1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45#include <linux/uaccess.h>
46#include <linux/module.h>
47#include <linux/types.h>
48#include <linux/kernel.h>
49#include <linux/mm.h>
50#include <linux/string.h>
51#include <linux/errno.h>
52#include <linux/highmem.h>
53#include <linux/slab.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 <net/lwtunnel.h>
77#include <linux/bpf-cgroup.h>
78#include <linux/igmp.h>
79#include <linux/netfilter_ipv4.h>
80#include <linux/netfilter_bridge.h>
81#include <linux/netlink.h>
82#include <linux/tcp.h>
83
84static int
85ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
86 unsigned int mtu,
87 int (*output)(struct net *, struct sock *, struct sk_buff *));
88
89
90void ip_send_check(struct iphdr *iph)
91{
92 iph->check = 0;
93 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
94}
95EXPORT_SYMBOL(ip_send_check);
96
97int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
98{
99 struct iphdr *iph = ip_hdr(skb);
100
101 iph->tot_len = htons(skb->len);
102 ip_send_check(iph);
103
104
105
106
107 skb = l3mdev_ip_out(sk, skb);
108 if (unlikely(!skb))
109 return 0;
110
111 skb->protocol = htons(ETH_P_IP);
112
113 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
114 net, sk, skb, NULL, skb_dst(skb)->dev,
115 dst_output);
116}
117
118int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
119{
120 int err;
121
122 err = __ip_local_out(net, sk, skb);
123 if (likely(err == 1))
124 err = dst_output(net, sk, skb);
125
126 return err;
127}
128EXPORT_SYMBOL_GPL(ip_local_out);
129
130static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
131{
132 int ttl = inet->uc_ttl;
133
134 if (ttl < 0)
135 ttl = ip4_dst_hoplimit(dst);
136 return ttl;
137}
138
139
140
141
142
143int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
144 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
145{
146 struct inet_sock *inet = inet_sk(sk);
147 struct rtable *rt = skb_rtable(skb);
148 struct net *net = sock_net(sk);
149 struct iphdr *iph;
150
151
152 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
153 skb_reset_network_header(skb);
154 iph = ip_hdr(skb);
155 iph->version = 4;
156 iph->ihl = 5;
157 iph->tos = inet->tos;
158 iph->ttl = ip_select_ttl(inet, &rt->dst);
159 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
160 iph->saddr = saddr;
161 iph->protocol = sk->sk_protocol;
162 if (ip_dont_fragment(sk, &rt->dst)) {
163 iph->frag_off = htons(IP_DF);
164 iph->id = 0;
165 } else {
166 iph->frag_off = 0;
167 __ip_select_ident(net, iph, 1);
168 }
169
170 if (opt && opt->opt.optlen) {
171 iph->ihl += opt->opt.optlen>>2;
172 ip_options_build(skb, &opt->opt, daddr, rt, 0);
173 }
174
175 skb->priority = sk->sk_priority;
176 if (!skb->mark)
177 skb->mark = sk->sk_mark;
178
179
180 return ip_local_out(net, skb->sk, skb);
181}
182EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
183
184static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
185{
186 struct dst_entry *dst = skb_dst(skb);
187 struct rtable *rt = (struct rtable *)dst;
188 struct net_device *dev = dst->dev;
189 unsigned int hh_len = LL_RESERVED_SPACE(dev);
190 struct neighbour *neigh;
191 u32 nexthop;
192
193 if (rt->rt_type == RTN_MULTICAST) {
194 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
195 } else if (rt->rt_type == RTN_BROADCAST)
196 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
197
198
199 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
200 struct sk_buff *skb2;
201
202 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
203 if (!skb2) {
204 kfree_skb(skb);
205 return -ENOMEM;
206 }
207 if (skb->sk)
208 skb_set_owner_w(skb2, skb->sk);
209 consume_skb(skb);
210 skb = skb2;
211 }
212
213 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
214 int res = lwtunnel_xmit(skb);
215
216 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
217 return res;
218 }
219
220 rcu_read_lock_bh();
221 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
222 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
223 if (unlikely(!neigh))
224 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
225 if (!IS_ERR(neigh)) {
226 int res;
227
228 sock_confirm_neigh(skb, neigh);
229 res = neigh_output(neigh, skb);
230
231 rcu_read_unlock_bh();
232 return res;
233 }
234 rcu_read_unlock_bh();
235
236 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
237 __func__);
238 kfree_skb(skb);
239 return -EINVAL;
240}
241
242static int ip_finish_output_gso(struct net *net, struct sock *sk,
243 struct sk_buff *skb, unsigned int mtu)
244{
