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