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 readability.
  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/gso.h>
  77#include <net/inetpeer.h>
  78#include <net/inet_ecn.h>
  79#include <net/lwtunnel.h>
  80#include <net/inet_dscp.h>
  81#include <linux/bpf-cgroup.h>
  82#include <linux/igmp.h>
  83#include <linux/netfilter_ipv4.h>
  84#include <linux/netfilter_bridge.h>
  85#include <linux/netlink.h>
  86#include <linux/tcp.h>
  87
  88static int
  89ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
  90	    unsigned int mtu,
  91	    int (*output)(struct net *, struct sock *, struct sk_buff *));
  92
  93/* Generate a checksum for an outgoing IP datagram. */
  94void ip_send_check(struct iphdr *iph)
  95{
  96	iph->check = 0;
  97	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
  98}
  99EXPORT_SYMBOL(ip_send_check);
 100
 101int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 102{
 103	struct iphdr *iph = ip_hdr(skb);
 104
 105	IP_INC_STATS(net, IPSTATS_MIB_OUTREQUESTS);
 106
 107	iph_set_totlen(iph, skb->len);
 108	ip_send_check(iph);
 109
 110	/* if egress device is enslaved to an L3 master device pass the
 111	 * skb to its handler for processing
 112	 */
 113	skb = l3mdev_ip_out(sk, skb);
 114	if (unlikely(!skb))
 115		return 0;
 116
 117	skb->protocol = htons(ETH_P_IP);
 118
 119	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
 120		       net, sk, skb, NULL, skb_dst(skb)->dev,
 121		       dst_output);
 122}
 123
 124int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 125{
 126	int err;
 127
 128	err = __ip_local_out(net, sk, skb);
 129	if (likely(err == 1))
 130		err = dst_output(net, sk, skb);
 131
 132	return err;
 133}
 134EXPORT_SYMBOL_GPL(ip_local_out);
 135
 136static inline int ip_select_ttl(const struct inet_sock *inet,
 137				const struct dst_entry *dst)
 138{
 139	int ttl = READ_ONCE(inet->uc_ttl);
 140
 141	if (ttl < 0)
 142		ttl = ip4_dst_hoplimit(dst);
 143	return ttl;
 144}
 145
 146/*
 147 *		Add an ip header to a skbuff and send it out.
 148 *
 149 */
 150int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
 151			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt,
 152			  u8 tos)
 153{
 154	const struct inet_sock *inet = inet_sk(sk);
 155	struct rtable *rt = skb_rtable(skb);
 156	struct net *net = sock_net(sk);
 157	struct iphdr *iph;
 158
 159	/* Build the IP header. */
 160	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
 161	skb_reset_network_header(skb);
 162	iph = ip_hdr(skb);
 163	iph->version  = 4;
 164	iph->ihl      = 5;
 165	iph->tos      = tos;
 166	iph->ttl      = ip_select_ttl(inet, &rt->dst);
 167	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
 168	iph->saddr    = saddr;
 169	iph->protocol = sk->sk_protocol;
 170	/* Do not bother generating IPID for small packets (eg SYNACK) */
 171	if (skb->len <= IPV4_MIN_MTU || ip_dont_fragment(sk, &rt->dst)) {
 172		iph->frag_off = htons(IP_DF);
 173		iph->id = 0;
 174	} else {
 175		iph->frag_off = 0;
 176		/* TCP packets here are SYNACK with fat IPv4/TCP options.
 177		 * Avoid using the hashed IP ident generator.
