Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The Internet Protocol (IP) output module.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Donald Becker, <becker@super.org>
12 * Alan Cox, <Alan.Cox@linux.org>
13 * Richard Underwood
14 * Stefan Becker, <stefanb@yello.ping.de>
15 * Jorge Cwik, <jorge@laser.satlink.net>
16 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
17 * Hirokazu Takahashi, <taka@valinux.co.jp>
18 *
19 * See ip_input.c for original log
20 *
21 * Fixes:
22 * Alan Cox : Missing nonblock feature in ip_build_xmit.
23 * Mike Kilburn : htons() missing in ip_build_xmit.
24 * Bradford Johnson: Fix faulty handling of some frames when
25 * no route is found.
26 * Alexander Demenshin: Missing sk/skb free in ip_queue_xmit
27 * (in case if packet not accepted by
28 * output firewall rules)
29 * Mike McLagan : Routing by source
30 * Alexey Kuznetsov: use new route cache
31 * Andi Kleen: Fix broken PMTU recovery and remove
32 * some redundant tests.
33 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
34 * Andi Kleen : Replace ip_reply with ip_send_reply.
35 * Andi Kleen : Split fast and slow ip_build_xmit path
36 * for decreased register pressure on x86
37 * and more readibility.
38 * Marc Boucher : When call_out_firewall returns FW_QUEUE,
39 * silently drop skb instead of failing with -EPERM.
40 * Detlev Wengorz : Copy protocol for fragments.
41 * Hirokazu Takahashi: HW checksumming for outgoing UDP
42 * datagrams.
43 * Hirokazu Takahashi: sendfile() on UDP works now.
44 */
45
46#include <linux/uaccess.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#include <linux/slab.h>
55
56#include <linux/socket.h>
57#include <linux/sockios.h>
58#include <linux/in.h>
59#include <linux/inet.h>
60#include <linux/netdevice.h>
61#include <linux/etherdevice.h>
62#include <linux/proc_fs.h>
63#include <linux/stat.h>
64#include <linux/init.h>
65
66#include <net/snmp.h>
67#include <net/ip.h>
68#include <net/protocol.h>
69#include <net/route.h>
70#include <net/xfrm.h>
71#include <linux/skbuff.h>
72#include <net/sock.h>
73#include <net/arp.h>
74#include <net/icmp.h>
75#include <net/checksum.h>
76#include <net/inetpeer.h>
77#include <net/inet_ecn.h>
78#include <net/lwtunnel.h>
79#include <linux/bpf-cgroup.h>
80#include <linux/igmp.h>
81#include <linux/netfilter_ipv4.h>
82#include <linux/netfilter_bridge.h>
83#include <linux/netlink.h>
84#include <linux/tcp.h>
85
86static int
87ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
88 unsigned int mtu,
89 int (*output)(struct net *, struct sock *, struct sk_buff *));
90
91/* Generate a checksum for an outgoing IP datagram. */
92void ip_send_check(struct iphdr *iph)
93{
94 iph->check = 0;
95 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
96}
97EXPORT_SYMBOL(ip_send_check);
98
99int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
100{
101 struct iphdr *iph = ip_hdr(skb);
102
103 iph->tot_len = htons(skb->len);
104 ip_send_check(iph);
105
106 /* if egress device is enslaved to an L3 master device pass the
107 * skb to its handler for processing
108 */
109 skb = l3mdev_ip_out(sk, skb);
110 if (unlikely(!skb))
111 return 0;
112
113 skb->protocol = htons(ETH_P_IP);
114
115 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
116 net, sk, skb, NULL, skb_dst(skb)->dev,
117 dst_output);
118}
119
120int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
121{
122 int err;
123
124 err = __ip_local_out(net, sk, skb);
125 if (likely(err == 1))
126 err = dst_output(net, sk, skb);
127
128 return err;
129}
130EXPORT_SYMBOL_GPL(ip_local_out);
131
132static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
133{
134 int ttl = inet->uc_ttl;
135
136 if (ttl < 0)
137 ttl = ip4_dst_hoplimit(dst);
138 return ttl;
139}
140
141/*
142 * Add an ip header to a skbuff and send it out.
143 *
144 */
145int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
146 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
147{
148 struct inet_sock *inet = inet_sk(sk);
149 struct rtable *rt = skb_rtable(skb);
150 struct net *net = sock_net(sk);
151 struct iphdr *iph;
152
153 /* Build the IP header. */
154 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
155 skb_reset_network_header(skb);
156 iph = ip_hdr(skb);
157 iph->version = 4;
158 iph->ihl = 5;
159 iph->tos = inet->tos;
160 iph->ttl = ip_select_ttl(inet, &rt->dst);
161 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
162 iph->saddr = saddr;
163 iph->protocol = sk->sk_protocol;
164 if (ip_dont_fragment(sk, &rt->dst)) {
165 iph->frag_off = htons(IP_DF);
166 iph->id = 0;
167 } else {
168 iph->frag_off = 0;
169 __ip_select_ident(net, iph, 1);
170 }
171
172 if (opt && opt->opt.optlen) {
173 iph->ihl += opt->opt.optlen>>2;
174 ip_options_build(skb, &opt->opt, daddr, rt, 0);
175 }
176
177 skb->priority = sk->sk_priority;
178 if (!skb->mark)
179 skb->mark = sk->sk_mark;
180
181 /* Send it out. */
182 return ip_local_out(net, skb->sk, skb);
183}
184EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
185
186static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
187{
188 struct dst_entry *dst = skb_dst(skb);
189 struct rtable *rt = (struct rtable *)dst;
190 struct net_device *dev = dst->dev;
191 unsigned int hh_len = LL_RESERVED_SPACE(dev);
192 struct neighbour *neigh;
193 bool is_v6gw = false;
194
195 if (rt->rt_type == RTN_MULTICAST) {
196 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
197 } else if (rt->rt_type == RTN_BROADCAST)
198 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
199
200 /* Be paranoid, rather than too clever. */
201 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
202 struct sk_buff *skb2;
203
204 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
205 if (!skb2) {
206 kfree_skb(skb);
207 return -ENOMEM;
208 }
209 if (skb->sk)
210 skb_set_owner_w(skb2, skb->sk);
211 consume_skb(skb);
212 skb = skb2;
213 }
214
215 if (lwtunnel_xmit_redirect(dst->lwtstate)) {
216 int res = lwtunnel_xmit(skb);
217
218 if (res < 0 || res == LWTUNNEL_XMIT_DONE)
219 return res;
220 }
221
222 rcu_read_lock_bh();
223 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
224 if (!IS_ERR(neigh)) {
225 int res;
226
227 sock_confirm_neigh(skb, neigh);
228 /* if crossing protocols, can not use the cached header */
229 res = neigh_output(neigh, skb, is_v6gw);
230 rcu_read_unlock_bh();
231 return res;
232 }
233 rcu_read_unlock_bh();
234
235 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
236 __func__);
237 kfree_skb(skb);
238 return -EINVAL;
239}
240
241static int ip_finish_output_gso(struct net *net, struct sock *sk,
242 struct sk_buff *skb, unsigned int mtu)
243{
244 struct sk_buff *segs, *nskb;
245 netdev_features_t features;
246 int ret = 0;
247
248 /* common case: seglen is <= mtu
249 */
250 if (skb_gso_validate_network_len(skb, mtu))
251 return ip_finish_output2(net, sk, skb);
252
253 /* Slowpath - GSO segment length exceeds the egress MTU.
254 *
255 * This can happen in several cases:
256 * - Forwarding of a TCP GRO skb, when DF flag is not set.
257 * - Forwarding of an skb that arrived on a virtualization interface
258 * (virtio-net/vhost/tap) with TSO/GSO size set by other network
259 * stack.
260 * - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
261 * interface with a smaller MTU.
262 * - Arriving GRO skb (or GSO skb in a virtualized environment) that is
263 * bridged to a NETIF_F_TSO tunnel stacked over an interface with an
264 * insufficent MTU.
265 */
266 features = netif_skb_features(skb);
267 BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
268 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
269 if (IS_ERR_OR_NULL(segs)) {
270 kfree_skb(skb);
271 return -ENOMEM;
272 }
273
274 consume_skb(skb);
275
276 skb_list_walk_safe(segs, segs, nskb) {
277 int err;
278
279 skb_mark_not_on_list(segs);
280 err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
281
282 if (err && ret == 0)
283 ret = err;
284 }
285
286 return ret;
287}
288
289static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
290{
291 unsigned int mtu;
292
293#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
294 /* Policy lookup after SNAT yielded a new policy */
295 if (skb_dst(skb)->xfrm) {
296 IPCB(skb)->flags |= IPSKB_REROUTED;
297 return dst_output(net, sk, skb);
298 }
299#endif
300 mtu = ip_skb_dst_mtu(sk, skb);
301 if (skb_is_gso(skb))
302 return ip_finish_output_gso(net, sk, skb, mtu);
303
304 if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
305 return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
306
307 return ip_finish_output2(net, sk, skb);
308}
309
310static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
311{
312 int ret;
313
314 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
315 switch (ret) {
316 case NET_XMIT_SUCCESS:
317 return __ip_finish_output(net, sk, skb);
318 case NET_XMIT_CN:
319 return __ip_finish_output(net, sk, skb) ? : ret;
320 default:
321 kfree_skb(skb);
322 return ret;
323 }
324}
325
326static int ip_mc_finish_output(struct net *net, struct sock *sk,
327 struct sk_buff *skb)
328{
329 struct rtable *new_rt;
330 bool do_cn = false;
331 int ret, err;
332
333 ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
334 switch (ret) {
335 case NET_XMIT_CN:
336 do_cn = true;
337 fallthrough;
338 case NET_XMIT_SUCCESS:
339 break;
340 default:
341 kfree_skb(skb);
342 return ret;
343 }
344
345 /* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
346 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
347 * see ipv4_pktinfo_prepare().
348 */
349 new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
350 if (new_rt) {
351 new_rt->rt_iif = 0;
352 skb_dst_drop(skb);
353 skb_dst_set(skb, &new_rt->dst);
354 }
355
356 err = dev_loopback_xmit(net, sk, skb);
357 return (do_cn && err) ? ret : err;
358}
359
360int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
361{
362 struct rtable *rt = skb_rtable(skb);
363 struct net_device *dev = rt->dst.dev;
364
365 /*
366 * If the indicated interface is up and running, send the packet.
367 */
368 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
369
370 skb->dev = dev;
371 skb->protocol = htons(ETH_P_IP);
372
373 /*
374 * Multicasts are looped back for other local users
375 */
376
377 if (rt->rt_flags&RTCF_MULTICAST) {
378 if (sk_mc_loop(sk)
379#ifdef CONFIG_IP_MROUTE
380 /* Small optimization: do not loopback not local frames,
381 which returned after forwarding; they will be dropped
382 by ip_mr_input in any case.
383 Note, that local frames are looped back to be delivered
384 to local recipients.
385
386 This check is duplicated in ip_mr_input at the moment.
387 */
388 &&
389 ((rt->rt_flags & RTCF_LOCAL) ||
390 !(IPCB(skb)->flags & IPSKB_FORWARDED))
391#endif
392 ) {
393 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
394 if (newskb)
395 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
396 net, sk, newskb, NULL, newskb->dev,
397 ip_mc_finish_output);
398 }
399
400 /* Multicasts with ttl 0 must not go beyond the host */
401
402 if (ip_hdr(skb)->ttl == 0) {
403 kfree_skb(skb);
404 return 0;
405 }
406 }
407
408 if (rt->rt_flags&RTCF_BROADCAST) {
409 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
410 if (newskb)
411 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
412 net, sk, newskb, NULL, newskb->dev,
413 ip_mc_finish_output);
414 }
415
416 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
417 net, sk, skb, NULL, skb->dev,
418 ip_finish_output,
419 !(IPCB(skb)->flags & IPSKB_REROUTED));
420}
421
422int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
423{
424 struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
425
426 IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
427
428 skb->dev = dev;
429 skb->protocol = htons(ETH_P_IP);
430
431 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
432 net, sk, skb, indev, dev,
433 ip_finish_output,
434 !(IPCB(skb)->flags & IPSKB_REROUTED));
435}
436
437/*
438 * copy saddr and daddr, possibly using 64bit load/stores
439 * Equivalent to :
440 * iph->saddr = fl4->saddr;
441 * iph->daddr = fl4->daddr;
442 */
443static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
444{
445 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
446 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
447 memcpy(&iph->saddr, &fl4->saddr,
448 sizeof(fl4->saddr) + sizeof(fl4->daddr));
449}
450
451/* Note: skb->sk can be different from sk, in case of tunnels */
452int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
453 __u8 tos)
454{
455 struct inet_sock *inet = inet_sk(sk);
456 struct net *net = sock_net(sk);
457 struct ip_options_rcu *inet_opt;
458 struct flowi4 *fl4;
459 struct rtable *rt;
460 struct iphdr *iph;
461 int res;
462
463 /* Skip all of this if the packet is already routed,
464 * f.e. by something like SCTP.
