xref: /kernel/linux/linux-5.10/net/ipv4/ip_output.c (revision 8c2ecf20)
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			  u8 tos)
148{
149	struct inet_sock *inet = inet_sk(sk);
150	struct rtable *rt = skb_rtable(skb);
151	struct net *net = sock_net(sk);
152	struct iphdr *iph;
153
154	/* Build the IP header. */
155	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
156	skb_reset_network_header(skb);
157	iph = ip_hdr(skb);
158	iph->version  = 4;
159	iph->ihl      = 5;
160	iph->tos      = tos;
161	iph->ttl      = ip_select_ttl(inet, &rt->dst);
162	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
163	iph->saddr    = saddr;
164	iph->protocol = sk->sk_protocol;
165	/* Do not bother generating IPID for small packets (eg SYNACK) */
166	if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
167		iph->frag_off = htons(IP_DF);
168		iph->id = 0;
169	} else {
170		iph->frag_off = 0;
171		/* TCP packets here are SYNACK with fat IPv4/TCP options.
172		 * Avoid using the hashed IP ident generator.
173		 */
174		if (sk->sk_protocol == IPPROTO_TCP)
175			iph->id = (__force __be16)prandom_u32();
176		else
177			__ip_select_ident(net, iph, 1);
178	}
179
180	if (opt && opt->opt.optlen) {
181		iph->ihl += opt->opt.optlen>>2;
182		ip_options_build(skb, &opt->opt, daddr, rt, 0);
183	}
184
185	skb->priority = sk->sk_priority;
186	if (!skb->mark)
187		skb->mark = sk->sk_mark;
188
189	/* Send it out. */
190	return ip_local_out(net, skb->sk, skb);
191}
192EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
193
194static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
195{
196	struct dst_entry *dst = skb_dst(skb);
197	struct rtable *rt = (struct rtable *)dst;
198	struct net_device *dev = dst->dev;
199	unsigned int hh_len = LL_RESERVED_SPACE(dev);
200	struct neighbour *neigh;
201	bool is_v6gw = false;
202
203	if (rt->rt_type == RTN_MULTICAST) {
204		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
205	} else if (rt->rt_type == RTN_BROADCAST)
206		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
207
208	/* Be paranoid, rather than too clever. */
209	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
210		struct sk_buff *skb2;
211
212		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
213		if (!skb2) {
214			kfree_skb(skb);
215			return -ENOMEM;
216		}
217		if (skb->sk)
218			skb_set_owner_w(skb2, skb->sk);
219		consume_skb(skb);
220		skb = skb2;
221	}
222
223	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
224		int res = lwtunnel_xmit(skb);
225
226		if (res != LWTUNNEL_XMIT_CONTINUE)
227			return res;
228	}
229
230	rcu_read_lock_bh();
231	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
232	if (!IS_ERR(neigh)) {
233		int res;
234
235		sock_confirm_neigh(skb, neigh);
236		/* if crossing protocols, can not use the cached header */
237		res = neigh_output(neigh, skb, is_v6gw);
238		rcu_read_unlock_bh();
239		return res;
240	}
241	rcu_read_unlock_bh();
242
243	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
244			    __func__);
245	kfree_skb(skb);
246	return -EINVAL;
247}
248
249static int ip_finish_output_gso(struct net *net, struct sock *sk,
250				struct sk_buff *skb, unsigned int mtu)
251{
252	struct sk_buff *segs, *nskb;
253	netdev_features_t features;
254	int ret = 0;
255
256	/* common case: seglen is <= mtu
257	 */
258	if (skb_gso_validate_network_len(skb, mtu))
259		return ip_finish_output2(net, sk, skb);
260
261	/* Slowpath -  GSO segment length exceeds the egress MTU.
262	 *
263	 * This can happen in several cases:
264	 *  - Forwarding of a TCP GRO skb, when DF flag is not set.
265	 *  - Forwarding of an skb that arrived on a virtualization interface
266	 *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
267	 *    stack.
268	 *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
269	 *    interface with a smaller MTU.
270	 *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
271	 *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
272	 *    insufficent MTU.
273	 */
274	features = netif_skb_features(skb);
275	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
276	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
277	if (IS_ERR_OR_NULL(segs)) {
278		kfree_skb(skb);
279		return -ENOMEM;
280	}
281
282	consume_skb(skb);
283
284	skb_list_walk_safe(segs, segs, nskb) {
285		int err;
286
287		skb_mark_not_on_list(segs);
288		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
289
290		if (err && ret == 0)
291			ret = err;
292	}
293
294	return ret;
295}
296
297static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
298{
299	unsigned int mtu;
300
301#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
302	/* Policy lookup after SNAT yielded a new policy */
303	if (skb_dst(skb)->xfrm) {
304		IPCB(skb)->flags |= IPSKB_REROUTED;
305		return dst_output(net, sk, skb);
306	}
307#endif
308	mtu = ip_skb_dst_mtu(sk, skb);
309	if (skb_is_gso(skb))
310		return ip_finish_output_gso(net, sk, skb, mtu);
311
312	if (skb->len > mtu || IPCB(skb)->frag_max_size)
313		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
314
315	return ip_finish_output2(net, sk, skb);
316}
317
318static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
319{
320	int ret;
321
322	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
323	switch (ret) {
324	case NET_XMIT_SUCCESS:
325		return __ip_finish_output(net, sk, skb);
326	case NET_XMIT_CN:
327		return __ip_finish_output(net, sk, skb) ? : ret;
328	default:
329		kfree_skb(skb);
330		return ret;
331	}
332}
333
334static int ip_mc_finish_output(struct net *net, struct sock *sk,
335			       struct sk_buff *skb)
336{
337	struct rtable *new_rt;
338	bool do_cn = false;
339	int ret, err;
340
341	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
342	switch (ret) {
343	case NET_XMIT_CN:
344		do_cn = true;
345		fallthrough;
346	case NET_XMIT_SUCCESS:
347		break;
348	default:
349		kfree_skb(skb);
350		return ret;
351	}
352
353	/* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
354	 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
355	 * see ipv4_pktinfo_prepare().