 178		 */
 179		if (sk->sk_protocol == IPPROTO_TCP)
 180			iph->id = (__force __be16)get_random_u16();
 181		else
 182			__ip_select_ident(net, iph, 1);
 183	}
 184
 185	if (opt && opt->opt.optlen) {
 186		iph->ihl += opt->opt.optlen>>2;
 187		ip_options_build(skb, &opt->opt, daddr, rt);
 188	}
 189
 190	skb->priority = READ_ONCE(sk->sk_priority);
 191	if (!skb->mark)
 192		skb->mark = READ_ONCE(sk->sk_mark);
 193
 194	/* Send it out. */
 195	return ip_local_out(net, skb->sk, skb);
 196}
 197EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
 198
 199static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
 200{
 201	struct dst_entry *dst = skb_dst(skb);
 202	struct rtable *rt = dst_rtable(dst);
 203	struct net_device *dev = dst->dev;
 204	unsigned int hh_len = LL_RESERVED_SPACE(dev);
 205	struct neighbour *neigh;
 206	bool is_v6gw = false;
 207
 208	if (rt->rt_type == RTN_MULTICAST) {
 209		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
 210	} else if (rt->rt_type == RTN_BROADCAST)
 211		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
 212
 213	/* OUTOCTETS should be counted after fragment */
 214	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
 215
 216	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
 217		skb = skb_expand_head(skb, hh_len);
 218		if (!skb)
 219			return -ENOMEM;
 220	}
 221
 222	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
 223		int res = lwtunnel_xmit(skb);
 224
 225		if (res != LWTUNNEL_XMIT_CONTINUE)
 226			return res;
 227	}
 228
 229	rcu_read_lock();
 230	neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
 231	if (!IS_ERR(neigh)) {
 232		int res;
 233
 234		sock_confirm_neigh(skb, neigh);
 235		/* if crossing protocols, can not use the cached header */
 236		res = neigh_output(neigh, skb, is_v6gw);
 237		rcu_read_unlock();
 238		return res;
 239	}
 240	rcu_read_unlock();
 241
 242	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
 243			    __func__);
 244	kfree_skb_reason(skb, SKB_DROP_REASON_NEIGH_CREATEFAIL);
 245	return PTR_ERR(neigh);
 246}
 247
 248static int ip_finish_output_gso(struct net *net, struct sock *sk,
 249				struct sk_buff *skb, unsigned int mtu)
 250{
 251	struct sk_buff *segs, *nskb;
 252	netdev_features_t features;
 253	int ret = 0;
 254
 255	/* common case: seglen is <= mtu
 256	 */
 257	if (skb_gso_validate_network_len(skb, mtu))
 258		return ip_finish_output2(net, sk, skb);
 259
 260	/* Slowpath -  GSO segment length exceeds the egress MTU.
 261	 *
 262	 * This can happen in several cases:
 263	 *  - Forwarding of a TCP GRO skb, when DF flag is not set.
 264	 *  - Forwarding of an skb that arrived on a virtualization interface
 265	 *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
 266	 *    stack.
 267	 *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
 268	 *    interface with a smaller MTU.
 269	 *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
 270	 *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
 271	 *    insufficient MTU.
 272	 */
 273	features = netif_skb_features(skb);
 274	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_GSO_CB_OFFSET);
 275	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
 276	if (IS_ERR_OR_NULL(segs)) {
 277		kfree_skb(skb);
 278		return -ENOMEM;
 279	}
 280
 281	consume_skb(skb);
 282
 283	skb_list_walk_safe(segs, segs, nskb) {
 284		int err;
 285
 286		skb_mark_not_on_list(segs);
 287		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
 288
 289		if (err && ret == 0)
 290			ret = err;
 291	}
 292
 293	return ret;
 294}
 295
 296static int __ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 297{
 298	unsigned int mtu;
 299
 300#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
 301	/* Policy lookup after SNAT yielded a new policy */
 302	if (skb_dst(skb)->xfrm) {
 303		IPCB(skb)->flags |= IPSKB_REROUTED;
 304		return dst_output(net, sk, skb);
 305	}
 306#endif
 307	mtu = ip_skb_dst_mtu(sk, skb);
 308	if (skb_is_gso(skb))
 309		return ip_finish_output_gso(net, sk, skb, mtu);
 310
 311	if (skb->len > mtu || IPCB(skb)->frag_max_size)
 312		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
 313
 314	return ip_finish_output2(net, sk, skb);
 315}
 316
 317static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 318{
 319	int ret;
 320
 321	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
 322	switch (ret) {
 323	case NET_XMIT_SUCCESS:
 324		return __ip_finish_output(net, sk, skb);
 325	case NET_XMIT_CN:
 326		return __ip_finish_output(net, sk, skb) ? : ret;
 327	default:
 328		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
 329		return ret;
 330	}
 331}
 332
 333static int ip_mc_finish_output(struct net *net, struct sock *sk,
 334			       struct sk_buff *skb)
 335{
 336	struct rtable *new_rt;
 337	bool do_cn = false;
 338	int ret, err;
 339
 340	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
 341	switch (ret) {
 342	case NET_XMIT_CN:
 343		do_cn = true;
 344		fallthrough;
 345	case NET_XMIT_SUCCESS:
 346		break;
 347	default:
 348		kfree_skb_reason(skb, SKB_DROP_REASON_BPF_CGROUP_EGRESS);
 349		return ret;
 350	}
 351
 352	/* Reset rt_iif so that inet_iif() will return skb->skb_iif. Setting
 353	 * this to non-zero causes ipi_ifindex in in_pktinfo to be overwritten,
 354	 * see ipv4_pktinfo_prepare().