465 */
466 rcu_read_lock();
467 inet_opt = rcu_dereference(inet->inet_opt);
468 fl4 = &fl->u.ip4;
469 rt = skb_rtable(skb);
470 if (rt)
471 goto packet_routed;
472
473 /* Make sure we can route this packet. */
474 rt = (struct rtable *)__sk_dst_check(sk, 0);
475 if (!rt) {
476 __be32 daddr;
477
478 /* Use correct destination address if we have options. */
479 daddr = inet->inet_daddr;
480 if (inet_opt && inet_opt->opt.srr)
481 daddr = inet_opt->opt.faddr;
482
483 /* If this fails, retransmit mechanism of transport layer will
484 * keep trying until route appears or the connection times
485 * itself out.
486 */
487 rt = ip_route_output_ports(net, fl4, sk,
488 daddr, inet->inet_saddr,
489 inet->inet_dport,
490 inet->inet_sport,
491 sk->sk_protocol,
492 RT_CONN_FLAGS_TOS(sk, tos),
493 sk->sk_bound_dev_if);
494 if (IS_ERR(rt))
495 goto no_route;
496 sk_setup_caps(sk, &rt->dst);
497 }
498 skb_dst_set_noref(skb, &rt->dst);
499
500packet_routed:
501 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
502 goto no_route;
503
504 /* OK, we know where to send it, allocate and build IP header. */
505 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
506 skb_reset_network_header(skb);
507 iph = ip_hdr(skb);
508 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
509 if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
510 iph->frag_off = htons(IP_DF);
511 else
512 iph->frag_off = 0;
513 iph->ttl = ip_select_ttl(inet, &rt->dst);
514 iph->protocol = sk->sk_protocol;
515 ip_copy_addrs(iph, fl4);
516
517 /* Transport layer set skb->h.foo itself. */
518
519 if (inet_opt && inet_opt->opt.optlen) {
520 iph->ihl += inet_opt->opt.optlen >> 2;
521 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
522 }
523
524 ip_select_ident_segs(net, skb, sk,
525 skb_shinfo(skb)->gso_segs ?: 1);
526
527 /* TODO : should we use skb->sk here instead of sk ? */
528 skb->priority = sk->sk_priority;
529 skb->mark = sk->sk_mark;
530
531 res = ip_local_out(net, sk, skb);
532 rcu_read_unlock();
533 return res;
534
535no_route:
536 rcu_read_unlock();
537 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
538 kfree_skb(skb);
539 return -EHOSTUNREACH;
540}
541EXPORT_SYMBOL(__ip_queue_xmit);
542
543int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
544{
545 return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
546}
547EXPORT_SYMBOL(ip_queue_xmit);
548
549static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
550{
551 to->pkt_type = from->pkt_type;
552 to->priority = from->priority;
553 to->protocol = from->protocol;
554 to->skb_iif = from->skb_iif;
555 skb_dst_drop(to);
556 skb_dst_copy(to, from);
557 to->dev = from->dev;
558 to->mark = from->mark;
559
560 skb_copy_hash(to, from);
561
562#ifdef CONFIG_NET_SCHED
563 to->tc_index = from->tc_index;
564#endif
565 nf_copy(to, from);
566 skb_ext_copy(to, from);
567#if IS_ENABLED(CONFIG_IP_VS)
568 to->ipvs_property = from->ipvs_property;
569#endif
570 skb_copy_secmark(to, from);
571}
572
573static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
574 unsigned int mtu,
575 int (*output)(struct net *, struct sock *, struct sk_buff *))
576{
577 struct iphdr *iph = ip_hdr(skb);
578
579 if ((iph->frag_off & htons(IP_DF)) == 0)
580 return ip_do_fragment(net, sk, skb, output);
581
582 if (unlikely(!skb->ignore_df ||
583 (IPCB(skb)->frag_max_size &&
584 IPCB(skb)->frag_max_size > mtu))) {
585 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
586 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
587 htonl(mtu));
588 kfree_skb(skb);
589 return -EMSGSIZE;
590 }
591
592 return ip_do_fragment(net, sk, skb, output);
593}
594
595void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
596 unsigned int hlen, struct ip_fraglist_iter *iter)
597{
598 unsigned int first_len = skb_pagelen(skb);
599
600 iter->frag = skb_shinfo(skb)->frag_list;
601 skb_frag_list_init(skb);
602
603 iter->offset = 0;
604 iter->iph = iph;
605 iter->hlen = hlen;
606
607 skb->data_len = first_len - skb_headlen(skb);
608 skb->len = first_len;
609 iph->tot_len = htons(first_len);
610 iph->frag_off = htons(IP_MF);
611 ip_send_check(iph);
612}
613EXPORT_SYMBOL(ip_fraglist_init);
614
615static void ip_fraglist_ipcb_prepare(struct sk_buff *skb,
616 struct ip_fraglist_iter *iter)
617{
618 struct sk_buff *to = iter->frag;
619
620 /* Copy the flags to each fragment. */
621 IPCB(to)->flags = IPCB(skb)->flags;
622
623 if (iter->offset == 0)
624 ip_options_fragment(to);
625}
626
627void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
628{
629 unsigned int hlen = iter->hlen;
630 struct iphdr *iph = iter->iph;
631 struct sk_buff *frag;
632
633 frag = iter->frag;
634 frag->ip_summed = CHECKSUM_NONE;
635 skb_reset_transport_header(frag);
636 __skb_push(frag, hlen);
637 skb_reset_network_header(frag);
638 memcpy(skb_network_header(frag), iph, hlen);
639 iter->iph = ip_hdr(frag);
640 iph = iter->iph;
641 iph->tot_len = htons(frag->len);
642 ip_copy_metadata(frag, skb);
643 iter->offset += skb->len - hlen;
644 iph->frag_off = htons(iter->offset >> 3);
645 if (frag->next)
646 iph->frag_off |= htons(IP_MF);
647 /* Ready, complete checksum */
648 ip_send_check(iph);
649}
650EXPORT_SYMBOL(ip_fraglist_prepare);
651
652void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
653 unsigned int ll_rs, unsigned int mtu, bool DF,
654 struct ip_frag_state *state)
655{
656 struct iphdr *iph = ip_hdr(skb);
657
658 state->DF = DF;
659 state->hlen = hlen;
660 state->ll_rs = ll_rs;
661 state->mtu = mtu;
662
663 state->left = skb->len - hlen; /* Space per frame */
664 state->ptr = hlen; /* Where to start from */
665
666 state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
667 state->not_last_frag = iph->frag_off & htons(IP_MF);
668}
669EXPORT_SYMBOL(ip_frag_init);
670
671static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
672 bool first_frag, struct ip_frag_state *state)
673{
674 /* Copy the flags to each fragment. */
675 IPCB(to)->flags = IPCB(from)->flags;
676
677 /* ANK: dirty, but effective trick. Upgrade options only if
678 * the segment to be fragmented was THE FIRST (otherwise,
679 * options are already fixed) and make it ONCE
680 * on the initial skb, so that all the following fragments
681 * will inherit fixed options.
682 */
683 if (first_frag)
684 ip_options_fragment(from);
685}
686
687struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
688{
689 unsigned int len = state->left;
690 struct sk_buff *skb2;
691 struct iphdr *iph;
692
693 len = state->left;
694 /* IF: it doesn't fit, use 'mtu' - the data space left */
695 if (len > state->mtu)
696 len = state->mtu;
697 /* IF: we are not sending up to and including the packet end
698 then align the next start on an eight byte boundary */
699 if (len < state->left) {
700 len &= ~7;
701 }
702
703 /* Allocate buffer */
704 skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
705 if (!skb2)
706 return ERR_PTR(-ENOMEM);
707
708 /*
709 * Set up data on packet
710 */
711
712 ip_copy_metadata(skb2, skb);
713 skb_reserve(skb2, state->ll_rs);
714 skb_put(skb2, len + state->hlen);
715 skb_reset_network_header(skb2);
716 skb2->transport_header = skb2->network_header + state->hlen;
717
718 /*
719 * Charge the memory for the fragment to any owner
720 * it might possess
721 */
722
723 if (skb->sk)
724 skb_set_owner_w(skb2, skb->sk);
725
726 /*
727 * Copy the packet header into the new buffer.
728 */
729
730 skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
731
732 /*
733 * Copy a block of the IP datagram.
734 */
735 if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
736 BUG();
737 state->left -= len;
738
739 /*
740 * Fill in the new header fields.
741 */
742 iph = ip_hdr(skb2);
743 iph->frag_off = htons((state->offset >> 3));
744 if (state->DF)
745 iph->frag_off |= htons(IP_DF);
746
747 /*
748 * Added AC : If we are fragmenting a fragment that's not the
749 * last fragment then keep MF on each bit
750 */
751 if (state->left > 0 || state->not_last_frag)
752 iph->frag_off |= htons(IP_MF);
753 state->ptr += len;
754 state->offset += len;
755
756 iph->tot_len = htons(len + state->hlen);
757
758 ip_send_check(iph);
759
760 return skb2;
761}
762EXPORT_SYMBOL(ip_frag_next);
763
764/*
765 * This IP datagram is too large to be sent in one piece. Break it up into
766 * smaller pieces (each of size equal to IP header plus
767 * a block of the data of the original IP data part) that will yet fit in a
768 * single device frame, and queue such a frame for sending.
769 */
770
771int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
772 int (*output)(struct net *, struct sock *, struct sk_buff *))
773{
774 struct iphdr *iph;
775 struct sk_buff *skb2;
776 struct rtable *rt = skb_rtable(skb);
777 unsigned int mtu, hlen, ll_rs;
778 struct ip_fraglist_iter iter;
779 ktime_t tstamp = skb->tstamp;
780 struct ip_frag_state state;
781 int err = 0;
782
783 /* for offloaded checksums cleanup checksum before fragmentation */
784 if (skb->ip_summed == CHECKSUM_PARTIAL &&
785 (err = skb_checksum_help(skb)))
786 goto fail;
787
788 /*
789 * Point into the IP datagram header.
790 */
791
792 iph = ip_hdr(skb);
793
794 mtu = ip_skb_dst_mtu(sk, skb);
795 if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
796 mtu = IPCB(skb)->frag_max_size;
797
798 /*
799 * Setup starting values.
800 */
801
802 hlen = iph->ihl * 4;
803 mtu = mtu - hlen; /* Size of data space */
804 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
805 ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
806
807 /* When frag_list is given, use it. First, check its validity:
808 * some transformers could create wrong frag_list or break existing
809 * one, it is not prohibited. In this case fall back to copying.
810 *
811 * LATER: this step can be merged to real generation of fragments,
812 * we can switch to copy when see the first bad fragment.