356	 */
357	new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
358	if (new_rt) {
359		new_rt->rt_iif = 0;
360		skb_dst_drop(skb);
361		skb_dst_set(skb, &new_rt->dst);
362	}
363
364	err = dev_loopback_xmit(net, sk, skb);
365	return (do_cn && err) ? ret : err;
366}
367
368int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
369{
370	struct rtable *rt = skb_rtable(skb);
371	struct net_device *dev = rt->dst.dev;
372
373	/*
374	 *	If the indicated interface is up and running, send the packet.
375	 */
376	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
377
378	skb->dev = dev;
379	skb->protocol = htons(ETH_P_IP);
380
381	/*
382	 *	Multicasts are looped back for other local users
383	 */
384
385	if (rt->rt_flags&RTCF_MULTICAST) {
386		if (sk_mc_loop(sk)
387#ifdef CONFIG_IP_MROUTE
388		/* Small optimization: do not loopback not local frames,
389		   which returned after forwarding; they will be  dropped
390		   by ip_mr_input in any case.
391		   Note, that local frames are looped back to be delivered
392		   to local recipients.
393
394		   This check is duplicated in ip_mr_input at the moment.
395		 */
396		    &&
397		    ((rt->rt_flags & RTCF_LOCAL) ||
398		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
399#endif
400		   ) {
401			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
402			if (newskb)
403				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
404					net, sk, newskb, NULL, newskb->dev,
405					ip_mc_finish_output);
406		}
407
408		/* Multicasts with ttl 0 must not go beyond the host */
409
410		if (ip_hdr(skb)->ttl == 0) {
411			kfree_skb(skb);
412			return 0;
413		}
414	}
415
416	if (rt->rt_flags&RTCF_BROADCAST) {
417		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
418		if (newskb)
419			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
420				net, sk, newskb, NULL, newskb->dev,
421				ip_mc_finish_output);
422	}
423
424	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
425			    net, sk, skb, NULL, skb->dev,
426			    ip_finish_output,
427			    !(IPCB(skb)->flags & IPSKB_REROUTED));
428}
429
430int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
431{
432	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
433
434	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
435
436	skb->dev = dev;
437	skb->protocol = htons(ETH_P_IP);
438
439	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
440			    net, sk, skb, indev, dev,
441			    ip_finish_output,
442			    !(IPCB(skb)->flags & IPSKB_REROUTED));
443}
444
445/*
446 * copy saddr and daddr, possibly using 64bit load/stores
447 * Equivalent to :
448 *   iph->saddr = fl4->saddr;
449 *   iph->daddr = fl4->daddr;
450 */
451static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
452{
453	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
454		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
455
456	iph->saddr = fl4->saddr;
457	iph->daddr = fl4->daddr;
458}
459
460/* Note: skb->sk can be different from sk, in case of tunnels */
461int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
462		    __u8 tos)
463{
464	struct inet_sock *inet = inet_sk(sk);
465	struct net *net = sock_net(sk);
466	struct ip_options_rcu *inet_opt;
467	struct flowi4 *fl4;
468	struct rtable *rt;
469	struct iphdr *iph;
470	int res;
471
472	/* Skip all of this if the packet is already routed,
473	 * f.e. by something like SCTP.
474	 */
475	rcu_read_lock();
476	inet_opt = rcu_dereference(inet->inet_opt);
477	fl4 = &fl->u.ip4;
478	rt = skb_rtable(skb);
479	if (rt)
480		goto packet_routed;
481
482	/* Make sure we can route this packet. */
483	rt = (struct rtable *)__sk_dst_check(sk, 0);
484	if (!rt) {
485		__be32 daddr;
486
487		/* Use correct destination address if we have options. */
488		daddr = inet->inet_daddr;
489		if (inet_opt && inet_opt->opt.srr)
490			daddr = inet_opt->opt.faddr;
491
492		/* If this fails, retransmit mechanism of transport layer will
493		 * keep trying until route appears or the connection times
494		 * itself out.
495		 */
496		rt = ip_route_output_ports(net, fl4, sk,
497					   daddr, inet->inet_saddr,
498					   inet->inet_dport,
499					   inet->inet_sport,
500					   sk->sk_protocol,
501					   RT_CONN_FLAGS_TOS(sk, tos),
502					   sk->sk_bound_dev_if);
503		if (IS_ERR(rt))
504			goto no_route;
505		sk_setup_caps(sk, &rt->dst);
506	}
507	skb_dst_set_noref(skb, &rt->dst);
508
509packet_routed:
510	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
511		goto no_route;
512
513	/* OK, we know where to send it, allocate and build IP header. */
514	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
515	skb_reset_network_header(skb);
516	iph = ip_hdr(skb);
517	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
518	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
519		iph->frag_off = htons(IP_DF);
520	else
521		iph->frag_off = 0;
522	iph->ttl      = ip_select_ttl(inet, &rt->dst);
523	iph->protocol = sk->sk_protocol;
524	ip_copy_addrs(iph, fl4);
525
526	/* Transport layer set skb->h.foo itself. */
527
528	if (inet_opt && inet_opt->opt.optlen) {
529		iph->ihl += inet_opt->opt.optlen >> 2;
530		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
531	}
532
533	ip_select_ident_segs(net, skb, sk,
534			     skb_shinfo(skb)->gso_segs ?: 1);
535
536	/* TODO : should we use skb->sk here instead of sk ? */
537	skb->priority = sk->sk_priority;
538	skb->mark = sk->sk_mark;
539
540	res = ip_local_out(net, sk, skb);
541	rcu_read_unlock();
542	return res;
543
544no_route:
545	rcu_read_unlock();
546	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
547	kfree_skb(skb);
548	return -EHOSTUNREACH;
549}
550EXPORT_SYMBOL(__ip_queue_xmit);
551
552int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
553{
554	return __ip_queue_xmit(sk, skb, fl, inet_sk(sk)->tos);
555}
556EXPORT_SYMBOL(ip_queue_xmit);
557