 355	 */
 356	new_rt = rt_dst_clone(net->loopback_dev, skb_rtable(skb));
 357	if (new_rt) {
 358		new_rt->rt_iif = 0;
 359		skb_dst_drop(skb);
 360		skb_dst_set(skb, &new_rt->dst);
 361	}
 362
 363	err = dev_loopback_xmit(net, sk, skb);
 364	return (do_cn && err) ? ret : err;
 365}
 366
 367int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 368{
 369	struct rtable *rt = skb_rtable(skb);
 370	struct net_device *dev = rt->dst.dev;
 371
 372	/*
 373	 *	If the indicated interface is up and running, send the packet.
 374	 */
 375	skb->dev = dev;
 376	skb->protocol = htons(ETH_P_IP);
 377
 378	/*
 379	 *	Multicasts are looped back for other local users
 380	 */
 381
 382	if (rt->rt_flags&RTCF_MULTICAST) {
 383		if (sk_mc_loop(sk)
 384#ifdef CONFIG_IP_MROUTE
 385		/* Small optimization: do not loopback not local frames,
 386		   which returned after forwarding; they will be  dropped
 387		   by ip_mr_input in any case.
 388		   Note, that local frames are looped back to be delivered
 389		   to local recipients.
 390
 391		   This check is duplicated in ip_mr_input at the moment.
 392		 */
 393		    &&
 394		    ((rt->rt_flags & RTCF_LOCAL) ||
 395		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
 396#endif
 397		   ) {
 398			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 399			if (newskb)
 400				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 401					net, sk, newskb, NULL, newskb->dev,
 402					ip_mc_finish_output);
 403		}
 404
 405		/* Multicasts with ttl 0 must not go beyond the host */
 406
 407		if (ip_hdr(skb)->ttl == 0) {
 408			kfree_skb(skb);
 409			return 0;
 410		}
 411	}
 412
 413	if (rt->rt_flags&RTCF_BROADCAST) {
 414		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
 415		if (newskb)
 416			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 417				net, sk, newskb, NULL, newskb->dev,
 418				ip_mc_finish_output);
 419	}
 420
 421	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 422			    net, sk, skb, NULL, skb->dev,
 423			    ip_finish_output,
 424			    !(IPCB(skb)->flags & IPSKB_REROUTED));
 425}
 426
 427int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 428{
 429	struct net_device *dev = skb_dst(skb)->dev, *indev = skb->dev;
 430
 431	skb->dev = dev;
 432	skb->protocol = htons(ETH_P_IP);
 433
 434	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
 435			    net, sk, skb, indev, dev,
 436			    ip_finish_output,
 437			    !(IPCB(skb)->flags & IPSKB_REROUTED));
 438}
 439EXPORT_SYMBOL(ip_output);
 440
 441/*
 442 * copy saddr and daddr, possibly using 64bit load/stores
 443 * Equivalent to :
 444 *   iph->saddr = fl4->saddr;
 445 *   iph->daddr = fl4->daddr;
 446 */
 447static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
 448{
 449	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
 450		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
 451
 452	iph->saddr = fl4->saddr;
 453	iph->daddr = fl4->daddr;
 454}
 455
 456/* Note: skb->sk can be different from sk, in case of tunnels */
 457int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
 458		    __u8 tos)
 459{
 460	struct inet_sock *inet = inet_sk(sk);
 461	struct net *net = sock_net(sk);
 462	struct ip_options_rcu *inet_opt;
 463	struct flowi4 *fl4;
 464	struct rtable *rt;
 465	struct iphdr *iph;
 466	int res;
 467
 468	/* Skip all of this if the packet is already routed,
 469	 * f.e. by something like SCTP.