813 */
814 if (skb_has_frag_list(skb)) {
815 struct sk_buff *frag, *frag2;
816 unsigned int first_len = skb_pagelen(skb);
817
818 if (first_len - hlen > mtu ||
819 ((first_len - hlen) & 7) ||
820 ip_is_fragment(iph) ||
821 skb_cloned(skb) ||
822 skb_headroom(skb) < ll_rs)
823 goto slow_path;
824
825 skb_walk_frags(skb, frag) {
826 /* Correct geometry. */
827 if (frag->len > mtu ||
828 ((frag->len & 7) && frag->next) ||
829 skb_headroom(frag) < hlen + ll_rs)
830 goto slow_path_clean;
831
832 /* Partially cloned skb? */
833 if (skb_shared(frag))
834 goto slow_path_clean;
835
836 BUG_ON(frag->sk);
837 if (skb->sk) {
838 frag->sk = skb->sk;
839 frag->destructor = sock_wfree;
840 }
841 skb->truesize -= frag->truesize;
842 }
843
844 /* Everything is OK. Generate! */
845 ip_fraglist_init(skb, iph, hlen, &iter);
846
847 for (;;) {
848 /* Prepare header of the next frame,
849 * before previous one went down. */
850 if (iter.frag) {
851 ip_fraglist_ipcb_prepare(skb, &iter);
852 ip_fraglist_prepare(skb, &iter);
853 }
854
855 skb->tstamp = tstamp;
856 err = output(net, sk, skb);
857
858 if (!err)
859 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
860 if (err || !iter.frag)
861 break;
862
863 skb = ip_fraglist_next(&iter);
864 }
865
866 if (err == 0) {
867 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
868 return 0;
869 }
870
871 kfree_skb_list(iter.frag);
872
873 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
874 return err;
875
876slow_path_clean:
877 skb_walk_frags(skb, frag2) {
878 if (frag2 == frag)
879 break;
880 frag2->sk = NULL;
881 frag2->destructor = NULL;
882 skb->truesize += frag2->truesize;
883 }
884 }
885
886slow_path:
887 /*
888 * Fragment the datagram.
889 */
890
891 ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
892 &state);
893
894 /*
895 * Keep copying data until we run out.
896 */
897
898 while (state.left > 0) {
899 bool first_frag = (state.offset == 0);
900
901 skb2 = ip_frag_next(skb, &state);
902 if (IS_ERR(skb2)) {
903 err = PTR_ERR(skb2);
904 goto fail;
905 }
906 ip_frag_ipcb(skb, skb2, first_frag, &state);
907
908 /*
909 * Put this fragment into the sending queue.
910 */
911 skb2->tstamp = tstamp;
912 err = output(net, sk, skb2);
913 if (err)
914 goto fail;
915
916 IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
917 }
918 consume_skb(skb);
919 IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
920 return err;
921
922fail:
923 kfree_skb(skb);
924 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
925 return err;
926}
927EXPORT_SYMBOL(ip_do_fragment);
928
929int
930ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
931{
932 struct msghdr *msg = from;
933
934 if (skb->ip_summed == CHECKSUM_PARTIAL) {
935 if (!copy_from_iter_full(to, len, &msg->msg_iter))
936 return -EFAULT;
937 } else {
938 __wsum csum = 0;
939 if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
940 return -EFAULT;
941 skb->csum = csum_block_add(skb->csum, csum, odd);
942 }
943 return 0;
944}
945EXPORT_SYMBOL(ip_generic_getfrag);
946
947static inline __wsum
948csum_page(struct page *page, int offset, int copy)
949{
950 char *kaddr;
951 __wsum csum;
952 kaddr = kmap(page);
953 csum = csum_partial(kaddr + offset, copy, 0);
954 kunmap(page);
955 return csum;
956}
957
958static int __ip_append_data(struct sock *sk,
959 struct flowi4 *fl4,
960 struct sk_buff_head *queue,
961 struct inet_cork *cork,
962 struct page_frag *pfrag,
963 int getfrag(void *from, char *to, int offset,
964 int len, int odd, struct sk_buff *skb),
965 void *from, int length, int transhdrlen,
966 unsigned int flags)
967{
968 struct inet_sock *inet = inet_sk(sk);
969 struct ubuf_info *uarg = NULL;
970 struct sk_buff *skb;
971
972 struct ip_options *opt = cork->opt;
973 int hh_len;
974 int exthdrlen;
975 int mtu;
976 int copy;
977 int err;
978 int offset = 0;
979 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
980 int csummode = CHECKSUM_NONE;
981 struct rtable *rt = (struct rtable *)cork->dst;
982 unsigned int wmem_alloc_delta = 0;
983 bool paged, extra_uref = false;
984 u32 tskey = 0;
985
986 skb = skb_peek_tail(queue);
987
988 exthdrlen = !skb ? rt->dst.header_len : 0;
989 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
990 paged = !!cork->gso_size;
991
992 if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
993 sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
994 tskey = sk->sk_tskey++;
995
996 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
997
998 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
999 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1000 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1001
1002 if (cork->length + length > maxnonfragsize - fragheaderlen) {
1003 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1004 mtu - (opt ? opt->optlen : 0));
1005 return -EMSGSIZE;
1006 }
1007
1008 /*
1009 * transhdrlen > 0 means that this is the first fragment and we wish
1010 * it won't be fragmented in the future.
1011 */
1012 if (transhdrlen &&
1013 length + fragheaderlen <= mtu &&
1014 rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1015 (!(flags & MSG_MORE) || cork->gso_size) &&
1016 (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1017 csummode = CHECKSUM_PARTIAL;
1018
1019 if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1020 uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
1021 if (!uarg)
1022 return -ENOBUFS;
1023 extra_uref = !skb_zcopy(skb); /* only ref on new uarg */
1024 if (rt->dst.dev->features & NETIF_F_SG &&
1025 csummode == CHECKSUM_PARTIAL) {
1026 paged = true;
1027 } else {
1028 uarg->zerocopy = 0;
1029 skb_zcopy_set(skb, uarg, &extra_uref);
1030 }
1031 }
1032
1033 cork->length += length;
1034
1035 /* So, what's going on in the loop below?
1036 *
1037 * We use calculated fragment length to generate chained skb,
1038 * each of segments is IP fragment ready for sending to network after
1039 * adding appropriate IP header.
1040 */
1041
1042 if (!skb)
1043 goto alloc_new_skb;
1044
1045 while (length > 0) {
1046 /* Check if the remaining data fits into current packet. */
1047 copy = mtu - skb->len;
1048 if (copy < length)
1049 copy = maxfraglen - skb->len;
1050 if (copy <= 0) {
1051 char *data;
1052 unsigned int datalen;
1053 unsigned int fraglen;
1054 unsigned int fraggap;
1055 unsigned int alloclen;
1056 unsigned int pagedlen;
1057 struct sk_buff *skb_prev;
1058alloc_new_skb:
1059 skb_prev = skb;
1060 if (skb_prev)
1061 fraggap = skb_prev->len - maxfraglen;
1062 else
1063 fraggap = 0;
1064
1065 /*
1066 * If remaining data exceeds the mtu,
1067 * we know we need more fragment(s).
1068 */
1069 datalen = length + fraggap;
1070 if (datalen > mtu - fragheaderlen)
1071 datalen = maxfraglen - fragheaderlen;
1072 fraglen = datalen + fragheaderlen;
1073 pagedlen = 0;
1074
1075 if ((flags & MSG_MORE) &&
1076 !(rt->dst.dev->features&NETIF_F_SG))
1077 alloclen = mtu;
1078 else if (!paged)
1079 alloclen = fraglen;
1080 else {
1081 alloclen = min_t(int, fraglen, MAX_HEADER);
1082 pagedlen = fraglen - alloclen;
1083 }
1084
1085 alloclen += exthdrlen;
1086
1087 /* The last fragment gets additional space at tail.
1088 * Note, with MSG_MORE we overallocate on fragments,
1089 * because we have no idea what fragment will be
1090 * the last.
1091 */
1092 if (datalen == length + fraggap)
1093 alloclen += rt->dst.trailer_len;
1094
1095 if (transhdrlen) {
1096 skb = sock_alloc_send_skb(sk,
1097 alloclen + hh_len + 15,
1098 (flags & MSG_DONTWAIT), &err);
1099 } else {
1100 skb = NULL;
1101 if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1102 2 * sk->sk_sndbuf)
1103 skb = alloc_skb(alloclen + hh_len + 15,
1104 sk->sk_allocation);
1105 if (unlikely(!skb))
1106 err = -ENOBUFS;
1107 }
1108 if (!skb)
1109 goto error;
1110
1111 /*
1112 * Fill in the control structures
1113 */
1114 skb->ip_summed = csummode;
1115 skb->csum = 0;
1116 skb_reserve(skb, hh_len);
1117
1118 /*
1119 * Find where to start putting bytes.
1120 */
1121 data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1122 skb_set_network_header(skb, exthdrlen);
1123 skb->transport_header = (skb->network_header +
1124 fragheaderlen);
1125 data += fragheaderlen + exthdrlen;
1126
1127 if (fraggap) {
1128 skb->csum = skb_copy_and_csum_bits(
1129 skb_prev, maxfraglen,
1130 data + transhdrlen, fraggap, 0);
1131 skb_prev->csum = csum_sub(skb_prev->csum,
1132 skb->csum);
1133 data += fraggap;
1134 pskb_trim_unique(skb_prev, maxfraglen);
1135 }
1136
1137 copy = datalen - transhdrlen - fraggap - pagedlen;
1138 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1139 err = -EFAULT;
1140 kfree_skb(skb);
1141 goto error;
1142 }
1143
1144 offset += copy;
1145 length -= copy + transhdrlen;
1146 transhdrlen = 0;
1147 exthdrlen = 0;
1148 csummode = CHECKSUM_NONE;
1149
1150 /* only the initial fragment is time stamped */
1151 skb_shinfo(skb)->tx_flags = cork->tx_flags;
1152 cork->tx_flags = 0;
1153 skb_shinfo(skb)->tskey = tskey;
1154 tskey = 0;
1155 skb_zcopy_set(skb, uarg, &extra_uref);
1156
1157 if ((flags & MSG_CONFIRM) && !skb_prev)
1158 skb_set_dst_pending_confirm(skb, 1);
1159
1160 /*
1161 * Put the packet on the pending queue.
1162 */
1163 if (!skb->destructor) {
1164 skb->destructor = sock_wfree;
1165 skb->sk = sk;
1166 wmem_alloc_delta += skb->truesize;
1167 }
1168 __skb_queue_tail(queue, skb);
1169 continue;
1170 }
1171
1172 if (copy > length)
1173 copy = length;
1174
1175 if (!(rt->dst.dev->features&NETIF_F_SG) &&
1176 skb_tailroom(skb) >= copy) {
1177 unsigned int off;
1178
1179 off = skb->len;
1180 if (getfrag(from, skb_put(skb, copy),
1181 offset, copy, off, skb) < 0) {
1182 __skb_trim(skb, off);
1183 err = -EFAULT;
1184 goto error;
1185 }
1186 } else if (!uarg || !uarg->zerocopy) {
1187 int i = skb_shinfo(skb)->nr_frags;
1188
1189 err = -ENOMEM;
1190 if (!sk_page_frag_refill(sk, pfrag))
1191 goto error;
1192
1193 if (!skb_can_coalesce(skb, i, pfrag->page,
1194 pfrag->offset)) {
1195 err = -EMSGSIZE;
1196 if (i == MAX_SKB_FRAGS)
1197 goto error;
1198
1199 __skb_fill_page_desc(skb, i, pfrag->page,
1200 pfrag->offset, 0);
1201 skb_shinfo(skb)->nr_frags = ++i;
1202 get_page(pfrag->page);
1203 }
1204 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1205 if (getfrag(from,
1206 page_address(pfrag->page) + pfrag->offset,
1207 offset, copy, skb->len, skb) < 0)
1208 goto error_efault;
1209
1210 pfrag->offset += copy;
1211 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1212 skb->len += copy;
1213 skb->data_len += copy;
1214 skb->truesize += copy;
1215 wmem_alloc_delta += copy;
1216 } else {
1217 err = skb_zerocopy_iter_dgram(skb, from, copy);
1218 if (err < 0)
1219 goto error;
1220 }
1221 offset += copy;
1222 length -= copy;
1223 }
1224
1225 if (wmem_alloc_delta)
1226 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1227 return 0;
1228
1229error_efault:
1230 err = -EFAULT;
1231error:
1232 if (uarg)
1233 sock_zerocopy_put_abort(uarg, extra_uref);
1234 cork->length -= length;
1235 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1236 refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1237 return err;
1238}
1239
1240static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1241 struct ipcm_cookie *ipc, struct rtable **rtp)
1242{
1243 struct ip_options_rcu *opt;
1244 struct rtable *rt;
1245
1246 rt = *rtp;
1247 if (unlikely(!rt))
1248 return -EFAULT;
1249
1250 /*
1251 * setup for corking.