558static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
559{
560	to->pkt_type = from->pkt_type;
561	to->priority = from->priority;
562	to->protocol = from->protocol;
563	to->skb_iif = from->skb_iif;
564	skb_dst_drop(to);
565	skb_dst_copy(to, from);
566	to->dev = from->dev;
567	to->mark = from->mark;
568
569	skb_copy_hash(to, from);
570
571#ifdef CONFIG_NET_SCHED
572	to->tc_index = from->tc_index;
573#endif
574	nf_copy(to, from);
575	skb_ext_copy(to, from);
576#if IS_ENABLED(CONFIG_IP_VS)
577	to->ipvs_property = from->ipvs_property;
578#endif
579	skb_copy_secmark(to, from);
580}
581
582static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
583		       unsigned int mtu,
584		       int (*output)(struct net *, struct sock *, struct sk_buff *))
585{
586	struct iphdr *iph = ip_hdr(skb);
587
588	if ((iph->frag_off & htons(IP_DF)) == 0)
589		return ip_do_fragment(net, sk, skb, output);
590
591	if (unlikely(!skb->ignore_df ||
592		     (IPCB(skb)->frag_max_size &&
593		      IPCB(skb)->frag_max_size > mtu))) {
594		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
595		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
596			  htonl(mtu));
597		kfree_skb(skb);
598		return -EMSGSIZE;
599	}
600
601	return ip_do_fragment(net, sk, skb, output);
602}
603
604void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
605		      unsigned int hlen, struct ip_fraglist_iter *iter)
606{
607	unsigned int first_len = skb_pagelen(skb);
608
609	iter->frag = skb_shinfo(skb)->frag_list;
610	skb_frag_list_init(skb);
611
612	iter->offset = 0;
613	iter->iph = iph;
614	iter->hlen = hlen;
615
616	skb->data_len = first_len - skb_headlen(skb);
617	skb->len = first_len;
618	iph->tot_len = htons(first_len);
619	iph->frag_off = htons(IP_MF);
620	ip_send_check(iph);
621}
622EXPORT_SYMBOL(ip_fraglist_init);
623
624void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
625{
626	unsigned int hlen = iter->hlen;
627	struct iphdr *iph = iter->iph;
628	struct sk_buff *frag;
629
630	frag = iter->frag;
631	frag->ip_summed = CHECKSUM_NONE;
632	skb_reset_transport_header(frag);
633	__skb_push(frag, hlen);
634	skb_reset_network_header(frag);
635	memcpy(skb_network_header(frag), iph, hlen);
636	iter->iph = ip_hdr(frag);
637	iph = iter->iph;
638	iph->tot_len = htons(frag->len);
639	ip_copy_metadata(frag, skb);
640	iter->offset += skb->len - hlen;
641	iph->frag_off = htons(iter->offset >> 3);
642	if (frag->next)
643		iph->frag_off |= htons(IP_MF);
644	/* Ready, complete checksum */
645	ip_send_check(iph);
646}
647EXPORT_SYMBOL(ip_fraglist_prepare);
648
649void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
650		  unsigned int ll_rs, unsigned int mtu, bool DF,
651		  struct ip_frag_state *state)
652{
653	struct iphdr *iph = ip_hdr(skb);
654
655	state->DF = DF;
656	state->hlen = hlen;
657	state->ll_rs = ll_rs;
658	state->mtu = mtu;
659
660	state->left = skb->len - hlen;	/* Space per frame */
661	state->ptr = hlen;		/* Where to start from */
662
663	state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
664	state->not_last_frag = iph->frag_off & htons(IP_MF);
665}
666EXPORT_SYMBOL(ip_frag_init);
667
668static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
669			 bool first_frag)
670{
671	/* Copy the flags to each fragment. */
672	IPCB(to)->flags = IPCB(from)->flags;
673
674	/* ANK: dirty, but effective trick. Upgrade options only if
675	 * the segment to be fragmented was THE FIRST (otherwise,
676	 * options are already fixed) and make it ONCE
677	 * on the initial skb, so that all the following fragments
678	 * will inherit fixed options.
679	 */
680	if (first_frag)
681		ip_options_fragment(from);
682}
683
684struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
685{
686	unsigned int len = state->left;
687	struct sk_buff *skb2;
688	struct iphdr *iph;
689
690	len = state->left;
691	/* IF: it doesn't fit, use 'mtu' - the data space left */
692	if (len > state->mtu)
693		len = state->mtu;
694	/* IF: we are not sending up to and including the packet end
695	   then align the next start on an eight byte boundary */
696	if (len < state->left)	{
697		len &= ~7;
698	}
699
700	/* Allocate buffer */
701	skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
702	if (!skb2)
703		return ERR_PTR(-ENOMEM);
704
705	/*
706	 *	Set up data on packet
707	 */
708
709	ip_copy_metadata(skb2, skb);
710	skb_reserve(skb2, state->ll_rs);
711	skb_put(skb2, len + state->hlen);
712	skb_reset_network_header(skb2);
713	skb2->transport_header = skb2->network_header + state->hlen;
714
715	/*
716	 *	Charge the memory for the fragment to any owner
717	 *	it might possess
718	 */
719
720	if (skb->sk)
721		skb_set_owner_w(skb2, skb->sk);
722
723	/*
724	 *	Copy the packet header into the new buffer.
725	 */
726
727	skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
728
729	/*
730	 *	Copy a block of the IP datagram.
731	 */
732	if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
733		BUG();
734	state->left -= len;
735
736	/*
737	 *	Fill in the new header fields.
738	 */
739	iph = ip_hdr(skb2);
740	iph->frag_off = htons((state->offset >> 3));
741	if (state->DF)
742		iph->frag_off |= htons(IP_DF);
743
744	/*
745	 *	Added AC : If we are fragmenting a fragment that's not the
746	 *		   last fragment then keep MF on each bit
747	 */
748	if (state->left > 0 || state->not_last_frag)
749		iph->frag_off |= htons(IP_MF);
750	state->ptr += len;
751	state->offset += len;
752
753	iph->tot_len = htons(len + state->hlen);
754
755	ip_send_check(iph);
756
757	return skb2;
758}
759EXPORT_SYMBOL(ip_frag_next);
760
761/*
762 *	This IP datagram is too large to be sent in one piece.  Break it up into
763 *	smaller pieces (each of size equal to IP header plus
764 *	a block of the data of the original IP data part) that will yet fit in a
765 *	single device frame, and queue such a frame for sending.