 470	 */
 471	rcu_read_lock();
 472	inet_opt = rcu_dereference(inet->inet_opt);
 473	fl4 = &fl->u.ip4;
 474	rt = skb_rtable(skb);
 475	if (rt)
 476		goto packet_routed;
 477
 478	/* Make sure we can route this packet. */
 479	rt = dst_rtable(__sk_dst_check(sk, 0));
 480	if (!rt) {
 481		__be32 daddr;
 482
 483		/* Use correct destination address if we have options. */
 484		daddr = inet->inet_daddr;
 485		if (inet_opt && inet_opt->opt.srr)
 486			daddr = inet_opt->opt.faddr;
 487
 488		/* If this fails, retransmit mechanism of transport layer will
 489		 * keep trying until route appears or the connection times
 490		 * itself out.
 491		 */
 492		rt = ip_route_output_ports(net, fl4, sk,
 493					   daddr, inet->inet_saddr,
 494					   inet->inet_dport,
 495					   inet->inet_sport,
 496					   sk->sk_protocol,
 497					   tos & INET_DSCP_MASK,
 498					   sk->sk_bound_dev_if);
 499		if (IS_ERR(rt))
 500			goto no_route;
 501		sk_setup_caps(sk, &rt->dst);
 502	}
 503	skb_dst_set_noref(skb, &rt->dst);
 504
 505packet_routed:
 506	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
 507		goto no_route;
 508
 509	/* OK, we know where to send it, allocate and build IP header. */
 510	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
 511	skb_reset_network_header(skb);
 512	iph = ip_hdr(skb);
 513	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
 514	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
 515		iph->frag_off = htons(IP_DF);
 516	else
 517		iph->frag_off = 0;
 518	iph->ttl      = ip_select_ttl(inet, &rt->dst);
 519	iph->protocol = sk->sk_protocol;
 520	ip_copy_addrs(iph, fl4);
 521
 522	/* Transport layer set skb->h.foo itself. */
 523
 524	if (inet_opt && inet_opt->opt.optlen) {
 525		iph->ihl += inet_opt->opt.optlen >> 2;
 526		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt);
 527	}
 528
 529	ip_select_ident_segs(net, skb, sk,
 530			     skb_shinfo(skb)->gso_segs ?: 1);
 531
 532	/* TODO : should we use skb->sk here instead of sk ? */
 533	skb->priority = READ_ONCE(sk->sk_priority);
 534	skb->mark = READ_ONCE(sk->sk_mark);
 535
 536	res = ip_local_out(net, sk, skb);
 537	rcu_read_unlock();
 538	return res;
 539
 540no_route:
 541	rcu_read_unlock();
 542	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 543	kfree_skb_reason(skb, SKB_DROP_REASON_IP_OUTNOROUTES);
 544	return -EHOSTUNREACH;
 545}
 546EXPORT_SYMBOL(__ip_queue_xmit);
 547
 548int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
 549{
 550	return __ip_queue_xmit(sk, skb, fl, READ_ONCE(inet_sk(sk)->tos));
 551}
 552EXPORT_SYMBOL(ip_queue_xmit);
 553
 554static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
 555{
 556	to->pkt_type = from->pkt_type;
 557	to->priority = from->priority;
 558	to->protocol = from->protocol;
 559	to->skb_iif = from->skb_iif;
 560	skb_dst_drop(to);
 561	skb_dst_copy(to, from);
 562	to->dev = from->dev;
 563	to->mark = from->mark;
 564
 565	skb_copy_hash(to, from);
 566
 567#ifdef CONFIG_NET_SCHED
 568	to->tc_index = from->tc_index;
 569#endif
 570	nf_copy(to, from);
 571	skb_ext_copy(to, from);
 572#if IS_ENABLED(CONFIG_IP_VS)
 573	to->ipvs_property = from->ipvs_property;
 574#endif