1252 */
1253 opt = ipc->opt;
1254 if (opt) {
1255 if (!cork->opt) {
1256 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1257 sk->sk_allocation);
1258 if (unlikely(!cork->opt))
1259 return -ENOBUFS;
1260 }
1261 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1262 cork->flags |= IPCORK_OPT;
1263 cork->addr = ipc->addr;
1264 }
1265
1266 cork->fragsize = ip_sk_use_pmtu(sk) ?
1267 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1268
1269 if (!inetdev_valid_mtu(cork->fragsize))
1270 return -ENETUNREACH;
1271
1272 cork->gso_size = ipc->gso_size;
1273
1274 cork->dst = &rt->dst;
1275 /* We stole this route, caller should not release it. */
1276 *rtp = NULL;
1277
1278 cork->length = 0;
1279 cork->ttl = ipc->ttl;
1280 cork->tos = ipc->tos;
1281 cork->mark = ipc->sockc.mark;
1282 cork->priority = ipc->priority;
1283 cork->transmit_time = ipc->sockc.transmit_time;
1284 cork->tx_flags = 0;
1285 sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1286
1287 return 0;
1288}
1289
1290/*
1291 * ip_append_data() and ip_append_page() can make one large IP datagram
1292 * from many pieces of data. Each pieces will be holded on the socket
1293 * until ip_push_pending_frames() is called. Each piece can be a page
1294 * or non-page data.
1295 *
1296 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1297 * this interface potentially.
1298 *
1299 * LATER: length must be adjusted by pad at tail, when it is required.
1300 */
1301int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1302 int getfrag(void *from, char *to, int offset, int len,
1303 int odd, struct sk_buff *skb),
1304 void *from, int length, int transhdrlen,
1305 struct ipcm_cookie *ipc, struct rtable **rtp,
1306 unsigned int flags)
1307{
1308 struct inet_sock *inet = inet_sk(sk);
1309 int err;
1310
1311 if (flags&MSG_PROBE)
1312 return 0;
1313
1314 if (skb_queue_empty(&sk->sk_write_queue)) {
1315 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1316 if (err)
1317 return err;
1318 } else {
1319 transhdrlen = 0;
1320 }
1321
1322 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1323 sk_page_frag(sk), getfrag,
1324 from, length, transhdrlen, flags);
1325}
1326
1327ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1328 int offset, size_t size, int flags)
1329{
1330 struct inet_sock *inet = inet_sk(sk);
1331 struct sk_buff *skb;
1332 struct rtable *rt;
1333 struct ip_options *opt = NULL;
1334 struct inet_cork *cork;
1335 int hh_len;
1336 int mtu;
1337 int len;
1338 int err;
1339 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1340
1341 if (inet->hdrincl)
1342 return -EPERM;
1343
1344 if (flags&MSG_PROBE)
1345 return 0;
1346
1347 if (skb_queue_empty(&sk->sk_write_queue))
1348 return -EINVAL;
1349
1350 cork = &inet->cork.base;
1351 rt = (struct rtable *)cork->dst;
1352 if (cork->flags & IPCORK_OPT)
1353 opt = cork->opt;
1354
1355 if (!(rt->dst.dev->features&NETIF_F_SG))
1356 return -EOPNOTSUPP;
1357
1358 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1359 mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
1360
1361 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1362 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1363 maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1364
1365 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1366 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1367 mtu - (opt ? opt->optlen : 0));
1368 return -EMSGSIZE;
1369 }
1370
1371 skb = skb_peek_tail(&sk->sk_write_queue);
1372 if (!skb)
1373 return -EINVAL;
1374
1375 cork->length += size;
1376
1377 while (size > 0) {
1378 /* Check if the remaining data fits into current packet. */
1379 len = mtu - skb->len;
1380 if (len < size)
1381 len = maxfraglen - skb->len;
1382
1383 if (len <= 0) {
1384 struct sk_buff *skb_prev;
1385 int alloclen;
1386
1387 skb_prev = skb;
1388 fraggap = skb_prev->len - maxfraglen;
1389
1390 alloclen = fragheaderlen + hh_len + fraggap + 15;
1391 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1392 if (unlikely(!skb)) {
1393 err = -ENOBUFS;
1394 goto error;
1395 }
1396
1397 /*
1398 * Fill in the control structures
1399 */
1400 skb->ip_summed = CHECKSUM_NONE;
1401 skb->csum = 0;
1402 skb_reserve(skb, hh_len);
1403
1404 /*
1405 * Find where to start putting bytes.
1406 */
1407 skb_put(skb, fragheaderlen + fraggap);
1408 skb_reset_network_header(skb);
1409 skb->transport_header = (skb->network_header +
1410 fragheaderlen);
1411 if (fraggap) {
1412 skb->csum = skb_copy_and_csum_bits(skb_prev,
1413 maxfraglen,
1414 skb_transport_header(skb),
1415 fraggap, 0);
1416 skb_prev->csum = csum_sub(skb_prev->csum,
1417 skb->csum);
1418 pskb_trim_unique(skb_prev, maxfraglen);
1419 }
1420
1421 /*
1422 * Put the packet on the pending queue.
1423 */
1424 __skb_queue_tail(&sk->sk_write_queue, skb);
1425 continue;
1426 }
1427
1428 if (len > size)
1429 len = size;
1430
1431 if (skb_append_pagefrags(skb, page, offset, len)) {
1432 err = -EMSGSIZE;
1433 goto error;
1434 }
1435
1436 if (skb->ip_summed == CHECKSUM_NONE) {
1437 __wsum csum;
1438 csum = csum_page(page, offset, len);
1439 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1440 }
1441
1442 skb->len += len;
1443 skb->data_len += len;
1444 skb->truesize += len;
1445 refcount_add(len, &sk->sk_wmem_alloc);
1446 offset += len;
1447 size -= len;
1448 }
1449 return 0;
1450
1451error:
1452 cork->length -= size;
1453 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1454 return err;
1455}
1456
1457static void ip_cork_release(struct inet_cork *cork)
1458{
1459 cork->flags &= ~IPCORK_OPT;
1460 kfree(cork->opt);
1461 cork->opt = NULL;
1462 dst_release(cork->dst);
1463 cork->dst = NULL;
1464}
1465
1466/*
1467 * Combined all pending IP fragments on the socket as one IP datagram
1468 * and push them out.
1469 */
1470struct sk_buff *__ip_make_skb(struct sock *sk,
1471 struct flowi4 *fl4,
1472 struct sk_buff_head *queue,
1473 struct inet_cork *cork)
1474{
1475 struct sk_buff *skb, *tmp_skb;
1476 struct sk_buff **tail_skb;
1477 struct inet_sock *inet = inet_sk(sk);
1478 struct net *net = sock_net(sk);
1479 struct ip_options *opt = NULL;
1480 struct rtable *rt = (struct rtable *)cork->dst;
1481 struct iphdr *iph;
1482 __be16 df = 0;
1483 __u8 ttl;
1484
1485 skb = __skb_dequeue(queue);
1486 if (!skb)
1487 goto out;
1488 tail_skb = &(skb_shinfo(skb)->frag_list);
1489
1490 /* move skb->data to ip header from ext header */
1491 if (skb->data < skb_network_header(skb))
1492 __skb_pull(skb, skb_network_offset(skb));
1493 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1494 __skb_pull(tmp_skb, skb_network_header_len(skb));
1495 *tail_skb = tmp_skb;
1496 tail_skb = &(tmp_skb->next);
1497 skb->len += tmp_skb->len;
1498 skb->data_len += tmp_skb->len;
1499 skb->truesize += tmp_skb->truesize;
1500 tmp_skb->destructor = NULL;
1501 tmp_skb->sk = NULL;
1502 }
1503
1504 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1505 * to fragment the frame generated here. No matter, what transforms
1506 * how transforms change size of the packet, it will come out.
1507 */
1508 skb->ignore_df = ip_sk_ignore_df(sk);
1509
1510 /* DF bit is set when we want to see DF on outgoing frames.
1511 * If ignore_df is set too, we still allow to fragment this frame
1512 * locally. */
1513 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1514 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1515 (skb->len <= dst_mtu(&rt->dst) &&
1516 ip_dont_fragment(sk, &rt->dst)))
1517 df = htons(IP_DF);
1518
1519 if (cork->flags & IPCORK_OPT)
1520 opt = cork->opt;
1521
1522 if (cork->ttl != 0)
1523 ttl = cork->ttl;
1524 else if (rt->rt_type == RTN_MULTICAST)
1525 ttl = inet->mc_ttl;
1526 else
1527 ttl = ip_select_ttl(inet, &rt->dst);
1528
1529 iph = ip_hdr(skb);
1530 iph->version = 4;
1531 iph->ihl = 5;
1532 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1533 iph->frag_off = df;
1534 iph->ttl = ttl;
1535 iph->protocol = sk->sk_protocol;
1536 ip_copy_addrs(iph, fl4);
1537 ip_select_ident(net, skb, sk);
1538
1539 if (opt) {
1540 iph->ihl += opt->optlen>>2;
1541 ip_options_build(skb, opt, cork->addr, rt, 0);
1542 }
1543
1544 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1545 skb->mark = cork->mark;
1546 skb->tstamp = cork->transmit_time;
1547 /*
1548 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1549 * on dst refcount
1550 */
1551 cork->dst = NULL;
1552 skb_dst_set(skb, &rt->dst);
1553
1554 if (iph->protocol == IPPROTO_ICMP)
1555 icmp_out_count(net, ((struct icmphdr *)
1556 skb_transport_header(skb))->type);
1557
1558 ip_cork_release(cork);
1559out:
1560 return skb;
1561}
1562
1563int ip_send_skb(struct net *net, struct sk_buff *skb)
1564{
1565 int err;
1566
1567 err = ip_local_out(net, skb->sk, skb);
1568 if (err) {
1569 if (err > 0)
1570 err = net_xmit_errno(err);
1571 if (err)
1572 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1573 }
1574
1575 return err;
1576}
1577
1578int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1579{
1580 struct sk_buff *skb;
1581
1582 skb = ip_finish_skb(sk, fl4);
1583 if (!skb)
1584 return 0;
1585
1586 /* Netfilter gets whole the not fragmented skb. */
1587 return ip_send_skb(sock_net(sk), skb);
1588}
1589
1590/*
1591 * Throw away all pending data on the socket.
1592 */
1593static void __ip_flush_pending_frames(struct sock *sk,
1594 struct sk_buff_head *queue,
1595 struct inet_cork *cork)
1596{
1597 struct sk_buff *skb;
1598
1599 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1600 kfree_skb(skb);
1601
1602 ip_cork_release(cork);
1603}
1604
1605void ip_flush_pending_frames(struct sock *sk)
1606{
1607 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1608}
1609
1610struct sk_buff *ip_make_skb(struct sock *sk,
1611 struct flowi4 *fl4,
1612 int getfrag(void *from, char *to, int offset,
1613 int len, int odd, struct sk_buff *skb),
1614 void *from, int length, int transhdrlen,
1615 struct ipcm_cookie *ipc, struct rtable **rtp,
1616 struct inet_cork *cork, unsigned int flags)
1617{
1618 struct sk_buff_head queue;
1619 int err;
1620
1621 if (flags & MSG_PROBE)
1622 return NULL;
1623
1624 __skb_queue_head_init(&queue);
1625
1626 cork->flags = 0;
1627 cork->addr = 0;
1628 cork->opt = NULL;
1629 err = ip_setup_cork(sk, cork, ipc, rtp);
1630 if (err)
1631 return ERR_PTR(err);
1632
1633 err = __ip_append_data(sk, fl4, &queue, cork,
1634 ¤t->task_frag, getfrag,
1635 from, length, transhdrlen, flags);
1636 if (err) {
1637 __ip_flush_pending_frames(sk, &queue, cork);
1638 return ERR_PTR(err);
1639 }
1640
1641 return __ip_make_skb(sk, fl4, &queue, cork);
1642}
1643
1644/*
1645 * Fetch data from kernel space and fill in checksum if needed.