766 */
767
768int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
769		   int (*output)(struct net *, struct sock *, struct sk_buff *))
770{
771	struct iphdr *iph;
772	struct sk_buff *skb2;
773	struct rtable *rt = skb_rtable(skb);
774	unsigned int mtu, hlen, ll_rs;
775	struct ip_fraglist_iter iter;
776	ktime_t tstamp = skb->tstamp;
777	struct ip_frag_state state;
778	int err = 0;
779
780	/* for offloaded checksums cleanup checksum before fragmentation */
781	if (skb->ip_summed == CHECKSUM_PARTIAL &&
782	    (err = skb_checksum_help(skb)))
783		goto fail;
784
785	/*
786	 *	Point into the IP datagram header.
787	 */
788
789	iph = ip_hdr(skb);
790
791	mtu = ip_skb_dst_mtu(sk, skb);
792	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
793		mtu = IPCB(skb)->frag_max_size;
794
795	/*
796	 *	Setup starting values.
797	 */
798
799	hlen = iph->ihl * 4;
800	mtu = mtu - hlen;	/* Size of data space */
801	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
802	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
803
804	/* When frag_list is given, use it. First, check its validity:
805	 * some transformers could create wrong frag_list or break existing
806	 * one, it is not prohibited. In this case fall back to copying.
807	 *
808	 * LATER: this step can be merged to real generation of fragments,
809	 * we can switch to copy when see the first bad fragment.
810	 */
811	if (skb_has_frag_list(skb)) {
812		struct sk_buff *frag, *frag2;
813		unsigned int first_len = skb_pagelen(skb);
814
815		if (first_len - hlen > mtu ||
816		    ((first_len - hlen) & 7) ||
817		    ip_is_fragment(iph) ||
818		    skb_cloned(skb) ||
819		    skb_headroom(skb) < ll_rs)
820			goto slow_path;
821
822		skb_walk_frags(skb, frag) {
823			/* Correct geometry. */
824			if (frag->len > mtu ||
825			    ((frag->len & 7) && frag->next) ||
826			    skb_headroom(frag) < hlen + ll_rs)
827				goto slow_path_clean;
828
829			/* Partially cloned skb? */
830			if (skb_shared(frag))
831				goto slow_path_clean;
832
833			BUG_ON(frag->sk);
834			if (skb->sk) {
835				frag->sk = skb->sk;
836				frag->destructor = sock_wfree;
837			}
838			skb->truesize -= frag->truesize;
839		}
840
841		/* Everything is OK. Generate! */
842		ip_fraglist_init(skb, iph, hlen, &iter);
843
844		for (;;) {
845			/* Prepare header of the next frame,
846			 * before previous one went down. */
847			if (iter.frag) {
848				bool first_frag = (iter.offset == 0);
849
850				IPCB(iter.frag)->flags = IPCB(skb)->flags;
851				ip_fraglist_prepare(skb, &iter);
852				if (first_frag && IPCB(skb)->opt.optlen) {
853					/* ipcb->opt is not populated for frags
854					 * coming from __ip_make_skb(),
855					 * ip_options_fragment() needs optlen
856					 */
857					IPCB(iter.frag)->opt.optlen =
858						IPCB(skb)->opt.optlen;
859					ip_options_fragment(iter.frag);
860					ip_send_check(iter.iph);
861				}
862			}
863
864			skb->tstamp = tstamp;
865			err = output(net, sk, skb);
866
867			if (!err)
868				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
869			if (err || !iter.frag)
870				break;
871
872			skb = ip_fraglist_next(&iter);
873		}
874
875		if (err == 0) {
876			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
877			return 0;
878		}
879
880		kfree_skb_list(iter.frag);
881
882		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
883		return err;
884
885slow_path_clean:
886		skb_walk_frags(skb, frag2) {
887			if (frag2 == frag)
888				break;
889			frag2->sk = NULL;
890			frag2->destructor = NULL;
891			skb->truesize += frag2->truesize;
892		}
893	}
894
895slow_path:
896	/*
897	 *	Fragment the datagram.
898	 */
899
900	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
901		     &state);
902
903	/*
904	 *	Keep copying data until we run out.
905	 */
906
907	while (state.left > 0) {
908		bool first_frag = (state.offset == 0);
909
910		skb2 = ip_frag_next(skb, &state);
911		if (IS_ERR(skb2)) {
912			err = PTR_ERR(skb2);
913			goto fail;
914		}
915		ip_frag_ipcb(skb, skb2, first_frag);
916
917		/*
918		 *	Put this fragment into the sending queue.
919		 */
920		skb2->tstamp = tstamp;
921		err = output(net, sk, skb2);
922		if (err)
923			goto fail;
924
925		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
926	}
927	consume_skb(skb);
928	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
929	return err;
930
931fail:
932	kfree_skb(skb);
933	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
934	return err;
935}
936EXPORT_SYMBOL(ip_do_fragment);
937
938int
939ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
940{
941	struct msghdr *msg = from;
942
943	if (skb->ip_summed == CHECKSUM_PARTIAL) {
944		if (!copy_from_iter_full(to, len, &msg->msg_iter))
945			return -EFAULT;
946	} else {
947		__wsum csum = 0;
948		if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
949			return -EFAULT;
950		skb->csum = csum_block_add(skb->csum, csum, odd);
951	}
952	return 0;
953}
954EXPORT_SYMBOL(ip_generic_getfrag);
955
956static inline __wsum
957csum_page(struct page *page, int offset, int copy)
958{
959	char *kaddr;
960	__wsum csum;
961	kaddr = kmap(page);
962	csum = csum_partial(kaddr + offset, copy, 0);
963	kunmap(page);
964	return csum;
965}
966
967static int __ip_append_data(struct sock *sk,
968			    struct flowi4 *fl4,
969			    struct sk_buff_head *queue,
970			    struct inet_cork *cork,
971			    struct page_frag *pfrag,
972			    int getfrag(void *from, char *to, int offset,
973					int len, int odd, struct sk_buff *skb),
974			    void *from, int length, int transhdrlen,
975			    unsigned int flags)
976{
977	struct inet_sock *inet = inet_sk(sk);
978	struct ubuf_info *uarg = NULL;
979	struct sk_buff *skb;
980
981	struct ip_options *opt = cork->opt;
982	int hh_len;
983	int exthdrlen;
984	int mtu;
985	int copy;
986	int err;
987	int offset = 0;
988	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
989	int csummode = CHECKSUM_NONE;
990	struct rtable *rt = (struct rtable *)cork->dst;
991	unsigned int wmem_alloc_delta = 0;
992	bool paged, extra_uref = false;
993	u32 tskey = 0;
994
995	skb = skb_peek_tail(queue);
996
997	exthdrlen = !skb ? rt->dst.header_len : 0;
998	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
999	paged = !!cork->gso_size;
1000
1001	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
1002	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
1003		tskey = sk->sk_tskey++;
1004
1005	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1006
1007	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1008	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1009	maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
1010
1011	if (cork->length + length > maxnonfragsize - fragheaderlen) {
1012		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1013			       mtu - (opt ? opt->optlen : 0));
1014		return -EMSGSIZE;
1015	}
1016
1017	/*
1018	 * transhdrlen > 0 means that this is the first fragment and we wish
1019	 * it won't be fragmented in the future.