 575	skb_copy_secmark(to, from);
 576}
 577
 578static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 579		       unsigned int mtu,
 580		       int (*output)(struct net *, struct sock *, struct sk_buff *))
 581{
 582	struct iphdr *iph = ip_hdr(skb);
 583
 584	if ((iph->frag_off & htons(IP_DF)) == 0)
 585		return ip_do_fragment(net, sk, skb, output);
 586
 587	if (unlikely(!skb->ignore_df ||
 588		     (IPCB(skb)->frag_max_size &&
 589		      IPCB(skb)->frag_max_size > mtu))) {
 590		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 591		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 592			  htonl(mtu));
 593		kfree_skb(skb);
 594		return -EMSGSIZE;
 595	}
 596
 597	return ip_do_fragment(net, sk, skb, output);
 598}
 599
 600void ip_fraglist_init(struct sk_buff *skb, struct iphdr *iph,
 601		      unsigned int hlen, struct ip_fraglist_iter *iter)
 602{
 603	unsigned int first_len = skb_pagelen(skb);
 604
 605	iter->frag = skb_shinfo(skb)->frag_list;
 606	skb_frag_list_init(skb);
 607
 608	iter->offset = 0;
 609	iter->iph = iph;
 610	iter->hlen = hlen;
 611
 612	skb->data_len = first_len - skb_headlen(skb);
 613	skb->len = first_len;
 614	iph->tot_len = htons(first_len);
 615	iph->frag_off = htons(IP_MF);
 616	ip_send_check(iph);
 617}
 618EXPORT_SYMBOL(ip_fraglist_init);
 619
 620void ip_fraglist_prepare(struct sk_buff *skb, struct ip_fraglist_iter *iter)
 621{
 622	unsigned int hlen = iter->hlen;
 623	struct iphdr *iph = iter->iph;
 624	struct sk_buff *frag;
 625
 626	frag = iter->frag;
 627	frag->ip_summed = CHECKSUM_NONE;
 628	skb_reset_transport_header(frag);
 629	__skb_push(frag, hlen);
 630	skb_reset_network_header(frag);
 631	memcpy(skb_network_header(frag), iph, hlen);
 632	iter->iph = ip_hdr(frag);
 633	iph = iter->iph;
 634	iph->tot_len = htons(frag->len);
 635	ip_copy_metadata(frag, skb);
 636	iter->offset += skb->len - hlen;
 637	iph->frag_off = htons(iter->offset >> 3);
 638	if (frag->next)
 639		iph->frag_off |= htons(IP_MF);
 640	/* Ready, complete checksum */
 641	ip_send_check(iph);
 642}
 643EXPORT_SYMBOL(ip_fraglist_prepare);
 644
 645void ip_frag_init(struct sk_buff *skb, unsigned int hlen,
 646		  unsigned int ll_rs, unsigned int mtu, bool DF,
 647		  struct ip_frag_state *state)
 648{
 649	struct iphdr *iph = ip_hdr(skb);
 650
 651	state->DF = DF;
 652	state->hlen = hlen;
 653	state->ll_rs = ll_rs;
 654	state->mtu = mtu;
 655
 656	state->left = skb->len - hlen;	/* Space per frame */
 657	state->ptr = hlen;		/* Where to start from */
 658
 659	state->offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
 660	state->not_last_frag = iph->frag_off & htons(IP_MF);
 661}
 662EXPORT_SYMBOL(ip_frag_init);
 663
 664static void ip_frag_ipcb(struct sk_buff *from, struct sk_buff *to,
 665			 bool first_frag)
 666{
 667	/* Copy the flags to each fragment. */
 668	IPCB(to)->flags = IPCB(from)->flags;
 669
 670	/* ANK: dirty, but effective trick. Upgrade options only if
 671	 * the segment to be fragmented was THE FIRST (otherwise,
 672	 * options are already fixed) and make it ONCE
 673	 * on the initial skb, so that all the following fragments
 674	 * will inherit fixed options.