1646 */
1647static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1648 int len, int odd, struct sk_buff *skb)
1649{
1650 __wsum csum;
1651
1652 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1653 skb->csum = csum_block_add(skb->csum, csum, odd);
1654 return 0;
1655}
1656
1657/*
1658 * Generic function to send a packet as reply to another packet.
1659 * Used to send some TCP resets/acks so far.
1660 */
1661void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1662 const struct ip_options *sopt,
1663 __be32 daddr, __be32 saddr,
1664 const struct ip_reply_arg *arg,
1665 unsigned int len, u64 transmit_time)
1666{
1667 struct ip_options_data replyopts;
1668 struct ipcm_cookie ipc;
1669 struct flowi4 fl4;
1670 struct rtable *rt = skb_rtable(skb);
1671 struct net *net = sock_net(sk);
1672 struct sk_buff *nskb;
1673 int err;
1674 int oif;
1675
1676 if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1677 return;
1678
1679 ipcm_init(&ipc);
1680 ipc.addr = daddr;
1681 ipc.sockc.transmit_time = transmit_time;
1682
1683 if (replyopts.opt.opt.optlen) {
1684 ipc.opt = &replyopts.opt;
1685
1686 if (replyopts.opt.opt.srr)
1687 daddr = replyopts.opt.opt.faddr;
1688 }
1689
1690 oif = arg->bound_dev_if;
1691 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1692 oif = skb->skb_iif;
1693
1694 flowi4_init_output(&fl4, oif,
1695 IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1696 RT_TOS(arg->tos),
1697 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1698 ip_reply_arg_flowi_flags(arg),
1699 daddr, saddr,
1700 tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1701 arg->uid);
1702 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1703 rt = ip_route_output_key(net, &fl4);
1704 if (IS_ERR(rt))
1705 return;
1706
1707 inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1708
1709 sk->sk_protocol = ip_hdr(skb)->protocol;
1710 sk->sk_bound_dev_if = arg->bound_dev_if;
1711 sk->sk_sndbuf = sysctl_wmem_default;
1712 ipc.sockc.mark = fl4.flowi4_mark;
1713 err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1714 len, 0, &ipc, &rt, MSG_DONTWAIT);
1715 if (unlikely(err)) {
1716 ip_flush_pending_frames(sk);
1717 goto out;
1718 }
1719
1720 nskb = skb_peek(&sk->sk_write_queue);
1721 if (nskb) {
1722 if (arg->csumoffset >= 0)
1723 *((__sum16 *)skb_transport_header(nskb) +
1724 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1725 arg->csum));
1726 nskb->ip_summed = CHECKSUM_NONE;
1727 ip_push_pending_frames(sk, &fl4);
1728 }
1729out:
1730 ip_rt_put(rt);
1731}
1732
1733void __init ip_init(void)
1734{
1735 ip_rt_init();
1736 inet_initpeers();
1737
1738#if defined(CONFIG_IP_MULTICAST)
1739 igmp_mc_init();
1740#endif
1741}
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 <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 <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
83int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84EXPORT_SYMBOL(sysctl_ip_default_ttl);
85
86/* Generate a checksum for an outgoing IP datagram. */
87void ip_send_check(struct iphdr *iph)
88{
89 iph->check = 0;
90 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
91}
92EXPORT_SYMBOL(ip_send_check);
93
94int __ip_local_out(struct sk_buff *skb)
95{
96 struct iphdr *iph = ip_hdr(skb);
97
98 iph->tot_len = htons(skb->len);
99 ip_send_check(iph);
100 return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
101 skb_dst(skb)->dev, dst_output);
102}
103
104int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
105{
106 int err;
107
108 err = __ip_local_out(skb);
109 if (likely(err == 1))
110 err = dst_output_sk(sk, skb);
111
112 return err;
113}
114EXPORT_SYMBOL_GPL(ip_local_out_sk);
115
116static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
117{
118 int ttl = inet->uc_ttl;
119
120 if (ttl < 0)
121 ttl = ip4_dst_hoplimit(dst);
122 return ttl;
123}
124
125/*
126 * Add an ip header to a skbuff and send it out.
127 *
128 */
129int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
130 __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
131{
132 struct inet_sock *inet = inet_sk(sk);
133 struct rtable *rt = skb_rtable(skb);
134 struct iphdr *iph;
135
136 /* Build the IP header. */
137 skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
138 skb_reset_network_header(skb);
139 iph = ip_hdr(skb);
140 iph->version = 4;
141 iph->ihl = 5;
142 iph->tos = inet->tos;
143 if (ip_dont_fragment(sk, &rt->dst))
144 iph->frag_off = htons(IP_DF);
145 else
146 iph->frag_off = 0;
147 iph->ttl = ip_select_ttl(inet, &rt->dst);
148 iph->daddr = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
149 iph->saddr = saddr;
150 iph->protocol = sk->sk_protocol;
151 ip_select_ident(skb, &rt->dst, sk);
152
153 if (opt && opt->opt.optlen) {
154 iph->ihl += opt->opt.optlen>>2;
155 ip_options_build(skb, &opt->opt, daddr, rt, 0);
156 }
157
158 skb->priority = sk->sk_priority;
159 skb->mark = sk->sk_mark;
160
161 /* Send it out. */
162 return ip_local_out(skb);
163}
164EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
165
166static inline int ip_finish_output2(struct sk_buff *skb)
167{
168 struct dst_entry *dst = skb_dst(skb);
169 struct rtable *rt = (struct rtable *)dst;
170 struct net_device *dev = dst->dev;
171 unsigned int hh_len = LL_RESERVED_SPACE(dev);
172 struct neighbour *neigh;
173 u32 nexthop;
174
175 if (rt->rt_type == RTN_MULTICAST) {
176 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTMCAST, skb->len);
177 } else if (rt->rt_type == RTN_BROADCAST)
178 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUTBCAST, skb->len);
179
180 /* Be paranoid, rather than too clever. */
181 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
182 struct sk_buff *skb2;
183
184 skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
185 if (skb2 == NULL) {
186 kfree_skb(skb);
187 return -ENOMEM;
188 }
189 if (skb->sk)
190 skb_set_owner_w(skb2, skb->sk);
191 consume_skb(skb);
192 skb = skb2;
193 }
194
195 rcu_read_lock_bh();
196 nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
197 neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
198 if (unlikely(!neigh))
199 neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
200 if (!IS_ERR(neigh)) {
201 int res = dst_neigh_output(dst, neigh, skb);
202
203 rcu_read_unlock_bh();
204 return res;
205 }
206 rcu_read_unlock_bh();
207
208 net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
209 __func__);
210 kfree_skb(skb);
211 return -EINVAL;
212}
213
214static int ip_finish_output_gso(struct sk_buff *skb)
215{
216 netdev_features_t features;
217 struct sk_buff *segs;
218 int ret = 0;
219
220 /* common case: locally created skb or seglen is <= mtu */
221 if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
222 skb_gso_network_seglen(skb) <= ip_skb_dst_mtu(skb))
223 return ip_finish_output2(skb);
224
225 /* Slowpath - GSO segment length is exceeding the dst MTU.
226 *
227 * This can happen in two cases:
228 * 1) TCP GRO packet, DF bit not set
229 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
230 * from host network stack.
231 */
232 features = netif_skb_features(skb);
233 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
234 if (IS_ERR(segs)) {
235 kfree_skb(skb);
236 return -ENOMEM;
237 }
238
239 consume_skb(skb);
240
241 do {
242 struct sk_buff *nskb = segs->next;
243 int err;
244
245 segs->next = NULL;
246 err = ip_fragment(segs, ip_finish_output2);
247
248 if (err && ret == 0)
249 ret = err;
250 segs = nskb;
251 } while (segs);
252
253 return ret;
254}
255
256static int ip_finish_output(struct sk_buff *skb)
257{
258#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
259 /* Policy lookup after SNAT yielded a new policy */
260 if (skb_dst(skb)->xfrm != NULL) {
261 IPCB(skb)->flags |= IPSKB_REROUTED;
262 return dst_output(skb);
263 }
264#endif
265 if (skb_is_gso(skb))
266 return ip_finish_output_gso(skb);
267
268 if (skb->len > ip_skb_dst_mtu(skb))
269 return ip_fragment(skb, ip_finish_output2);
270
271 return ip_finish_output2(skb);
272}
273
274int ip_mc_output(struct sock *sk, struct sk_buff *skb)
275{
276 struct rtable *rt = skb_rtable(skb);
277 struct net_device *dev = rt->dst.dev;
278
279 /*
280 * If the indicated interface is up and running, send the packet.
281 */
282 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
283
284 skb->dev = dev;
285 skb->protocol = htons(ETH_P_IP);
286
287 /*
288 * Multicasts are looped back for other local users
289 */
290
291 if (rt->rt_flags&RTCF_MULTICAST) {
292 if (sk_mc_loop(sk)
293#ifdef CONFIG_IP_MROUTE
294 /* Small optimization: do not loopback not local frames,
295 which returned after forwarding; they will be dropped
296 by ip_mr_input in any case.
297 Note, that local frames are looped back to be delivered
298 to local recipients.
299
300 This check is duplicated in ip_mr_input at the moment.
301 */
302 &&
303 ((rt->rt_flags & RTCF_LOCAL) ||
304 !(IPCB(skb)->flags & IPSKB_FORWARDED))
305#endif
306 ) {
307 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
308 if (newskb)
309 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
310 newskb, NULL, newskb->dev,
311 dev_loopback_xmit);
312 }
313
314 /* Multicasts with ttl 0 must not go beyond the host */
315
316 if (ip_hdr(skb)->ttl == 0) {
317 kfree_skb(skb);
318 return 0;
319 }
320 }
321
322 if (rt->rt_flags&RTCF_BROADCAST) {
323 struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
324 if (newskb)
325 NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING, newskb,
326 NULL, newskb->dev, dev_loopback_xmit);
327 }
328
329 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL,
330 skb->dev, ip_finish_output,
331 !(IPCB(skb)->flags & IPSKB_REROUTED));
332}
333
334int ip_output(struct sock *sk, struct sk_buff *skb)
335{
336 struct net_device *dev = skb_dst(skb)->dev;
337
338 IP_UPD_PO_STATS(dev_net(dev), IPSTATS_MIB_OUT, skb->len);
339
340 skb->dev = dev;
341 skb->protocol = htons(ETH_P_IP);
342
343 return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING, skb, NULL, dev,
344 ip_finish_output,
345 !(IPCB(skb)->flags & IPSKB_REROUTED));
346}
347
348/*
349 * copy saddr and daddr, possibly using 64bit load/stores
350 * Equivalent to :
351 * iph->saddr = fl4->saddr;
352 * iph->daddr = fl4->daddr;
353 */
354static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
355{
356 BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
357 offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
358 memcpy(&iph->saddr, &fl4->saddr,
359 sizeof(fl4->saddr) + sizeof(fl4->daddr));
360}
361
362/* Note: skb->sk can be different from sk, in case of tunnels */
363int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
364{
365 struct inet_sock *inet = inet_sk(sk);
366 struct ip_options_rcu *inet_opt;
367 struct flowi4 *fl4;
368 struct rtable *rt;
369 struct iphdr *iph;
370 int res;
371
372 /* Skip all of this if the packet is already routed,
373 * f.e. by something like SCTP.
374 */
375 rcu_read_lock();
376 inet_opt = rcu_dereference(inet->inet_opt);
377 fl4 = &fl->u.ip4;
378 rt = skb_rtable(skb);
379 if (rt != NULL)
380 goto packet_routed;
381
382 /* Make sure we can route this packet. */
383 rt = (struct rtable *)__sk_dst_check(sk, 0);
384 if (rt == NULL) {
385 __be32 daddr;
386
387 /* Use correct destination address if we have options. */
388 daddr = inet->inet_daddr;
389 if (inet_opt && inet_opt->opt.srr)
390 daddr = inet_opt->opt.faddr;
391
392 /* If this fails, retransmit mechanism of transport layer will
393 * keep trying until route appears or the connection times
394 * itself out.