1020	 */
1021	if (transhdrlen &&
1022	    length + fragheaderlen <= mtu &&
1023	    rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1024	    (!(flags & MSG_MORE) || cork->gso_size) &&
1025	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1026		csummode = CHECKSUM_PARTIAL;
1027
1028	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
1029		uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
1030		if (!uarg)
1031			return -ENOBUFS;
1032		extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1033		if (rt->dst.dev->features & NETIF_F_SG &&
1034		    csummode == CHECKSUM_PARTIAL) {
1035			paged = true;
1036		} else {
1037			uarg->zerocopy = 0;
1038			skb_zcopy_set(skb, uarg, &extra_uref);
1039		}
1040	}
1041
1042	cork->length += length;
1043
1044	/* So, what's going on in the loop below?
1045	 *
1046	 * We use calculated fragment length to generate chained skb,
1047	 * each of segments is IP fragment ready for sending to network after
1048	 * adding appropriate IP header.
1049	 */
1050
1051	if (!skb)
1052		goto alloc_new_skb;
1053
1054	while (length > 0) {
1055		/* Check if the remaining data fits into current packet. */
1056		copy = mtu - skb->len;
1057		if (copy < length)
1058			copy = maxfraglen - skb->len;
1059		if (copy <= 0) {
1060			char *data;
1061			unsigned int datalen;
1062			unsigned int fraglen;
1063			unsigned int fraggap;
1064			unsigned int alloclen, alloc_extra;
1065			unsigned int pagedlen;
1066			struct sk_buff *skb_prev;
1067alloc_new_skb:
1068			skb_prev = skb;
1069			if (skb_prev)
1070				fraggap = skb_prev->len - maxfraglen;
1071			else
1072				fraggap = 0;
1073
1074			/*
1075			 * If remaining data exceeds the mtu,
1076			 * we know we need more fragment(s).
1077			 */
1078			datalen = length + fraggap;
1079			if (datalen > mtu - fragheaderlen)
1080				datalen = maxfraglen - fragheaderlen;
1081			fraglen = datalen + fragheaderlen;
1082			pagedlen = 0;
1083
1084			alloc_extra = hh_len + 15;
1085			alloc_extra += 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				alloc_extra += rt->dst.trailer_len;
1094
1095			if ((flags & MSG_MORE) &&
1096			    !(rt->dst.dev->features&NETIF_F_SG))
1097				alloclen = mtu;
1098			else if (!paged &&
1099				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1100				  !(rt->dst.dev->features & NETIF_F_SG)))
1101				alloclen = fraglen;
1102			else {
1103				alloclen = min_t(int, fraglen, MAX_HEADER);
1104				pagedlen = fraglen - alloclen;
1105			}
1106
1107			alloclen += alloc_extra;
1108
1109			if (transhdrlen) {
1110				skb = sock_alloc_send_skb(sk, alloclen,
1111						(flags & MSG_DONTWAIT), &err);
1112			} else {
1113				skb = NULL;
1114				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1115				    2 * sk->sk_sndbuf)
1116					skb = alloc_skb(alloclen,
1117							sk->sk_allocation);
1118				if (unlikely(!skb))
1119					err = -ENOBUFS;
1120			}
1121			if (!skb)
1122				goto error;
1123
1124			/*
1125			 *	Fill in the control structures
1126			 */
1127			skb->ip_summed = csummode;
1128			skb->csum = 0;
1129			skb_reserve(skb, hh_len);
1130
1131			/*
1132			 *	Find where to start putting bytes.
1133			 */
1134			data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1135			skb_set_network_header(skb, exthdrlen);
1136			skb->transport_header = (skb->network_header +
1137						 fragheaderlen);
1138			data += fragheaderlen + exthdrlen;
1139
1140			if (fraggap) {
1141				skb->csum = skb_copy_and_csum_bits(
1142					skb_prev, maxfraglen,
1143					data + transhdrlen, fraggap);
1144				skb_prev->csum = csum_sub(skb_prev->csum,
1145							  skb->csum);
1146				data += fraggap;
1147				pskb_trim_unique(skb_prev, maxfraglen);
1148			}
1149
1150			copy = datalen - transhdrlen - fraggap - pagedlen;
1151			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1152				err = -EFAULT;
1153				kfree_skb(skb);
1154				goto error;
1155			}
1156
1157			offset += copy;
1158			length -= copy + transhdrlen;
1159			transhdrlen = 0;
1160			exthdrlen = 0;
1161			csummode = CHECKSUM_NONE;
1162
1163			/* only the initial fragment is time stamped */
1164			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1165			cork->tx_flags = 0;
1166			skb_shinfo(skb)->tskey = tskey;
1167			tskey = 0;
1168			skb_zcopy_set(skb, uarg, &extra_uref);
1169
1170			if ((flags & MSG_CONFIRM) && !skb_prev)
1171				skb_set_dst_pending_confirm(skb, 1);
1172
1173			/*
1174			 * Put the packet on the pending queue.