245 netdev_features_t features;
246 struct sk_buff *segs;
247 int ret = 0;
248
249
250
251 if (skb_gso_validate_mtu(skb, mtu))
252 return ip_finish_output2(net, sk, skb);
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267 features = netif_skb_features(skb);
268 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
269 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
270 if (IS_ERR_OR_NULL(segs)) {
271 kfree_skb(skb);
272 return -ENOMEM;
273 }
274
275 consume_skb(skb);
276
277 do {
278 struct sk_buff *nskb = segs->next;
279 int err;
280
281 segs->next = NULL;
282 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
283
284 if (err && ret == 0)
285 ret = err;
286 segs = nskb;
287 } while (segs);
288
289 return ret;
290}
291
292static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
293{
294 unsigned int mtu;
295 int ret;
296
297 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
298 if (ret) {
299 kfree_skb(skb);
300 return ret;
301 }
302
303#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
304
305 if (skb_dst(skb)->xfrm) {
306 IPCB(skb)->flags |= IPSKB_REROUTED;
307 return dst_output(net, sk, skb);
308 }
309#endif
310 mtu = ip_skb_dst_mtu(sk, skb);
311 if (skb_is_gso(skb))
312 return ip_finish_output_gso(net, sk, skb, mtu);
313
314 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
315 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
316
317 return ip_finish_output2(net, sk, skb);
318}
319
320static int ip_mc_finish_output(struct net *net, struct sock *sk,
321 struct sk_buff *skb)
322{
323 int ret;
324
325 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
326 if (ret) {
327 kfree_skb(skb);
328 return ret;
329 }
330
331 return dev_loopback_xmit(net, sk, skb);
332}
333
334int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
335{
336 struct rtable *rt = skb_rtable(skb);
337 struct net_device *dev = rt->dst.dev;
338
339
340
341
342 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
343
344 skb->dev = dev;
345 skb->protocol = htons(ETH_P_IP);
346
347
348
349
350
351 if (rt->rt_flags&RTCF_MULTICAST) {
352 if (sk_mc_loop(sk)
353#ifdef CONFIG_IP_MROUTE
354
355
356
357
358
359
360
361
362 &&
363 ((rt->rt_flags & RTCF_LOCAL) ||
364 !(IPCB(skb)->flags & IPSKB_FORWARDED))
365#endif
366 ) {
367 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
368 if (newskb)
369 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
370 net, sk, newskb, NULL, newskb->dev,
371 ip_mc_finish_output);
372 }
373
374
375
376 if (ip_hdr(skb)->ttl == 0) {
377 kfree_skb(skb);
378 return 0;
379 }
380 }
381
382 if (rt->rt_flags&RTCF_BROADCAST) {
383 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
384 if (newskb)
385 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
386 net, sk, newskb, NULL, newskb->dev,
387 ip_mc_finish_output);
388 }
389
390 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
391 net, sk, skb, NULL, skb->dev,
392 ip_finish_output,
393 !(IPCB(skb)->flags & IPSKB_REROUTED));
394}
395
396int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
397{
398 struct net_device *dev = skb_dst(skb)->dev;
399
400 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
401
402 skb->dev = dev;
403 skb->protocol = htons(ETH_P_IP);
404
405 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
406 net, sk, skb, NULL, dev,
407 ip_finish_output,
408 !(IPCB(skb)->flags & IPSKB_REROUTED));
409}
410
411
412
413
414
415
416
417static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
418{
419 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
420 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
421 memcpy(&iph->saddr, &fl4->saddr,
422 sizeof(fl4->saddr) + sizeof(fl4->daddr));
423}
424
425
426int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
427{
428 struct inet_sock *inet = inet_sk(sk);
429 struct net *net = sock_net(sk);
430 struct ip_options_rcu *inet_opt;
431 struct flowi4 *fl4;
432 struct rtable *rt;
433 struct iphdr *iph;
434 int res;
435
436
437
438
439 rcu_read_lock();
440 inet_opt = rcu_dereference(inet->inet_opt);
441 fl4 = &fl->u.ip4;
442 rt = skb_rtable(skb);
443 if (rt)
444 goto packet_routed;
445
446
447 rt = (struct rtable *)__sk_dst_check(sk, 0);
448 if (!rt) {
449 __be32 daddr;
450
451
452 daddr = inet->inet_daddr;
453 if (inet_opt && inet_opt->opt.srr)
454 daddr = inet_opt->opt.faddr;
455
456
457
458
459
460 rt = ip_route_output_ports(net, fl4, sk,
461 daddr, inet->inet_saddr,
462 inet->inet_dport,
463 inet->inet_sport,
464 sk->sk_protocol,
465 RT_CONN_FLAGS(sk),
466 sk->sk_bound_dev_if);
467 if (IS_ERR(rt))
468 goto no_route;