395 */
396 rt = ip_route_output_ports(sock_net(sk), fl4, sk,
397 daddr, inet->inet_saddr,
398 inet->inet_dport,
399 inet->inet_sport,
400 sk->sk_protocol,
401 RT_CONN_FLAGS(sk),
402 sk->sk_bound_dev_if);
403 if (IS_ERR(rt))
404 goto no_route;
405 sk_setup_caps(sk, &rt->dst);
406 }
407 skb_dst_set_noref(skb, &rt->dst);
408
409packet_routed:
410 if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
411 goto no_route;
412
413 /* OK, we know where to send it, allocate and build IP header. */
414 skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
415 skb_reset_network_header(skb);
416 iph = ip_hdr(skb);
417 *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
418 if (ip_dont_fragment(sk, &rt->dst) && !skb->local_df)
419 iph->frag_off = htons(IP_DF);
420 else
421 iph->frag_off = 0;
422 iph->ttl = ip_select_ttl(inet, &rt->dst);
423 iph->protocol = sk->sk_protocol;
424 ip_copy_addrs(iph, fl4);
425
426 /* Transport layer set skb->h.foo itself. */
427
428 if (inet_opt && inet_opt->opt.optlen) {
429 iph->ihl += inet_opt->opt.optlen >> 2;
430 ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
431 }
432
433 ip_select_ident_more(skb, &rt->dst, sk,
434 (skb_shinfo(skb)->gso_segs ?: 1) - 1);
435
436 /* TODO : should we use skb->sk here instead of sk ? */
437 skb->priority = sk->sk_priority;
438 skb->mark = sk->sk_mark;
439
440 res = ip_local_out(skb);
441 rcu_read_unlock();
442 return res;
443
444no_route:
445 rcu_read_unlock();
446 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
447 kfree_skb(skb);
448 return -EHOSTUNREACH;
449}
450EXPORT_SYMBOL(ip_queue_xmit);
451
452
453static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
454{
455 to->pkt_type = from->pkt_type;
456 to->priority = from->priority;
457 to->protocol = from->protocol;
458 skb_dst_drop(to);
459 skb_dst_copy(to, from);
460 to->dev = from->dev;
461 to->mark = from->mark;
462
463 /* Copy the flags to each fragment. */
464 IPCB(to)->flags = IPCB(from)->flags;
465
466#ifdef CONFIG_NET_SCHED
467 to->tc_index = from->tc_index;
468#endif
469 nf_copy(to, from);
470#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
471 to->ipvs_property = from->ipvs_property;
472#endif
473 skb_copy_secmark(to, from);
474}
475
476/*
477 * This IP datagram is too large to be sent in one piece. Break it up into
478 * smaller pieces (each of size equal to IP header plus
479 * a block of the data of the original IP data part) that will yet fit in a
480 * single device frame, and queue such a frame for sending.
481 */
482
483int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *))
484{
485 struct iphdr *iph;
486 int ptr;
487 struct net_device *dev;
488 struct sk_buff *skb2;
489 unsigned int mtu, hlen, left, len, ll_rs;
490 int offset;
491 __be16 not_last_frag;
492 struct rtable *rt = skb_rtable(skb);
493 int err = 0;
494
495 dev = rt->dst.dev;
496
497 /*
498 * Point into the IP datagram header.
499 */
500
501 iph = ip_hdr(skb);
502
503 mtu = ip_skb_dst_mtu(skb);
504 if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->local_df) ||
505 (IPCB(skb)->frag_max_size &&
506 IPCB(skb)->frag_max_size > mtu))) {
507 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
508 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
509 htonl(mtu));
510 kfree_skb(skb);
511 return -EMSGSIZE;
512 }
513
514 /*
515 * Setup starting values.
516 */
517
518 hlen = iph->ihl * 4;
519 mtu = mtu - hlen; /* Size of data space */
520#ifdef CONFIG_BRIDGE_NETFILTER
521 if (skb->nf_bridge)
522 mtu -= nf_bridge_mtu_reduction(skb);
523#endif
524 IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
525
526 /* When frag_list is given, use it. First, check its validity:
527 * some transformers could create wrong frag_list or break existing
528 * one, it is not prohibited. In this case fall back to copying.
529 *
530 * LATER: this step can be merged to real generation of fragments,
531 * we can switch to copy when see the first bad fragment.
532 */
533 if (skb_has_frag_list(skb)) {
534 struct sk_buff *frag, *frag2;
535 int first_len = skb_pagelen(skb);
536
537 if (first_len - hlen > mtu ||
538 ((first_len - hlen) & 7) ||
539 ip_is_fragment(iph) ||
540 skb_cloned(skb))
541 goto slow_path;
542
543 skb_walk_frags(skb, frag) {
544 /* Correct geometry. */
545 if (frag->len > mtu ||
546 ((frag->len & 7) && frag->next) ||
547 skb_headroom(frag) < hlen)
548 goto slow_path_clean;
549
550 /* Partially cloned skb? */
551 if (skb_shared(frag))
552 goto slow_path_clean;
553
554 BUG_ON(frag->sk);
555 if (skb->sk) {
556 frag->sk = skb->sk;
557 frag->destructor = sock_wfree;
558 }
559 skb->truesize -= frag->truesize;
560 }
561
562 /* Everything is OK. Generate! */
563
564 err = 0;
565 offset = 0;
566 frag = skb_shinfo(skb)->frag_list;
567 skb_frag_list_init(skb);
568 skb->data_len = first_len - skb_headlen(skb);
569 skb->len = first_len;
570 iph->tot_len = htons(first_len);
571 iph->frag_off = htons(IP_MF);
572 ip_send_check(iph);
573
574 for (;;) {
575 /* Prepare header of the next frame,
576 * before previous one went down. */
577 if (frag) {
578 frag->ip_summed = CHECKSUM_NONE;
579 skb_reset_transport_header(frag);
580 __skb_push(frag, hlen);
581 skb_reset_network_header(frag);
582 memcpy(skb_network_header(frag), iph, hlen);
583 iph = ip_hdr(frag);
584 iph->tot_len = htons(frag->len);
585 ip_copy_metadata(frag, skb);
586 if (offset == 0)
587 ip_options_fragment(frag);
588 offset += skb->len - hlen;
589 iph->frag_off = htons(offset>>3);
590 if (frag->next != NULL)
591 iph->frag_off |= htons(IP_MF);
592 /* Ready, complete checksum */
593 ip_send_check(iph);
594 }
595
596 err = output(skb);
597
598 if (!err)
599 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
600 if (err || !frag)
601 break;
602
603 skb = frag;
604 frag = skb->next;
605 skb->next = NULL;
606 }
607
608 if (err == 0) {
609 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
610 return 0;
611 }
612
613 while (frag) {
614 skb = frag->next;
615 kfree_skb(frag);
616 frag = skb;
617 }
618 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
619 return err;
620
621slow_path_clean:
622 skb_walk_frags(skb, frag2) {
623 if (frag2 == frag)
624 break;
625 frag2->sk = NULL;
626 frag2->destructor = NULL;
627 skb->truesize += frag2->truesize;
628 }
629 }
630
631slow_path:
632 /* for offloaded checksums cleanup checksum before fragmentation */
633 if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
634 goto fail;
635 iph = ip_hdr(skb);
636
637 left = skb->len - hlen; /* Space per frame */
638 ptr = hlen; /* Where to start from */
639
640 /* for bridged IP traffic encapsulated inside f.e. a vlan header,
641 * we need to make room for the encapsulating header
642 */
643 ll_rs = LL_RESERVED_SPACE_EXTRA(rt->dst.dev, nf_bridge_pad(skb));
644
645 /*
646 * Fragment the datagram.
647 */
648
649 offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
650 not_last_frag = iph->frag_off & htons(IP_MF);
651
652 /*
653 * Keep copying data until we run out.
654 */
655
656 while (left > 0) {
657 len = left;
658 /* IF: it doesn't fit, use 'mtu' - the data space left */
659 if (len > mtu)
660 len = mtu;
661 /* IF: we are not sending up to and including the packet end
662 then align the next start on an eight byte boundary */
663 if (len < left) {
664 len &= ~7;
665 }
666 /*
667 * Allocate buffer.
668 */
669
670 if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
671 NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
672 err = -ENOMEM;
673 goto fail;
674 }
675
676 /*
677 * Set up data on packet
678 */
679
680 ip_copy_metadata(skb2, skb);
681 skb_reserve(skb2, ll_rs);
682 skb_put(skb2, len + hlen);
683 skb_reset_network_header(skb2);
684 skb2->transport_header = skb2->network_header + hlen;
685
686 /*
687 * Charge the memory for the fragment to any owner
688 * it might possess
689 */
690
691 if (skb->sk)
692 skb_set_owner_w(skb2, skb->sk);
693
694 /*
695 * Copy the packet header into the new buffer.
696 */
697
698 skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
699
700 /*
701 * Copy a block of the IP datagram.
702 */
703 if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
704 BUG();
705 left -= len;
706
707 /*
708 * Fill in the new header fields.
709 */
710 iph = ip_hdr(skb2);
711 iph->frag_off = htons((offset >> 3));
712
713 /* ANK: dirty, but effective trick. Upgrade options only if
714 * the segment to be fragmented was THE FIRST (otherwise,
715 * options are already fixed) and make it ONCE
716 * on the initial skb, so that all the following fragments
717 * will inherit fixed options.
718 */
719 if (offset == 0)
720 ip_options_fragment(skb);
721
722 /*
723 * Added AC : If we are fragmenting a fragment that's not the
724 * last fragment then keep MF on each bit
725 */
726 if (left > 0 || not_last_frag)
727 iph->frag_off |= htons(IP_MF);
728 ptr += len;
729 offset += len;
730
731 /*
732 * Put this fragment into the sending queue.
733 */
734 iph->tot_len = htons(len + hlen);
735
736 ip_send_check(iph);
737
738 err = output(skb2);
739 if (err)
740 goto fail;
741
742 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGCREATES);
743 }
744 consume_skb(skb);
745 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGOKS);
746 return err;
747
748fail:
749 kfree_skb(skb);
750 IP_INC_STATS(dev_net(dev), IPSTATS_MIB_FRAGFAILS);
751 return err;
752}
753EXPORT_SYMBOL(ip_fragment);
754
755int
756ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
757{
758 struct iovec *iov = from;
759
760 if (skb->ip_summed == CHECKSUM_PARTIAL) {
761 if (memcpy_fromiovecend(to, iov, offset, len) < 0)
762 return -EFAULT;
763 } else {
764 __wsum csum = 0;
765 if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
766 return -EFAULT;
767 skb->csum = csum_block_add(skb->csum, csum, odd);
768 }
769 return 0;
770}
771EXPORT_SYMBOL(ip_generic_getfrag);
772
773static inline __wsum
774csum_page(struct page *page, int offset, int copy)
775{
776 char *kaddr;
777 __wsum csum;
778 kaddr = kmap(page);
779 csum = csum_partial(kaddr + offset, copy, 0);
780 kunmap(page);
781 return csum;
782}
783
784static inline int ip_ufo_append_data(struct sock *sk,
785 struct sk_buff_head *queue,
786 int getfrag(void *from, char *to, int offset, int len,
787 int odd, struct sk_buff *skb),
788 void *from, int length, int hh_len, int fragheaderlen,
789 int transhdrlen, int maxfraglen, unsigned int flags)
790{
791 struct sk_buff *skb;
792 int err;
793
794 /* There is support for UDP fragmentation offload by network
795 * device, so create one single skb packet containing complete
796 * udp datagram
797 */
798 if ((skb = skb_peek_tail(queue)) == NULL) {
799 skb = sock_alloc_send_skb(sk,
800 hh_len + fragheaderlen + transhdrlen + 20,
801 (flags & MSG_DONTWAIT), &err);
802
803 if (skb == NULL)
804 return err;
805
806 /* reserve space for Hardware header */
807 skb_reserve(skb, hh_len);
808
809 /* create space for UDP/IP header */
810 skb_put(skb, fragheaderlen + transhdrlen);
811
812 /* initialize network header pointer */
813 skb_reset_network_header(skb);
814
815 /* initialize protocol header pointer */
816 skb->transport_header = skb->network_header + fragheaderlen;
817
818 skb->csum = 0;
819
820
821 __skb_queue_tail(queue, skb);
822 } else if (skb_is_gso(skb)) {
823 goto append;
824 }
825
826 skb->ip_summed = CHECKSUM_PARTIAL;
827 /* specify the length of each IP datagram fragment */
828 skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
829 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
830
831append:
832 return skb_append_datato_frags(sk, skb, getfrag, from,
833 (length - transhdrlen));
834}
835
836static int __ip_append_data(struct sock *sk,
837 struct flowi4 *fl4,
838 struct sk_buff_head *queue,
839 struct inet_cork *cork,
840 struct page_frag *pfrag,
841 int getfrag(void *from, char *to, int offset,
842 int len, int odd, struct sk_buff *skb),
843 void *from, int length, int transhdrlen,
844 unsigned int flags)
845{
846 struct inet_sock *inet = inet_sk(sk);
847 struct sk_buff *skb;
848
849 struct ip_options *opt = cork->opt;
850 int hh_len;
851 int exthdrlen;
852 int mtu;
853 int copy;
854 int err;
855 int offset = 0;
856 unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
857 int csummode = CHECKSUM_NONE;
858 struct rtable *rt = (struct rtable *)cork->dst;
859
860 skb = skb_peek_tail(queue);
861
862 exthdrlen = !skb ? rt->dst.header_len : 0;
863 mtu = cork->fragsize;
864
865 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
866
867 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
868 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
869 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu;
870
871 if (cork->length + length > maxnonfragsize - fragheaderlen) {
872 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
873 mtu - (opt ? opt->optlen : 0));
874 return -EMSGSIZE;
875 }
876
877 /*
878 * transhdrlen > 0 means that this is the first fragment and we wish
879 * it won't be fragmented in the future.