1175			 */
1176			if (!skb->destructor) {
1177				skb->destructor = sock_wfree;
1178				skb->sk = sk;
1179				wmem_alloc_delta += skb->truesize;
1180			}
1181			__skb_queue_tail(queue, skb);
1182			continue;
1183		}
1184
1185		if (copy > length)
1186			copy = length;
1187
1188		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1189		    skb_tailroom(skb) >= copy) {
1190			unsigned int off;
1191
1192			off = skb->len;
1193			if (getfrag(from, skb_put(skb, copy),
1194					offset, copy, off, skb) < 0) {
1195				__skb_trim(skb, off);
1196				err = -EFAULT;
1197				goto error;
1198			}
1199		} else if (!uarg || !uarg->zerocopy) {
1200			int i = skb_shinfo(skb)->nr_frags;
1201
1202			err = -ENOMEM;
1203			if (!sk_page_frag_refill(sk, pfrag))
1204				goto error;
1205
1206			if (!skb_can_coalesce(skb, i, pfrag->page,
1207					      pfrag->offset)) {
1208				err = -EMSGSIZE;
1209				if (i == MAX_SKB_FRAGS)
1210					goto error;
1211
1212				__skb_fill_page_desc(skb, i, pfrag->page,
1213						     pfrag->offset, 0);
1214				skb_shinfo(skb)->nr_frags = ++i;
1215				get_page(pfrag->page);
1216			}
1217			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1218			if (getfrag(from,
1219				    page_address(pfrag->page) + pfrag->offset,
1220				    offset, copy, skb->len, skb) < 0)
1221				goto error_efault;
1222
1223			pfrag->offset += copy;
1224			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1225			skb->len += copy;
1226			skb->data_len += copy;
1227			skb->truesize += copy;
1228			wmem_alloc_delta += copy;
1229		} else {
1230			err = skb_zerocopy_iter_dgram(skb, from, copy);
1231			if (err < 0)
1232				goto error;
1233		}
1234		offset += copy;
1235		length -= copy;
1236	}
1237
1238	if (wmem_alloc_delta)
1239		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1240	return 0;
1241
1242error_efault:
1243	err = -EFAULT;
1244error:
1245	if (uarg)
1246		sock_zerocopy_put_abort(uarg, extra_uref);
1247	cork->length -= length;
1248	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1249	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1250	return err;
1251}
1252
1253static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1254			 struct ipcm_cookie *ipc, struct rtable **rtp)
1255{
1256	struct ip_options_rcu *opt;
1257	struct rtable *rt;
1258
1259	rt = *rtp;
1260	if (unlikely(!rt))
1261		return -EFAULT;
1262
1263	cork->fragsize = ip_sk_use_pmtu(sk) ?
1264			 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1265
1266	if (!inetdev_valid_mtu(cork->fragsize))
1267		return -ENETUNREACH;
1268
1269	/*
1270	 * setup for corking.
1271	 */
1272	opt = ipc->opt;
1273	if (opt) {
1274		if (!cork->opt) {
1275			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1276					    sk->sk_allocation);
1277			if (unlikely(!cork->opt))
1278				return -ENOBUFS;
1279		}
1280		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1281		cork->flags |= IPCORK_OPT;
1282		cork->addr = ipc->addr;
1283	}
1284
1285	cork->gso_size = ipc->gso_size;
1286
1287	cork->dst = &rt->dst;
1288	/* We stole this route, caller should not release it. */
1289	*rtp = NULL;
1290
1291	cork->length = 0;
1292	cork->ttl = ipc->ttl;
1293	cork->tos = ipc->tos;
1294	cork->mark = ipc->sockc.mark;
1295	cork->priority = ipc->priority;
1296	cork->transmit_time = ipc->sockc.transmit_time;
1297	cork->tx_flags = 0;
1298	sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1299
1300	return 0;
1301}
1302
1303/*
1304 *	ip_append_data() and ip_append_page() can make one large IP datagram
1305 *	from many pieces of data. Each pieces will be holded on the socket
1306 *	until ip_push_pending_frames() is called. Each piece can be a page
1307 *	or non-page data.
1308 *
1309 *	Not only UDP, other transport protocols - e.g. raw sockets - can use
1310 *	this interface potentially.
1311 *
1312 *	LATER: length must be adjusted by pad at tail, when it is required.
1313 */
1314int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1315		   int getfrag(void *from, char *to, int offset, int len,
1316			       int odd, struct sk_buff *skb),
1317		   void *from, int length, int transhdrlen,
1318		   struct ipcm_cookie *ipc, struct rtable **rtp,
1319		   unsigned int flags)
1320{
1321	struct inet_sock *inet = inet_sk(sk);
1322	int err;
1323
1324	if (flags&MSG_PROBE)
1325		return 0;
1326
1327	if (skb_queue_empty(&sk->sk_write_queue)) {
1328		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1329		if (err)
1330			return err;
1331	} else {
1332		transhdrlen = 0;
1333	}
1334
1335	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1336				sk_page_frag(sk), getfrag,
1337				from, length, transhdrlen, flags);
1338}
1339
1340ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
1341		       int offset, size_t size, int flags)
1342{
1343	struct inet_sock *inet = inet_sk(sk);
1344	struct sk_buff *skb;
1345	struct rtable *rt;
1346	struct ip_options *opt = NULL;
1347	struct inet_cork *cork;
1348	int hh_len;
1349	int mtu;
1350	int len;
1351	int err;
1352	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
1353
1354	if (inet->hdrincl)
1355		return -EPERM;
1356
1357	if (flags&MSG_PROBE)
1358		return 0;
1359
1360	if (skb_queue_empty(&sk->sk_write_queue))
1361		return -EINVAL;
1362
1363	cork = &inet->cork.base;
1364	rt = (struct rtable *)cork->dst;
1365	if (cork->flags & IPCORK_OPT)
1366		opt = cork->opt;
1367
1368	if (!(rt->dst.dev->features & NETIF_F_SG))
1369		return -EOPNOTSUPP;
1370
1371	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1372	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
1373
1374	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
1375	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
1376	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
1377
1378	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1379		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
1380			       mtu - (opt ? opt->optlen : 0));
1381		return -EMSGSIZE;
1382	}
1383
1384	skb = skb_peek_tail(&sk->sk_write_queue);
1385	if (!skb)
1386		return -EINVAL;
1387
1388	cork->length += size;
1389
1390	while (size > 0) {
1391		/* Check if the remaining data fits into current packet. */
1392		len = mtu - skb->len;
1393		if (len < size)
1394			len = maxfraglen - skb->len;
1395
1396		if (len <= 0) {
1397			struct sk_buff *skb_prev;
1398			int alloclen;
1399
1400			skb_prev = skb;
1401			fraggap = skb_prev->len - maxfraglen;
1402
1403			alloclen = fragheaderlen + hh_len + fraggap + 15;
1404			skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
1405			if (unlikely(!skb)) {
1406				err = -ENOBUFS;
1407				goto error;
1408			}
1409
1410			/*
1411			 *	Fill in the control structures
1412			 */
1413			skb->ip_summed = CHECKSUM_NONE;
1414			skb->csum = 0;
1415			skb_reserve(skb, hh_len);
1416
1417			/*
1418			 *	Find where to start putting bytes.