469 sk_setup_caps(sk, &rt->dst);
470 }
471 skb_dst_set_noref(skb, &rt->dst);
472
473packet_routed:
474 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
475 goto no_route;
476
477
478 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
479 skb_reset_network_header(skb);
480 iph = ip_hdr(skb);
481 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
482 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
483 iph->frag_off = htons(IP_DF);
484 else
485 iph->frag_off = 0;
486 iph->ttl = ip_select_ttl(inet, &rt->dst);
487 iph->protocol = sk->sk_protocol;
488 ip_copy_addrs(iph, fl4);
489
490
491
492 if (inet_opt && inet_opt->opt.optlen) {
493 iph->ihl += inet_opt->opt.optlen >> 2;
494 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
495 }
496
497 ip_select_ident_segs(net, skb, sk,
498 skb_shinfo(skb)->gso_segs ?: 1);
499
500
501 skb->priority = sk->sk_priority;
502 skb->mark = sk->sk_mark;
503
504 res = ip_local_out(net, sk, skb);
505 rcu_read_unlock();
506 return res;
507
508no_route:
509 rcu_read_unlock();
510 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
511 kfree_skb(skb);
512 return -EHOSTUNREACH;
513}
514EXPORT_SYMBOL(ip_queue_xmit);
515
516static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
517{
518 to->pkt_type = from->pkt_type;
519 to->priority = from->priority;
520 to->protocol = from->protocol;
521 skb_dst_drop(to);
522 skb_dst_copy(to, from);
523 to->dev = from->dev;
524 to->mark = from->mark;
525
526
527 IPCB(to)->flags = IPCB(from)->flags;
528
529#ifdef CONFIG_NET_SCHED
530 to->tc_index = from->tc_index;
531#endif
532 nf_copy(to, from);
533#if IS_ENABLED(CONFIG_IP_VS)
534 to->ipvs_property = from->ipvs_property;
535#endif
536 skb_copy_secmark(to, from);
537}
538
539static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
540 unsigned int mtu,
541 int (*output)(struct net *, struct sock *, struct sk_buff *))
542{
543 struct iphdr *iph = ip_hdr(skb);
544
545 if ((iph->frag_off & htons(IP_DF)) == 0)
546 return ip_do_fragment(net, sk, skb, output);
547
548 if (unlikely(!skb->ignore_df ||
549 (IPCB(skb)->frag_max_size &&
550 IPCB(skb)->frag_max_size > mtu))) {
551 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
552 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
553 htonl(mtu));
554 kfree_skb(skb);
555 return -EMSGSIZE;
556 }
557
558 return ip_do_fragment(net, sk, skb, output);
559}
560
561
562
563
564
565
566
567
568int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
569 int (*output)(struct net *, struct sock *, struct sk_buff *))
570{
571 struct iphdr *iph;
572 int ptr;
573 struct sk_buff *skb2;
574 unsigned int mtu, hlen, left, len, ll_rs;
575 int offset;
576 __be16 not_last_frag;
577 struct rtable *rt = skb_rtable(skb);
578 int err = 0;
579
580
581 if (skb->ip_summed == CHECKSUM_PARTIAL &&
582 (err = skb_checksum_help(skb)))
583 goto fail;
584
585
586
587
588
589 iph = ip_hdr(skb);
590
591 mtu = ip_skb_dst_mtu(sk, skb);
592 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
593 mtu = IPCB(skb)->frag_max_size;
594
595
596
597
598
599 hlen = iph->ihl * 4;
600 mtu = mtu - hlen;
601 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
602 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
603
604
605
606
607
608
609
610
611 if (skb_has_frag_list(skb)) {
612 struct sk_buff *frag, *frag2;
613 unsigned int first_len = skb_pagelen(skb);
614
615 if (first_len - hlen > mtu ||
616 ((first_len - hlen) & 7) ||
617 ip_is_fragment(iph) ||
618 skb_cloned(skb) ||
619 skb_headroom(skb) < ll_rs)
620 goto slow_path;
621
622 skb_walk_frags(skb, frag) {
623
624 if (frag->len > mtu ||
625 ((frag->len & 7) && frag->next) ||
626 skb_headroom(frag) < hlen + ll_rs)
627 goto slow_path_clean;
628
629
630 if (skb_shared(frag))
631 goto slow_path_clean;
632
633 BUG_ON(frag->sk);
634 if (skb->sk) {
635 frag->sk = skb->sk;
636 frag->destructor = sock_wfree;
637 }
638 skb->truesize -= frag->truesize;
639 }
640
641
642
643 err = 0;
644 offset = 0;
645 frag = skb_shinfo(skb)->frag_list;
646 skb_frag_list_init(skb);
647 skb->data_len = first_len - skb_headlen(skb);
648 skb->len = first_len;
649 iph->tot_len = htons(first_len);
650 iph->frag_off = htons(IP_MF);
651 ip_send_check(iph);
652
653 for (;;) {
654
655
656 if (frag) {
657 frag->ip_summed = CHECKSUM_NONE;
658 skb_reset_transport_header(frag);
659 __skb_push(frag, hlen);
660 skb_reset_network_header(frag);
661 memcpy(skb_network_header(frag), iph, hlen);
662 iph = ip_hdr(frag);
663 iph->tot_len = htons(frag->len);
664 ip_copy_metadata(frag, skb);
665 if (offset == 0)