880 */
881 if (transhdrlen &&
882 length + fragheaderlen <= mtu &&
883 rt->dst.dev->features & NETIF_F_V4_CSUM &&
884 !exthdrlen)
885 csummode = CHECKSUM_PARTIAL;
886
887 cork->length += length;
888 if (((length > mtu) || (skb && skb_is_gso(skb))) &&
889 (sk->sk_protocol == IPPROTO_UDP) &&
890 (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len) {
891 err = ip_ufo_append_data(sk, queue, getfrag, from, length,
892 hh_len, fragheaderlen, transhdrlen,
893 maxfraglen, flags);
894 if (err)
895 goto error;
896 return 0;
897 }
898
899 /* So, what's going on in the loop below?
900 *
901 * We use calculated fragment length to generate chained skb,
902 * each of segments is IP fragment ready for sending to network after
903 * adding appropriate IP header.
904 */
905
906 if (!skb)
907 goto alloc_new_skb;
908
909 while (length > 0) {
910 /* Check if the remaining data fits into current packet. */
911 copy = mtu - skb->len;
912 if (copy < length)
913 copy = maxfraglen - skb->len;
914 if (copy <= 0) {
915 char *data;
916 unsigned int datalen;
917 unsigned int fraglen;
918 unsigned int fraggap;
919 unsigned int alloclen;
920 struct sk_buff *skb_prev;
921alloc_new_skb:
922 skb_prev = skb;
923 if (skb_prev)
924 fraggap = skb_prev->len - maxfraglen;
925 else
926 fraggap = 0;
927
928 /*
929 * If remaining data exceeds the mtu,
930 * we know we need more fragment(s).
931 */
932 datalen = length + fraggap;
933 if (datalen > mtu - fragheaderlen)
934 datalen = maxfraglen - fragheaderlen;
935 fraglen = datalen + fragheaderlen;
936
937 if ((flags & MSG_MORE) &&
938 !(rt->dst.dev->features&NETIF_F_SG))
939 alloclen = mtu;
940 else
941 alloclen = fraglen;
942
943 alloclen += exthdrlen;
944
945 /* The last fragment gets additional space at tail.
946 * Note, with MSG_MORE we overallocate on fragments,
947 * because we have no idea what fragment will be
948 * the last.
949 */
950 if (datalen == length + fraggap)
951 alloclen += rt->dst.trailer_len;
952
953 if (transhdrlen) {
954 skb = sock_alloc_send_skb(sk,
955 alloclen + hh_len + 15,
956 (flags & MSG_DONTWAIT), &err);
957 } else {
958 skb = NULL;
959 if (atomic_read(&sk->sk_wmem_alloc) <=
960 2 * sk->sk_sndbuf)
961 skb = sock_wmalloc(sk,
962 alloclen + hh_len + 15, 1,
963 sk->sk_allocation);
964 if (unlikely(skb == NULL))
965 err = -ENOBUFS;
966 else
967 /* only the initial fragment is
968 time stamped */
969 cork->tx_flags = 0;
970 }
971 if (skb == NULL)
972 goto error;
973
974 /*
975 * Fill in the control structures
976 */
977 skb->ip_summed = csummode;
978 skb->csum = 0;
979 skb_reserve(skb, hh_len);
980 skb_shinfo(skb)->tx_flags = cork->tx_flags;
981
982 /*
983 * Find where to start putting bytes.
984 */
985 data = skb_put(skb, fraglen + exthdrlen);
986 skb_set_network_header(skb, exthdrlen);
987 skb->transport_header = (skb->network_header +
988 fragheaderlen);
989 data += fragheaderlen + exthdrlen;
990
991 if (fraggap) {
992 skb->csum = skb_copy_and_csum_bits(
993 skb_prev, maxfraglen,
994 data + transhdrlen, fraggap, 0);
995 skb_prev->csum = csum_sub(skb_prev->csum,
996 skb->csum);
997 data += fraggap;
998 pskb_trim_unique(skb_prev, maxfraglen);
999 }
1000
1001 copy = datalen - transhdrlen - fraggap;
1002 if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1003 err = -EFAULT;
1004 kfree_skb(skb);
1005 goto error;
1006 }
1007
1008 offset += copy;
1009 length -= datalen - fraggap;
1010 transhdrlen = 0;
1011 exthdrlen = 0;
1012 csummode = CHECKSUM_NONE;
1013
1014 /*
1015 * Put the packet on the pending queue.
1016 */
1017 __skb_queue_tail(queue, skb);
1018 continue;
1019 }
1020
1021 if (copy > length)
1022 copy = length;
1023
1024 if (!(rt->dst.dev->features&NETIF_F_SG)) {
1025 unsigned int off;
1026
1027 off = skb->len;
1028 if (getfrag(from, skb_put(skb, copy),
1029 offset, copy, off, skb) < 0) {
1030 __skb_trim(skb, off);
1031 err = -EFAULT;
1032 goto error;
1033 }
1034 } else {
1035 int i = skb_shinfo(skb)->nr_frags;
1036
1037 err = -ENOMEM;
1038 if (!sk_page_frag_refill(sk, pfrag))
1039 goto error;
1040
1041 if (!skb_can_coalesce(skb, i, pfrag->page,
1042 pfrag->offset)) {
1043 err = -EMSGSIZE;
1044 if (i == MAX_SKB_FRAGS)
1045 goto error;
1046
1047 __skb_fill_page_desc(skb, i, pfrag->page,
1048 pfrag->offset, 0);
1049 skb_shinfo(skb)->nr_frags = ++i;
1050 get_page(pfrag->page);
1051 }
1052 copy = min_t(int, copy, pfrag->size - pfrag->offset);
1053 if (getfrag(from,
1054 page_address(pfrag->page) + pfrag->offset,
1055 offset, copy, skb->len, skb) < 0)
1056 goto error_efault;
1057
1058 pfrag->offset += copy;
1059 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1060 skb->len += copy;
1061 skb->data_len += copy;
1062 skb->truesize += copy;
1063 atomic_add(copy, &sk->sk_wmem_alloc);
1064 }
1065 offset += copy;
1066 length -= copy;
1067 }
1068
1069 return 0;
1070
1071error_efault:
1072 err = -EFAULT;
1073error:
1074 cork->length -= length;
1075 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1076 return err;
1077}
1078
1079static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1080 struct ipcm_cookie *ipc, struct rtable **rtp)
1081{
1082 struct ip_options_rcu *opt;
1083 struct rtable *rt;
1084
1085 /*
1086 * setup for corking.
1087 */
1088 opt = ipc->opt;
1089 if (opt) {
1090 if (cork->opt == NULL) {
1091 cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1092 sk->sk_allocation);
1093 if (unlikely(cork->opt == NULL))
1094 return -ENOBUFS;
1095 }
1096 memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1097 cork->flags |= IPCORK_OPT;
1098 cork->addr = ipc->addr;
1099 }
1100 rt = *rtp;
1101 if (unlikely(!rt))
1102 return -EFAULT;
1103 /*
1104 * We steal reference to this route, caller should not release it
1105 */
1106 *rtp = NULL;
1107 cork->fragsize = ip_sk_use_pmtu(sk) ?
1108 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1109 cork->dst = &rt->dst;
1110 cork->length = 0;
1111 cork->ttl = ipc->ttl;
1112 cork->tos = ipc->tos;
1113 cork->priority = ipc->priority;
1114 cork->tx_flags = ipc->tx_flags;
1115
1116 return 0;
1117}
1118
1119/*
1120 * ip_append_data() and ip_append_page() can make one large IP datagram
1121 * from many pieces of data. Each pieces will be holded on the socket
1122 * until ip_push_pending_frames() is called. Each piece can be a page
1123 * or non-page data.
1124 *
1125 * Not only UDP, other transport protocols - e.g. raw sockets - can use
1126 * this interface potentially.
1127 *
1128 * LATER: length must be adjusted by pad at tail, when it is required.
1129 */
1130int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1131 int getfrag(void *from, char *to, int offset, int len,
1132 int odd, struct sk_buff *skb),
1133 void *from, int length, int transhdrlen,
1134 struct ipcm_cookie *ipc, struct rtable **rtp,
1135 unsigned int flags)
1136{
1137 struct inet_sock *inet = inet_sk(sk);
1138 int err;
1139
1140 if (flags&MSG_PROBE)
1141 return 0;
1142
1143 if (skb_queue_empty(&sk->sk_write_queue)) {
1144 err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1145 if (err)
1146 return err;
1147 } else {
1148 transhdrlen = 0;
1149 }
1150
1151 return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1152 sk_page_frag(sk), getfrag,
1153 from, length, transhdrlen, flags);
1154}
1155
1156ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1157 int offset, size_t size, int flags)
1158{
1159 struct inet_sock *inet = inet_sk(sk);
1160 struct sk_buff *skb;
1161 struct rtable *rt;
1162 struct ip_options *opt = NULL;
1163 struct inet_cork *cork;
1164 int hh_len;
1165 int mtu;
1166 int len;
1167 int err;
1168 unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1169
1170 if (inet->hdrincl)
1171 return -EPERM;
1172
1173 if (flags&MSG_PROBE)
1174 return 0;
1175
1176 if (skb_queue_empty(&sk->sk_write_queue))
1177 return -EINVAL;
1178
1179 cork = &inet->cork.base;
1180 rt = (struct rtable *)cork->dst;
1181 if (cork->flags & IPCORK_OPT)
1182 opt = cork->opt;
1183
1184 if (!(rt->dst.dev->features&NETIF_F_SG))
1185 return -EOPNOTSUPP;
1186
1187 hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1188 mtu = cork->fragsize;
1189
1190 fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1191 maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1192 maxnonfragsize = ip_sk_local_df(sk) ? 0xFFFF : mtu;
1193
1194 if (cork->length + size > maxnonfragsize - fragheaderlen) {
1195 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1196 mtu - (opt ? opt->optlen : 0));
1197 return -EMSGSIZE;
1198 }
1199
1200 if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
1201 return -EINVAL;
1202
1203 cork->length += size;
1204 if ((size + skb->len > mtu) &&
1205 (sk->sk_protocol == IPPROTO_UDP) &&
1206 (rt->dst.dev->features & NETIF_F_UFO)) {
1207 skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
1208 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1209 }
1210
1211
1212 while (size > 0) {
1213 int i;
1214
1215 if (skb_is_gso(skb))
1216 len = size;
1217 else {
1218
1219 /* Check if the remaining data fits into current packet. */
1220 len = mtu - skb->len;
1221 if (len < size)
1222 len = maxfraglen - skb->len;
1223 }
1224 if (len <= 0) {
1225 struct sk_buff *skb_prev;
1226 int alloclen;
1227
1228 skb_prev = skb;
1229 fraggap = skb_prev->len - maxfraglen;
1230
1231 alloclen = fragheaderlen + hh_len + fraggap + 15;
1232 skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1233 if (unlikely(!skb)) {
1234 err = -ENOBUFS;
1235 goto error;
1236 }
1237
1238 /*
1239 * Fill in the control structures
1240 */
1241 skb->ip_summed = CHECKSUM_NONE;
1242 skb->csum = 0;
1243 skb_reserve(skb, hh_len);
1244
1245 /*
1246 * Find where to start putting bytes.