1419			 */
1420			skb_put(skb, fragheaderlen + fraggap);
1421			skb_reset_network_header(skb);
1422			skb->transport_header = (skb->network_header +
1423						 fragheaderlen);
1424			if (fraggap) {
1425				skb->csum = skb_copy_and_csum_bits(skb_prev,
1426								   maxfraglen,
1427						    skb_transport_header(skb),
1428								   fraggap);
1429				skb_prev->csum = csum_sub(skb_prev->csum,
1430							  skb->csum);
1431				pskb_trim_unique(skb_prev, maxfraglen);
1432			}
1433
1434			/*
1435			 * Put the packet on the pending queue.
1436			 */
1437			__skb_queue_tail(&sk->sk_write_queue, skb);
1438			continue;
1439		}
1440
1441		if (len > size)
1442			len = size;
1443
1444		if (skb_append_pagefrags(skb, page, offset, len)) {
1445			err = -EMSGSIZE;
1446			goto error;
1447		}
1448
1449		if (skb->ip_summed == CHECKSUM_NONE) {
1450			__wsum csum;
1451			csum = csum_page(page, offset, len);
1452			skb->csum = csum_block_add(skb->csum, csum, skb->len);
1453		}
1454
1455		skb->len += len;
1456		skb->data_len += len;
1457		skb->truesize += len;
1458		refcount_add(len, &sk->sk_wmem_alloc);
1459		offset += len;
1460		size -= len;
1461	}
1462	return 0;
1463
1464error:
1465	cork->length -= size;
1466	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1467	return err;
1468}
1469
1470static void ip_cork_release(struct inet_cork *cork)
1471{
1472	cork->flags &= ~IPCORK_OPT;
1473	kfree(cork->opt);
1474	cork->opt = NULL;
1475	dst_release(cork->dst);
1476	cork->dst = NULL;
1477}
1478
1479/*
1480 *	Combined all pending IP fragments on the socket as one IP datagram
1481 *	and push them out.
1482 */
1483struct sk_buff *__ip_make_skb(struct sock *sk,
1484			      struct flowi4 *fl4,
1485			      struct sk_buff_head *queue,
1486			      struct inet_cork *cork)
1487{
1488	struct sk_buff *skb, *tmp_skb;
1489	struct sk_buff **tail_skb;
1490	struct inet_sock *inet = inet_sk(sk);
1491	struct net *net = sock_net(sk);
1492	struct ip_options *opt = NULL;
1493	struct rtable *rt = (struct rtable *)cork->dst;
1494	struct iphdr *iph;
1495	__be16 df = 0;
1496	__u8 ttl;
1497
1498	skb = __skb_dequeue(queue);
1499	if (!skb)
1500		goto out;
1501	tail_skb = &(skb_shinfo(skb)->frag_list);
1502
1503	/* move skb->data to ip header from ext header */
1504	if (skb->data < skb_network_header(skb))
1505		__skb_pull(skb, skb_network_offset(skb));
1506	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1507		__skb_pull(tmp_skb, skb_network_header_len(skb));
1508		*tail_skb = tmp_skb;
1509		tail_skb = &(tmp_skb->next);
1510		skb->len += tmp_skb->len;
1511		skb->data_len += tmp_skb->len;
1512		skb->truesize += tmp_skb->truesize;
1513		tmp_skb->destructor = NULL;
1514		tmp_skb->sk = NULL;
1515	}
1516
1517	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1518	 * to fragment the frame generated here. No matter, what transforms
1519	 * how transforms change size of the packet, it will come out.
1520	 */
1521	skb->ignore_df = ip_sk_ignore_df(sk);
1522
1523	/* DF bit is set when we want to see DF on outgoing frames.
1524	 * If ignore_df is set too, we still allow to fragment this frame
1525	 * locally. */
1526	if (inet->pmtudisc == IP_PMTUDISC_DO ||
1527	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
1528	    (skb->len <= dst_mtu(&rt->dst) &&
1529	     ip_dont_fragment(sk, &rt->dst)))
1530		df = htons(IP_DF);
1531
1532	if (cork->flags & IPCORK_OPT)
1533		opt = cork->opt;
1534
1535	if (cork->ttl != 0)
1536		ttl = cork->ttl;
1537	else if (rt->rt_type == RTN_MULTICAST)
1538		ttl = inet->mc_ttl;
1539	else
1540		ttl = ip_select_ttl(inet, &rt->dst);
1541
1542	iph = ip_hdr(skb);
1543	iph->version = 4;
1544	iph->ihl = 5;
1545	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
1546	iph->frag_off = df;
1547	iph->ttl = ttl;
1548	iph->protocol = sk->sk_protocol;
1549	ip_copy_addrs(iph, fl4);
1550	ip_select_ident(net, skb, sk);
1551
1552	if (opt) {
1553		iph->ihl += opt->optlen >> 2;
1554		ip_options_build(skb, opt, cork->addr, rt, 0);
1555	}
1556
1557	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1558	skb->mark = cork->mark;
1559	skb->tstamp = cork->transmit_time;
1560	/*
1561	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1562	 * on dst refcount
1563	 */
1564	cork->dst = NULL;
1565	skb_dst_set(skb, &rt->dst);
1566
1567	if (iph->protocol == IPPROTO_ICMP) {
1568		u8 icmp_type;
1569
1570		/* For such sockets, transhdrlen is zero when do ip_append_data(),
1571		 * so icmphdr does not in skb linear region and can not get icmp_type
1572		 * by icmp_hdr(skb)->type.