666 ip_options_fragment(frag);
667 offset += skb->len - hlen;
668 iph->frag_off = htons(offset>>3);
669 if (frag->next)
670 iph->frag_off |= htons(IP_MF);
671
672 ip_send_check(iph);
673 }
674
675 err = output(net, sk, skb);
676
677 if (!err)
678 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
679 if (err || !frag)
680 break;
681
682 skb = frag;
683 frag = skb->next;
684 skb->next = NULL;
685 }
686
687 if (err == 0) {
688 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
689 return 0;
690 }
691
692 while (frag) {
693 skb = frag->next;
694 kfree_skb(frag);
695 frag = skb;
696 }
697 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
698 return err;
699
700slow_path_clean:
701 skb_walk_frags(skb, frag2) {
702 if (frag2 == frag)
703 break;
704 frag2->sk = NULL;
705 frag2->destructor = NULL;
706 skb->truesize += frag2->truesize;
707 }
708 }
709
710slow_path:
711 iph = ip_hdr(skb);
712
713 left = skb->len - hlen;
714 ptr = hlen;
715
716
717
718
719
720 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
721 not_last_frag = iph->frag_off & htons(IP_MF);
722
723
724
725
726
727 while (left > 0) {
728 len = left;
729
730 if (len > mtu)
731 len = mtu;
732
733
734 if (len < left) {
735 len &= ~7;
736 }
737
738
739 skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
740 if (!skb2) {
741 err = -ENOMEM;
742 goto fail;
743 }
744
745
746
747
748
749 ip_copy_metadata(skb2, skb);
750 skb_reserve(skb2, ll_rs);
751 skb_put(skb2, len + hlen);
752 skb_reset_network_header(skb2);
753 skb2->transport_header = skb2->network_header + hlen;
754
755
756
757
758
759
760 if (skb->sk)
761 skb_set_owner_w(skb2, skb->sk);
762
763
764
765
766
767 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
768
769
770
771
772 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
773 BUG();
774 left -= len;
775
776
777
778
779 iph = ip_hdr(skb2);
780 iph->frag_off = htons((offset >> 3));
781
782 if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
783 iph->frag_off |= htons(IP_DF);
784
785
786
787
788
789
790
791 if (offset == 0)
792 ip_options_fragment(skb);
793
794
795
796
797
798 if (left > 0 || not_last_frag)
799 iph->frag_off |= htons(IP_MF);
800 ptr += len;
801 offset += len;
802
803
804
805
806 iph->tot_len = htons(len + hlen);
807
808 ip_send_check(iph);
809
810 err = output(net, sk, skb2);
811 if (err)
812 goto fail;
813
814 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
815 }
816 consume_skb(skb);
817 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
818 return err;
819
820fail:
821 kfree_skb(skb);
822 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
823 return err;
824}
825EXPORT_SYMBOL(ip_do_fragment);
826
827int
828ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
829{
830 struct msghdr *msg = from;
831
832 if (skb->ip_summed == CHECKSUM_PARTIAL) {
833 if (!copy_from_iter_full(to, len, &msg->msg_iter))
834 return -EFAULT;
835 } else {
836 __wsum csum = 0;
837 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
838 return -EFAULT;
839 skb->csum = csum_block_add(skb->csum, csum, odd);
840 }
841 return 0;
842}
843EXPORT_SYMBOL(ip_generic_getfrag);
844
845static inline __wsum
846csum_page(struct page *page, int offset, int copy)
847{
848 char *kaddr;
849 __wsum csum;
850 kaddr = kmap(page);
851 csum = csum_partial(kaddr + offset, copy, 0);
852 kunmap(page);
853 return csum;
854}
855
856static inline int ip_ufo_append_data(struct sock *sk,
857 struct sk_buff_head *queue,
858 int getfrag(void *from, char *to, int offset, int len,
859 int odd, struct sk_buff *skb),
860 void *from, int length, int hh_len, int fragheaderlen,
861 int transhdrlen, int maxfraglen, unsigned int flags)
862{
863 struct sk_buff *skb;
864 int err;
865
866
867
868
869
870 skb = skb_peek_tail(queue);
871 if (!skb) {
872 skb = sock_alloc_send_skb(sk,
873 hh_len + fragheaderlen + transhdrlen + 20,
874 (flags & MSG_DONTWAIT), &err);
875
876 if (!skb)
877 return err;
878
879
880 skb_reserve(skb, hh_len);
881
882
883 skb_put(skb, fragheaderlen + transhdrlen);
884
885
886 skb_reset_network_header(skb);
887
888
889 skb->transport_header = skb->network_header + fragheaderlen;
890
891 skb->csum = 0;
892
893 if (flags & MSG_CONFIRM)
894 skb_set_dst_pending_confirm(skb, 1);
895
896 __skb_queue_tail(queue, skb);
897 } else if (skb_is_gso(skb)) {
898 goto append;
899 }
900
901 skb->ip_summed = CHECKSUM_PARTIAL;
902
903 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
904 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
905
906append:
907 return skb_append_datato_frags(sk, skb, getfrag, from,