1247 */
1248 skb_put(skb, fragheaderlen + fraggap);
1249 skb_reset_network_header(skb);
1250 skb->transport_header = (skb->network_header +
1251 fragheaderlen);
1252 if (fraggap) {
1253 skb->csum = skb_copy_and_csum_bits(skb_prev,
1254 maxfraglen,
1255 skb_transport_header(skb),
1256 fraggap, 0);
1257 skb_prev->csum = csum_sub(skb_prev->csum,
1258 skb->csum);
1259 pskb_trim_unique(skb_prev, maxfraglen);
1260 }
1261
1262 /*
1263 * Put the packet on the pending queue.
1264 */
1265 __skb_queue_tail(&sk->sk_write_queue, skb);
1266 continue;
1267 }
1268
1269 i = skb_shinfo(skb)->nr_frags;
1270 if (len > size)
1271 len = size;
1272 if (skb_can_coalesce(skb, i, page, offset)) {
1273 skb_frag_size_add(&skb_shinfo(skb)->frags[i-1], len);
1274 } else if (i < MAX_SKB_FRAGS) {
1275 get_page(page);
1276 skb_fill_page_desc(skb, i, page, offset, len);
1277 } else {
1278 err = -EMSGSIZE;
1279 goto error;
1280 }
1281
1282 if (skb->ip_summed == CHECKSUM_NONE) {
1283 __wsum csum;
1284 csum = csum_page(page, offset, len);
1285 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1286 }
1287
1288 skb->len += len;
1289 skb->data_len += len;
1290 skb->truesize += len;
1291 atomic_add(len, &sk->sk_wmem_alloc);
1292 offset += len;
1293 size -= len;
1294 }
1295 return 0;
1296
1297error:
1298 cork->length -= size;
1299 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1300 return err;
1301}
1302
1303static void ip_cork_release(struct inet_cork *cork)
1304{
1305 cork->flags &= ~IPCORK_OPT;
1306 kfree(cork->opt);
1307 cork->opt = NULL;
1308 dst_release(cork->dst);
1309 cork->dst = NULL;
1310}
1311
1312/*
1313 * Combined all pending IP fragments on the socket as one IP datagram
1314 * and push them out.
1315 */
1316struct sk_buff *__ip_make_skb(struct sock *sk,
1317 struct flowi4 *fl4,
1318 struct sk_buff_head *queue,
1319 struct inet_cork *cork)
1320{
1321 struct sk_buff *skb, *tmp_skb;
1322 struct sk_buff **tail_skb;
1323 struct inet_sock *inet = inet_sk(sk);
1324 struct net *net = sock_net(sk);
1325 struct ip_options *opt = NULL;
1326 struct rtable *rt = (struct rtable *)cork->dst;
1327 struct iphdr *iph;
1328 __be16 df = 0;
1329 __u8 ttl;
1330
1331 if ((skb = __skb_dequeue(queue)) == NULL)
1332 goto out;
1333 tail_skb = &(skb_shinfo(skb)->frag_list);
1334
1335 /* move skb->data to ip header from ext header */
1336 if (skb->data < skb_network_header(skb))
1337 __skb_pull(skb, skb_network_offset(skb));
1338 while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1339 __skb_pull(tmp_skb, skb_network_header_len(skb));
1340 *tail_skb = tmp_skb;
1341 tail_skb = &(tmp_skb->next);
1342 skb->len += tmp_skb->len;
1343 skb->data_len += tmp_skb->len;
1344 skb->truesize += tmp_skb->truesize;
1345 tmp_skb->destructor = NULL;
1346 tmp_skb->sk = NULL;
1347 }
1348
1349 /* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1350 * to fragment the frame generated here. No matter, what transforms
1351 * how transforms change size of the packet, it will come out.
1352 */
1353 skb->local_df = ip_sk_local_df(sk);
1354
1355 /* DF bit is set when we want to see DF on outgoing frames.
1356 * If local_df is set too, we still allow to fragment this frame
1357 * locally. */
1358 if (inet->pmtudisc == IP_PMTUDISC_DO ||
1359 inet->pmtudisc == IP_PMTUDISC_PROBE ||
1360 (skb->len <= dst_mtu(&rt->dst) &&
1361 ip_dont_fragment(sk, &rt->dst)))
1362 df = htons(IP_DF);
1363
1364 if (cork->flags & IPCORK_OPT)
1365 opt = cork->opt;
1366
1367 if (cork->ttl != 0)
1368 ttl = cork->ttl;
1369 else if (rt->rt_type == RTN_MULTICAST)
1370 ttl = inet->mc_ttl;
1371 else
1372 ttl = ip_select_ttl(inet, &rt->dst);
1373
1374 iph = ip_hdr(skb);
1375 iph->version = 4;
1376 iph->ihl = 5;
1377 iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1378 iph->frag_off = df;
1379 iph->ttl = ttl;
1380 iph->protocol = sk->sk_protocol;
1381 ip_copy_addrs(iph, fl4);
1382 ip_select_ident(skb, &rt->dst, sk);
1383
1384 if (opt) {
1385 iph->ihl += opt->optlen>>2;
1386 ip_options_build(skb, opt, cork->addr, rt, 0);
1387 }
1388
1389 skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1390 skb->mark = sk->sk_mark;
1391 /*
1392 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1393 * on dst refcount
1394 */
1395 cork->dst = NULL;
1396 skb_dst_set(skb, &rt->dst);
1397
1398 if (iph->protocol == IPPROTO_ICMP)
1399 icmp_out_count(net, ((struct icmphdr *)
1400 skb_transport_header(skb))->type);
1401
1402 ip_cork_release(cork);
1403out:
1404 return skb;
1405}
1406
1407int ip_send_skb(struct net *net, struct sk_buff *skb)
1408{
1409 int err;
1410
1411 err = ip_local_out(skb);
1412 if (err) {
1413 if (err > 0)
1414 err = net_xmit_errno(err);
1415 if (err)
1416 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1417 }
1418
1419 return err;
1420}
1421
1422int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1423{
1424 struct sk_buff *skb;
1425
1426 skb = ip_finish_skb(sk, fl4);
1427 if (!skb)
1428 return 0;
1429
1430 /* Netfilter gets whole the not fragmented skb. */
1431 return ip_send_skb(sock_net(sk), skb);
1432}
1433
1434/*
1435 * Throw away all pending data on the socket.
1436 */
1437static void __ip_flush_pending_frames(struct sock *sk,
1438 struct sk_buff_head *queue,
1439 struct inet_cork *cork)
1440{
1441 struct sk_buff *skb;
1442
1443 while ((skb = __skb_dequeue_tail(queue)) != NULL)
1444 kfree_skb(skb);
1445
1446 ip_cork_release(cork);
1447}
1448
1449void ip_flush_pending_frames(struct sock *sk)
1450{
1451 __ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1452}
1453
1454struct sk_buff *ip_make_skb(struct sock *sk,
1455 struct flowi4 *fl4,
1456 int getfrag(void *from, char *to, int offset,
1457 int len, int odd, struct sk_buff *skb),
1458 void *from, int length, int transhdrlen,
1459 struct ipcm_cookie *ipc, struct rtable **rtp,
1460 unsigned int flags)
1461{
1462 struct inet_cork cork;
1463 struct sk_buff_head queue;
1464 int err;
1465
1466 if (flags & MSG_PROBE)
1467 return NULL;
1468
1469 __skb_queue_head_init(&queue);
1470
1471 cork.flags = 0;
1472 cork.addr = 0;
1473 cork.opt = NULL;
1474 err = ip_setup_cork(sk, &cork, ipc, rtp);
1475 if (err)
1476 return ERR_PTR(err);
1477
1478 err = __ip_append_data(sk, fl4, &queue, &cork,
1479 ¤t->task_frag, getfrag,
1480 from, length, transhdrlen, flags);
1481 if (err) {
1482 __ip_flush_pending_frames(sk, &queue, &cork);
1483 return ERR_PTR(err);
1484 }
1485
1486 return __ip_make_skb(sk, fl4, &queue, &cork);
1487}
1488
1489/*
1490 * Fetch data from kernel space and fill in checksum if needed.
1491 */
1492static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1493 int len, int odd, struct sk_buff *skb)
1494{
1495 __wsum csum;
1496
1497 csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
1498 skb->csum = csum_block_add(skb->csum, csum, odd);
1499 return 0;
1500}
1501
1502/*
1503 * Generic function to send a packet as reply to another packet.
1504 * Used to send some TCP resets/acks so far.
1505 *
1506 * Use a fake percpu inet socket to avoid false sharing and contention.
1507 */
1508static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
1509 .sk = {
1510 .__sk_common = {
1511 .skc_refcnt = ATOMIC_INIT(1),
1512 },
1513 .sk_wmem_alloc = ATOMIC_INIT(1),
1514 .sk_allocation = GFP_ATOMIC,
1515 .sk_flags = (1UL << SOCK_USE_WRITE_QUEUE),
1516 },
1517 .pmtudisc = IP_PMTUDISC_WANT,
1518 .uc_ttl = -1,
1519};
1520
1521void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
1522 __be32 saddr, const struct ip_reply_arg *arg,
1523 unsigned int len)
1524{
1525 struct ip_options_data replyopts;
1526 struct ipcm_cookie ipc;
1527 struct flowi4 fl4;
1528 struct rtable *rt = skb_rtable(skb);
1529 struct sk_buff *nskb;
1530 struct sock *sk;
1531 struct inet_sock *inet;
1532
1533 if (ip_options_echo(&replyopts.opt.opt, skb))
1534 return;
1535
1536 ipc.addr = daddr;
1537 ipc.opt = NULL;
1538 ipc.tx_flags = 0;
1539 ipc.ttl = 0;
1540 ipc.tos = -1;
1541
1542 if (replyopts.opt.opt.optlen) {
1543 ipc.opt = &replyopts.opt;
1544
1545 if (replyopts.opt.opt.srr)
1546 daddr = replyopts.opt.opt.faddr;
1547 }
1548
1549 flowi4_init_output(&fl4, arg->bound_dev_if, 0,
1550 RT_TOS(arg->tos),
1551 RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1552 ip_reply_arg_flowi_flags(arg),
1553 daddr, saddr,
1554 tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
1555 security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1556 rt = ip_route_output_key(net, &fl4);
1557 if (IS_ERR(rt))
1558 return;
1559
1560 inet = &get_cpu_var(unicast_sock);
1561
1562 inet->tos = arg->tos;
1563 sk = &inet->sk;
1564 sk->sk_priority = skb->priority;
1565 sk->sk_protocol = ip_hdr(skb)->protocol;
1566 sk->sk_bound_dev_if = arg->bound_dev_if;
1567 sock_net_set(sk, net);
1568 __skb_queue_head_init(&sk->sk_write_queue);
1569 sk->sk_sndbuf = sysctl_wmem_default;
1570 ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base, len, 0,
1571 &ipc, &rt, MSG_DONTWAIT);
1572 nskb = skb_peek(&sk->sk_write_queue);
1573 if (nskb) {
1574 if (arg->csumoffset >= 0)
1575 *((__sum16 *)skb_transport_header(nskb) +
1576 arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1577 arg->csum));
1578 nskb->ip_summed = CHECKSUM_NONE;
1579 skb_orphan(nskb);
1580 skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1581 ip_push_pending_frames(sk, &fl4);
1582 }
1583
1584 put_cpu_var(unicast_sock);
1585
1586 ip_rt_put(rt);
1587}
1588
1589void __init ip_init(void)
1590{
1591 ip_rt_init();
1592 inet_initpeers();
1593
1594#if defined(CONFIG_IP_MULTICAST)
1595 igmp_mc_init();
1596#endif
1597}