1573		 */
1574		if (sk->sk_type == SOCK_RAW && !inet_sk(sk)->hdrincl)
1575			icmp_type = fl4->fl4_icmp_type;
1576		else
1577			icmp_type = icmp_hdr(skb)->type;
1578		icmp_out_count(net, icmp_type);
1579	}
1580
1581	ip_cork_release(cork);
1582out:
1583	return skb;
1584}
1585
1586int ip_send_skb(struct net *net, struct sk_buff *skb)
1587{
1588	int err;
1589
1590	err = ip_local_out(net, skb->sk, skb);
1591	if (err) {
1592		if (err > 0)
1593			err = net_xmit_errno(err);
1594		if (err)
1595			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1596	}
1597
1598	return err;
1599}
1600
1601int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1602{
1603	struct sk_buff *skb;
1604
1605	skb = ip_finish_skb(sk, fl4);
1606	if (!skb)
1607		return 0;
1608
1609	/* Netfilter gets whole the not fragmented skb. */
1610	return ip_send_skb(sock_net(sk), skb);
1611}
1612
1613/*
1614 *	Throw away all pending data on the socket.
1615 */
1616static void __ip_flush_pending_frames(struct sock *sk,
1617				      struct sk_buff_head *queue,
1618				      struct inet_cork *cork)
1619{
1620	struct sk_buff *skb;
1621
1622	while ((skb = __skb_dequeue_tail(queue)) != NULL)
1623		kfree_skb(skb);
1624
1625	ip_cork_release(cork);
1626}
1627
1628void ip_flush_pending_frames(struct sock *sk)
1629{
1630	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1631}
1632
1633struct sk_buff *ip_make_skb(struct sock *sk,
1634			    struct flowi4 *fl4,
1635			    int getfrag(void *from, char *to, int offset,
1636					int len, int odd, struct sk_buff *skb),
1637			    void *from, int length, int transhdrlen,
1638			    struct ipcm_cookie *ipc, struct rtable **rtp,
1639			    struct inet_cork *cork, unsigned int flags)
1640{
1641	struct sk_buff_head queue;
1642	int err;
1643
1644	if (flags & MSG_PROBE)
1645		return NULL;
1646
1647	__skb_queue_head_init(&queue);
1648
1649	cork->flags = 0;
1650	cork->addr = 0;
1651	cork->opt = NULL;
1652	err = ip_setup_cork(sk, cork, ipc, rtp);
1653	if (err)
1654		return ERR_PTR(err);
1655
1656	err = __ip_append_data(sk, fl4, &queue, cork,
1657			       &current->task_frag, getfrag,
1658			       from, length, transhdrlen, flags);
1659	if (err) {
1660		__ip_flush_pending_frames(sk, &queue, cork);
1661		return ERR_PTR(err);
1662	}
1663
1664	return __ip_make_skb(sk, fl4, &queue, cork);
1665}
1666
1667/*
1668 *	Fetch data from kernel space and fill in checksum if needed.
1669 */
1670static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1671			      int len, int odd, struct sk_buff *skb)
1672{
1673	__wsum csum;
1674
1675	csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1676	skb->csum = csum_block_add(skb->csum, csum, odd);
1677	return 0;
1678}
1679
1680/*
1681 *	Generic function to send a packet as reply to another packet.
1682 *	Used to send some TCP resets/acks so far.
1683 */
1684void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1685			   const struct ip_options *sopt,
1686			   __be32 daddr, __be32 saddr,
1687			   const struct ip_reply_arg *arg,
1688			   unsigned int len, u64 transmit_time)
1689{
1690	struct ip_options_data replyopts;
1691	struct ipcm_cookie ipc;
1692	struct flowi4 fl4;
1693	struct rtable *rt = skb_rtable(skb);
1694	struct net *net = sock_net(sk);
1695	struct sk_buff *nskb;
1696	int err;
1697	int oif;
1698
1699	if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1700		return;
1701
1702	ipcm_init(&ipc);
1703	ipc.addr = daddr;
1704	ipc.sockc.transmit_time = transmit_time;
1705
1706	if (replyopts.opt.opt.optlen) {
1707		ipc.opt = &replyopts.opt;
1708
1709		if (replyopts.opt.opt.srr)
1710			daddr = replyopts.opt.opt.faddr;
1711	}
1712
1713	oif = arg->bound_dev_if;
1714	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1715		oif = skb->skb_iif;
1716
1717	flowi4_init_output(&fl4, oif,
1718			   IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1719			   RT_TOS(arg->tos),
1720			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1721			   ip_reply_arg_flowi_flags(arg),
1722			   daddr, saddr,
1723			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1724			   arg->uid);
1725	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1726	rt = ip_route_output_flow(net, &fl4, sk);
1727	if (IS_ERR(rt))
1728		return;
1729
1730	inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1731
1732	sk->sk_protocol = ip_hdr(skb)->protocol;
1733	sk->sk_bound_dev_if = arg->bound_dev_if;
1734	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1735	ipc.sockc.mark = fl4.flowi4_mark;
1736	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1737			     len, 0, &ipc, &rt, MSG_DONTWAIT);
1738	if (unlikely(err)) {
1739		ip_flush_pending_frames(sk);
1740		goto out;
1741	}
1742
1743	nskb = skb_peek(&sk->sk_write_queue);
1744	if (nskb) {
1745		if (arg->csumoffset >= 0)
1746			*((__sum16 *)skb_transport_header(nskb) +
1747			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1748								arg->csum));
1749		nskb->ip_summed = CHECKSUM_NONE;
1750		ip_push_pending_frames(sk, &fl4);
1751	}
1752out:
1753	ip_rt_put(rt);
1754}
1755
1756void __init ip_init(void)
1757{
1758	ip_rt_init();
1759	inet_initpeers();
1760
1761#if defined(CONFIG_IP_MULTICAST)
1762	igmp_mc_init();
1763#endif
1764}
1765