908 (length - transhdrlen));
909}
910
911static int __ip_append_data(struct sock *sk,
912 struct flowi4 *fl4,
913 struct sk_buff_head *queue,
914 struct inet_cork *cork,
915 struct page_frag *pfrag,
916 int getfrag(void *from, char *to, int offset,
917 int len, int odd, struct sk_buff *skb),
918 void *from, int length, int transhdrlen,
919 unsigned int flags)
920{
921 struct inet_sock *inet = inet_sk(sk);
922 struct sk_buff *skb;
923
924 struct ip_options *opt = cork->opt;
925 int hh_len;
926 int exthdrlen;
927 int mtu;
928 int copy;
929 int err;
930 int offset = 0;
931 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
932 int csummode = CHECKSUM_NONE;
933 struct rtable *rt = (struct rtable *)cork->dst;
934 u32 tskey = 0;
935
936 skb = skb_peek_tail(queue);
937
938 exthdrlen = !skb ? rt->dst.header_len : 0;
939 mtu = cork->fragsize;
940 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
941 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
942 tskey = sk->sk_tskey++;
943
944 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
945
946 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
947 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
948 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
949
950 if (cork->length + length > maxnonfragsize - fragheaderlen) {
951 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
952 mtu - (opt ? opt->optlen : 0));
953 return -EMSGSIZE;
954 }
955
956
957
958
959
960 if (transhdrlen &&
961 length + fragheaderlen <= mtu &&
962 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
963 !(flags & MSG_MORE) &&
964 !exthdrlen)
965 csummode = CHECKSUM_PARTIAL;
966
967 cork->length += length;
968 if ((skb && skb_is_gso(skb)) ||
969 (((length + (skb ? skb->len : fragheaderlen)) > mtu) &&
970 (skb_queue_len(queue) <= 1) &&
971 (sk->sk_protocol == IPPROTO_UDP) &&
972 (rt->dst.dev->features & NETIF_F_UFO) && !dst_xfrm(&rt->dst) &&
973 (sk->sk_type == SOCK_DGRAM) && !sk->sk_no_check_tx)) {
974 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
975 hh_len, fragheaderlen, transhdrlen,
976 maxfraglen, flags);
977 if (err)
978 goto error;
979 return 0;
980 }
981
982
983
984
985
986
987
988
989 if (!skb)
990 goto alloc_new_skb;
991
992 while (length > 0) {
993
994 copy = mtu - skb->len;
995 if (copy < length)
996 copy = maxfraglen - skb->len;
997 if (copy <= 0) {
998 char *data;
999 unsigned int datalen;
1000 unsigned int fraglen;
1001 unsigned int fraggap;
1002 unsigned int alloclen;
1003 struct sk_buff *skb_prev;
1004alloc_new_skb:
1005 skb_prev = skb;
1006 if (skb_prev)
1007 fraggap = skb_prev->len - maxfraglen;
1008 else
1009 fraggap = 0;
1010
1011
1012
1013
1014
1015 datalen = length + fraggap;
1016 if (datalen > mtu - fragheaderlen)
1017 datalen = maxfraglen - fragheaderlen;
1018 fraglen = datalen + fragheaderlen;
1019
1020 if ((flags & MSG_MORE) &&
1021 !(rt->dst.dev->features&NETIF_F_SG))
1022 alloclen = mtu;
1023 else
1024 alloclen = fraglen;
1025
1026 alloclen += exthdrlen;
1027
1028
1029
1030
1031
1032
1033 if (datalen == length + fraggap)
1034 alloclen += rt->dst.trailer_len;
1035
1036 if (transhdrlen) {
1037 skb = sock_alloc_send_skb(sk,
1038 alloclen + hh_len + 15,
1039 (flags & MSG_DONTWAIT), &err);
1040 } else {
1041 skb = NULL;
1042 if (refcount_read(&sk->sk_wmem_alloc) <=
1043 2 * sk->sk_sndbuf)
1044 skb = sock_wmalloc(sk,
1045 alloclen + hh_len + 15, 1,
1046 sk->sk_allocation);
1047 if (unlikely(!skb))
1048 err = -ENOBUFS;
1049 }
1050 if (!skb)
1051 goto error;
1052
1053
1054
1055
1056 skb->ip_summed = csummode;
1057 skb->csum = 0;
1058 skb_reserve(skb, hh_len);
1059
1060
1061 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1062 cork->tx_flags = 0;
1063 skb_shinfo(skb)->tskey = tskey;
1064 tskey = 0;
1065
1066
1067
1068
1069 data = skb_put(skb, fraglen + exthdrlen);
1070 skb_set_network_header(skb, exthdrlen);
1071 skb->transport_header = (skb->network_header +
1072 fragheaderlen);
1073 data += fragheaderlen + exthdrlen;
1074
1075 if (fraggap) {
1076 skb->csum = skb_copy_and_csum_bits(
1077 skb_prev, maxfraglen,
1078 data + transhdrlen, fraggap, 0);
1079 skb_prev->csum = csum_sub(skb_prev->csum,
1080 skb->csum);
1081 data += fraggap;
1082 pskb_trim_unique(skb_prev, maxfraglen);
1083 }
1084
1085 copy = datalen - transhdrlen - fraggap;
1086 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1087 err = -EFAULT;
1088 kfree_skb(skb);
1089 goto error;
1090 }
1091
1092 offset += copy;
1093 length -= datalen - fraggap;
1094 transhdrlen = 0;
1095 exthdrlen = 0;
1096 csummode = CHECKSUM_NONE;
