
1346 struct ip_options opt;
1347 char data[40];
1348 } replyopts;
1349 struct ipcm_cookie ipc;
1350 __be32 daddr;
1351 struct rtable *rt = (struct rtable*)skb->dst;
1352
1353 if (ip_options_echo(&replyopts.opt, skb))
1354 return;
1355
1356 daddr = ipc.addr = rt->rt_src;
1357 ipc.opt = NULL;
1358
1359 if (replyopts.opt.optlen) {
1360 ipc.opt = &replyopts.opt;
1361
1362 if (ipc.opt->srr)
1363 daddr = replyopts.opt.faddr;
1364 }
1365
1366 {
1367 struct flowi fl = { .oif = arg->bound_dev_if,
1368 .nl_u = { .ip4_u =
1369 { .daddr = daddr,
1370 .saddr = rt->rt_spec_dst,
1371 .tos = RT_TOS(ip_hdr(skb)->tos) } },
1372 /* Not quite clean, but right. */
1373 .uli_u = { .ports =
1374 { .sport = tcp_hdr(skb)->dest,
1375 .dport = tcp_hdr(skb)->source } },
1376 .proto = sk->sk_protocol };
1377 security_skb_classify_flow(skb, &fl);
1378 if (ip_route_output_key(&rt, &fl))
1379 return;
1380 }
1381
1382 /* And let IP do all the hard work.
1383
1384 This chunk is not reenterable, hence spinlock.
1385 Note that it uses the fact, that this function is called
1386 with locally disabled BH and that sk cannot be already spinlocked.
1387 */
1388 bh_lock_sock(sk);
1389 inet->tos = ip_hdr(skb)->tos;
1390 sk->sk_priority = skb->priority;
1391 sk->sk_protocol = ip_hdr(skb)->protocol;
1392 sk->sk_bound_dev_if = arg->bound_dev_if;
1393 ip_append_data(sk, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1394 &ipc, rt, MSG_DONTWAIT);
1395 if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
1396 if (arg->csumoffset >= 0)
1397 *((__sum16 *)skb_transport_header(skb) +
1398 arg->csumoffset) = csum_fold(csum_add(skb->csum,
1399 arg->csum));
1400 skb->ip_summed = CHECKSUM_NONE;
1401 ip_push_pending_frames(sk);
1402 }
1403
1404 bh_unlock_sock(sk);
1405
1406 ip_rt_put(rt);
1407 }
1408
1409 void __init ip_init(void)
1410 {
1411 ip_rt_init();
1412 inet_initpeers();
1413
1414 #if defined(CONFIG_IP_MULTICAST) && defined(CONFIG_PROC_FS)
1415 igmp_mc_proc_init();
1416 #endif
1417 }
1418
1419 EXPORT_SYMBOL(ip_generic_getfrag);
1420 EXPORT_SYMBOL(ip_queue_xmit);
1421 EXPORT_SYMBOL(ip_send_check);
Tomato firmware and modifications.