1097
1098 if ((flags & MSG_CONFIRM) && !skb_prev)
1099 skb_set_dst_pending_confirm(skb, 1);
1100
1101
1102
1103
1104 __skb_queue_tail(queue, skb);
1105 continue;
1106 }
1107
1108 if (copy > length)
1109 copy = length;
1110
1111 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1112 unsigned int off;
1113
1114 off = skb->len;
1115 if (getfrag(from, skb_put(skb, copy),
1116 offset, copy, off, skb) < 0) {
1117 __skb_trim(skb, off);
1118 err = -EFAULT;
1119 goto error;
1120 }
1121 } else {
1122 int i = skb_shinfo(skb)->nr_frags;
1123
1124 err = -ENOMEM;
1125 if (!sk_page_frag_refill(sk, pfrag))
1126 goto error;
1127
1128 if (!skb_can_coalesce(skb, i, pfrag->page,
1129 pfrag->offset)) {
1130 err = -EMSGSIZE;
1131 if (i == MAX_SKB_FRAGS)
1132 goto error;
1133
1134 __skb_fill_page_desc(skb, i, pfrag->page,
1135 pfrag->offset, 0);
1136 skb_shinfo(skb)->nr_frags = ++i;
1137 get_page(pfrag->page);
1138 }
1139 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1140 if (getfrag(from,
1141 page_address(pfrag->page) + pfrag->offset,
1142 offset, copy, skb->len, skb) < 0)
1143 goto error_efault;
1144
1145 pfrag->offset += copy;
1146 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1147 skb->len += copy;
1148 skb->data_len += copy;
1149 skb->truesize += copy;
1150 refcount_add(copy, &sk->sk_wmem_alloc);
1151 }
1152 offset += copy;
1153 length -= copy;
1154 }
1155
1156 return 0;
1157
1158error_efault:
1159 err = -EFAULT;
1160error:
1161 cork->length -= length;
1162 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1163 return err;
1164}
1165
1166static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1167 struct ipcm_cookie *ipc, struct rtable **rtp)
1168{
1169 struct ip_options_rcu *opt;
1170 struct rtable *rt;
1171
1172
1173
1174
1175 opt = ipc->opt;
1176 if (opt) {
1177 if (!cork->opt) {
1178 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1179 sk->sk_allocation);
1180 if (unlikely(!cork->opt))
1181 return -ENOBUFS;
1182 }
1183 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1184 cork->flags |= IPCORK_OPT;
1185 cork->addr = ipc->addr;
1186 }
1187 rt = *rtp;
1188 if (unlikely(!rt))
1189 return -EFAULT;
1190
1191
1192
1193 *rtp = NULL;
1194 cork->fragsize = ip_sk_use_pmtu(sk) ?
1195 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1196 cork->dst = &rt->dst;
1197 cork->length = 0;
1198 cork->ttl = ipc->ttl;
1199 cork->tos = ipc->tos;
1200 cork->priority = ipc->priority;
1201 cork->tx_flags = ipc->tx_flags;
1202
1203 return 0;
1204}
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1218 int getfrag(void *from, char *to, int offset, int len,
1219 int odd, struct sk_buff *skb),
1220 void *from, int length, int transhdrlen,
1221 struct ipcm_cookie *ipc, struct rtable **rtp,
1222 unsigned int flags)
1223{
1224 struct inet_sock *inet = inet_sk(sk);
1225 int err;
1226
1227 if (flags&MSG_PROBE)
1228 return 0;
1229
1230 if (skb_queue_empty(&sk->sk_write_queue)) {
1231 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1232 if (err)
1233 return err;
1234 } else {
1235 transhdrlen = 0;
1236 }
1237
1238 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1239 sk_page_frag(sk), getfrag,
1240 from, length, transhdrlen, flags);
1241}
1242
1243ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1244 int offset, size_t size, int flags)
1245{
1246 struct inet_sock *inet = inet_sk(sk);
1247 struct sk_buff *skb;
1248 struct rtable *rt;
1249 struct ip_options *opt = NULL;
1250 struct inet_cork *cork;
1251 int hh_len;
1252 int mtu;
1253 int len;
1254 int err;
1255 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1256
1257 if (inet->hdrincl)
1258 return -EPERM;
1259
1260 if (flags&MSG_PROBE)
1261 return 0;
1262
1263 if (skb_queue_empty(&sk->sk_write_queue))
1264 return -EINVAL;
1265
1266 cork = &inet->cork.base;
1267 rt = (struct rtable *)cork->dst;
1268 if (cork->flags & IPCORK_OPT)
1269 opt = cork->opt;
1270
1271 if (!(rt->dst.dev->features&NETIF_F_SG))
1272 return -EOPNOTSUPP;
1273
1274 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1275 mtu = cork->fragsize;
1276
1277 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1278 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1279 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1280
1281 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1282 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1283 mtu - (opt ? opt->optlen : 0));
1284 return -EMSGSIZE;
1285 }
1286
1287 skb = skb_peek_tail(&sk->sk_write_queue);
1288 if (!skb)
1289 return -EINVAL;
1290
1291 if ((size + skb->len > mtu) &&
1292 (skb_queue_len(&sk->sk_write_queue) == 1) &&
1293 (sk->sk_protocol == IPPROTO_UDP) &&
1294 (rt->dst.dev->features & NETIF_F_UFO)) {
1295 if (skb->ip_summed != CHECKSUM_PARTIAL)
1296 return -EOPNOTSUPP;
1297
1298 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1299 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1300 }
1301 cork->length += size;
1302
1303 while (size > 0) {
1304 if (skb_is_gso(skb)) {
1305 len = size;
1306 } else {
1307
1308
1309 len = mtu - skb->len;
1310 if (len < size)
1311 len = maxfraglen - skb->len;
1312 }
1313 if (len <= 0) {
1314 struct sk_buff *skb_prev;
1315 int alloclen;
1316
1317 skb_prev = skb;
1318 fraggap = skb_prev->len - maxfraglen;
1319
1320 alloclen = fragheaderlen + hh_len + fraggap + 15;
1321 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1322 if (unlikely(!skb)) {
1323 err = -ENOBUFS;
1324 goto error;
1325 }
1326
1327
1328
1329
1330 skb->ip_summed = CHECKSUM_NONE;
1331 skb->csum = 0;
1332 skb_reserve(skb, hh_len);
1333
1334
1335
1336
1337 skb_put(skb, fragheaderlen + fraggap);
1338 skb_reset_network_header(skb);
1339 skb->transport_header = (skb->network_header +
1340 fragheaderlen);
1341 if (fraggap) {
1342 skb->csum = skb_copy_and_csum_bits(skb_prev,
1343 maxfraglen,
1344 skb_transport_header(skb),
1345 fraggap, 0);
1346 skb_prev->csum = csum_sub(skb_prev->csum,
1347 skb->csum);
1348 pskb_trim_unique(skb_prev, maxfraglen);
1349 }
1350
1351
1352
1353
1354 __skb_queue_tail(&sk->sk_write_queue, skb);
1355 continue;
1356 }
1357
1358 if (len > size)
1359 len = size;
1360
1361 if (skb_append_pagefrags(skb, page, offset, len)) {
1362 err = -EMSGSIZE;
1363 goto error;
1364 }
1365
1366 if (skb->ip_summed == CHECKSUM_NONE) {
1367 __wsum csum;
1368 csum = csum_page(page, offset, len);
1369 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1370 }
1371
1372 skb->len += len;
1373 skb->data_len += len;
1374 skb->truesize += len;
1375 refcount_add(len, &sk->sk_wmem_alloc);
1376 offset += len;
1377 size -= len;
1378 }
1379 return 0;
1380
1381error:
1382 cork->length -= size;
1383 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1384 return err;
1385}
1386
1387static void ip_cork_release(struct inet_cork *cork)
1388{
1389 cork->flags &= ~IPCORK_OPT;
1390 kfree(cork->opt);
1391 cork->opt = NULL;
1392 dst_release(cork->dst);
1393 cork->dst = NULL;
1394}
1395
1396
1397
1398
1399
1400struct sk_buff *__ip_make_skb(struct sock *sk,
1401 struct flowi4 *fl4,
1402 struct sk_buff_head *queue,
1403 struct inet_cork *cork)
1404{
1405 struct sk_buff *skb, *tmp_skb;
1406 struct sk_buff **tail_skb;
1407 struct inet_sock *inet = inet_sk(sk);
1408 struct net *net = sock_net(sk);
1409 struct ip_options *opt = NULL;
1410 struct rtable *rt = (struct rtable *)cork->dst;
1411 struct iphdr *iph;
1412 __be16 df = 0;
1413 __u8 ttl;
1414
1415 skb = __skb_dequeue(queue);
1416 if (!skb)
1417 goto out;
1418 tail_skb = &(skb_shinfo(skb)->frag_list);
1419
1420
1421 if (skb->data < skb_network_header(skb))
1422 __skb_pull(skb, skb_network_offset(skb));
1423 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1424 __skb_pull(tmp_skb, skb_network_header_len(skb));
1425 *tail_skb = tmp_skb;
1426 tail_skb = &(tmp_skb->next);
1427 skb->len += tmp_skb->len;
1428 skb->data_len += tmp_skb->len;
1429 skb->truesize += tmp_skb->truesize;
1430 tmp_skb->destructor = NULL;
1431 tmp_skb->sk = NULL;
1432 }
1433
1434
1435
1436
1437
1438 skb->ignore_df = ip_sk_ignore_df(sk);
1439
1440
1441
1442
1443 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1444 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1445 (skb->len <= dst_mtu(&rt->dst) &&
1446 ip_dont_fragment(sk, &rt->dst)))
1447 df = htons(IP_DF);
1448
1449 if (cork->flags & IPCORK_OPT)
1450 opt = cork->opt;
1451
1452 if (cork->ttl != 0)
1453 ttl = cork->ttl;
1454 else if (rt->rt_type == RTN_MULTICAST)
1455 ttl = inet->mc_ttl;
1456 else
1457 ttl = ip_select_ttl(inet, &rt->dst);
1458
1459 iph = ip_hdr(skb);
1460 iph->version = 4;
1461 iph->ihl = 5;
1462 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1463 iph->frag_off = df;
1464 iph->ttl = ttl;
1465 iph->protocol = sk->sk_protocol;
1466 ip_copy_addrs(iph, fl4);
1467 ip_select_ident(net, skb, sk);
1468
1469 if (opt) {
1470 iph->ihl += opt->optlen>>2;
1471 ip_options_build(skb, opt, cork->addr, rt, 0);
1472 }
1473
1474 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1475 skb->mark = sk->sk_mark;
1476
1477
1478
1479
1480 cork->dst = NULL;
1481 skb_dst_set(skb, &rt->dst);
1482
1483 if (iph->protocol == IPPROTO_ICMP)
1484 icmp_out_count(net, ((struct icmphdr *)
1485 skb_transport_header(skb))->type);
1486
1487 ip_cork_release(cork);
1488out:
1489 return skb;
1490}
1491
1492int ip_send_skb(struct net *net, struct sk_buff *skb)
1493{
1494 int err;
1495
1496 err = ip_local_out(net, skb->sk, skb);
1497 if (err) {
1498 if (err > 0)
1499 err = net_xmit_errno(err);
1500 if (err)
1501 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1502 }
1503
1504 return err;
1505}
1506
1507int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1508{
1509 struct sk_buff *skb;
1510
1511 skb = ip_finish_skb(sk, fl4);
1512 if (!skb)
1513 return 0;
1514
1515
1516 return ip_send_skb(sock_net(sk), skb);
1517}
1518
1519
1520
1521
1522static void __ip_flush_pending_frames(struct sock *sk,
1523 struct sk_buff_head *queue,
1524 struct inet_cork *cork)
1525{
1526 struct sk_buff *skb;
1527
1528 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1529 kfree_skb(skb);
1530
1531 ip_cork_release(cork);
1532}
1533
1534void ip_flush_pending_frames(struct sock *sk)
1535{
1536 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1537}
1538
1539struct sk_buff *ip_make_skb(struct sock *sk,
1540 struct flowi4 *fl4,
1541 int getfrag(void *from, char *to, int offset,
1542 int len, int odd, struct sk_buff *skb),
1543 void *from, int length, int transhdrlen,
1544 struct ipcm_cookie *ipc, struct rtable **rtp,
1545 unsigned int flags)
1546{
1547 struct inet_cork cork;
1548 struct sk_buff_head queue;
1549 int err;
1550
1551 if (flags & MSG_PROBE)
1552 return NULL;
1553
1554 __skb_queue_head_init(&queue);
1555
1556 cork.flags = 0;
1557 cork.addr = 0;
1558 cork.opt = NULL;
1559 err = ip_setup_cork(sk, &cork, ipc, rtp);
1560 if (err)
1561 return ERR_PTR(err);
1562
1563 err = __ip_append_data(sk, fl4, &queue, &cork,
1564 ¤t->task_frag, getfrag,
1565 from, length, transhdrlen, flags);
1566 if (err) {
1567 __ip_flush_pending_frames(sk, &queue, &cork);
1568 return ERR_PTR(err);
1569 }
1570
1571 return __ip_make_skb(sk, fl4, &queue, &cork);
1572}
1573
1574
1575
1576
1577static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1578 int len, int odd, struct sk_buff *skb)
1579{
1580 __wsum csum;
1581
1582 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1583 skb->csum = csum_block_add(skb->csum, csum, odd);
1584 return 0;
1585}
1586
1587
1588
1589
1590
1591void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1592 const struct ip_options *sopt,
1593 __be32 daddr, __be32 saddr,
1594 const struct ip_reply_arg *arg,
1595 unsigned int len)
1596{
1597 struct ip_options_data replyopts;
1598 struct ipcm_cookie ipc;
1599 struct flowi4 fl4;
1600 struct rtable *rt = skb_rtable(skb);
1601 struct net *net = sock_net(sk);
1602 struct sk_buff *nskb;
1603 int err;
1604 int oif;
1605
1606 if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
1607 return;
1608
1609 ipc.addr = daddr;
1610 ipc.opt = NULL;
1611 ipc.tx_flags = 0;
1612 ipc.ttl = 0;
1613 ipc.tos = -1;
1614
1615 if (replyopts.opt.opt.optlen) {
1616 ipc.opt = &replyopts.opt;
1617
1618 if (replyopts.opt.opt.srr)
1619 daddr = replyopts.opt.opt.faddr;
1620 }
1621
1622 oif = arg->bound_dev_if;
1623 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1624 oif = skb->skb_iif;
1625
1626 flowi4_init_output(&fl4, oif,
1627 IP4_REPLY_MARK(net, skb->mark),
1628 RT_TOS(arg->tos),
1629 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1630 ip_reply_arg_flowi_flags(arg),
1631 daddr, saddr,
1632 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1633 arg->uid);
1634 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1635 rt = ip_route_output_key(net, &fl4);
1636 if (IS_ERR(rt))
1637 return;
1638
1639 inet_sk(sk)->tos = arg->tos;
1640
1641 sk->sk_priority = skb->priority;
1642 sk->sk_protocol = ip_hdr(skb)->protocol;
1643 sk->sk_bound_dev_if = arg->bound_dev_if;
1644 sk->sk_sndbuf = sysctl_wmem_default;
1645 sk->sk_mark = fl4.flowi4_mark;
1646 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1647 len, 0, &ipc, &rt, MSG_DONTWAIT);
1648 if (unlikely(err)) {
1649 ip_flush_pending_frames(sk);
1650 goto out;
1651 }
1652
1653 nskb = skb_peek(&sk->sk_write_queue);
1654 if (nskb) {
1655 if (arg->csumoffset >= 0)
1656 *((__sum16 *)skb_transport_header(nskb) +
1657 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1658 arg->csum));
1659 nskb->ip_summed = CHECKSUM_NONE;
1660 ip_push_pending_frames(sk, &fl4);
1661 }
1662out:
1663 ip_rt_put(rt);
1664}
1665
1666void __init ip_init(void)
1667{
1668 ip_rt_init();
1669 inet_initpeers();
1670
1671#if defined(CONFIG_IP_MULTICAST)
1672 igmp_mc_init();
1673#endif
1674}
1675