 675	 */
 676	if (first_frag)
 677		ip_options_fragment(from);
 678}
 679
 680struct sk_buff *ip_frag_next(struct sk_buff *skb, struct ip_frag_state *state)
 681{
 682	unsigned int len = state->left;
 683	struct sk_buff *skb2;
 684	struct iphdr *iph;
 685
 686	/* IF: it doesn't fit, use 'mtu' - the data space left */
 687	if (len > state->mtu)
 688		len = state->mtu;
 689	/* IF: we are not sending up to and including the packet end
 690	   then align the next start on an eight byte boundary */
 691	if (len < state->left)	{
 692		len &= ~7;
 693	}
 694
 695	/* Allocate buffer */
 696	skb2 = alloc_skb(len + state->hlen + state->ll_rs, GFP_ATOMIC);
 697	if (!skb2)
 698		return ERR_PTR(-ENOMEM);
 699
 700	/*
 701	 *	Set up data on packet
 702	 */
 703
 704	ip_copy_metadata(skb2, skb);
 705	skb_reserve(skb2, state->ll_rs);
 706	skb_put(skb2, len + state->hlen);
 707	skb_reset_network_header(skb2);
 708	skb2->transport_header = skb2->network_header + state->hlen;
 709
 710	/*
 711	 *	Charge the memory for the fragment to any owner
 712	 *	it might possess
 713	 */
 714
 715	if (skb->sk)
 716		skb_set_owner_w(skb2, skb->sk);
 717
 718	/*
 719	 *	Copy the packet header into the new buffer.
 720	 */
 721
 722	skb_copy_from_linear_data(skb, skb_network_header(skb2), state->hlen);
 723
 724	/*
 725	 *	Copy a block of the IP datagram.
 726	 */
 727	if (skb_copy_bits(skb, state->ptr, skb_transport_header(skb2), len))
 728		BUG();
 729	state->left -= len;
 730
 731	/*
 732	 *	Fill in the new header fields.
 733	 */
 734	iph = ip_hdr(skb2);
 735	iph->frag_off = htons((state->offset >> 3));
 736	if (state->DF)
 737		iph->frag_off |= htons(IP_DF);
 738
 739	/*
 740	 *	Added AC : If we are fragmenting a fragment that's not the
 741	 *		   last fragment then keep MF on each bit
 742	 */
 743	if (state->left > 0 || state->not_last_frag)
 744		iph->frag_off |= htons(IP_MF);
 745	state->ptr += len;
 746	state->offset += len;
 747
 748	iph->tot_len = htons(len + state->hlen);
 749
 750	ip_send_check(iph);
 751
 752	return skb2;
 753}
 754EXPORT_SYMBOL(ip_frag_next);
 755
 756/*
 757 *	This IP datagram is too large to be sent in one piece.  Break it up into
 758 *	smaller pieces (each of size equal to IP header plus
 759 *	a block of the data of the original IP data part) that will yet fit in a
 760 *	single device frame, and queue such a frame for sending.
 761 */
 762
 763int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
 764		   int (*output)(struct net *, struct sock *, struct sk_buff *))
 765{
 766	struct iphdr *iph;
 767	struct sk_buff *skb2;
 768	u8 tstamp_type = skb->tstamp_type;
 769	struct rtable *rt = skb_rtable(skb);
 770	unsigned int mtu, hlen, ll_rs;
 771	struct ip_fraglist_iter iter;
 772	ktime_t tstamp = skb->tstamp;
 773	struct ip_frag_state state;
 774	int err = 0;
 775
 776	/* for offloaded checksums cleanup checksum before fragmentation */
 777	if (skb->ip_summed == CHECKSUM_PARTIAL &&
 778	    (err = skb_checksum_help(skb)))
 779		goto fail;
 780
 781	/*
 782	 *	Point into the IP datagram header.
 783	 */
 784
 785	iph = ip_hdr(skb);
 786
 787	mtu = ip_skb_dst_mtu(sk, skb);
 788	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
 789		mtu = IPCB(skb)->frag_max_size;
 790
 791	/*
 792	 *	Setup starting values.
 793	 */
 794
 795	hlen = iph->ihl * 4;
 796	mtu = mtu - hlen;	/* Size of data space */
 797	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 798	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
 799
 800	/* When frag_list is given, use it. First, check its validity:
 801	 * some transformers could create wrong frag_list or break existing
 802	 * one, it is not prohibited. In this case fall back to copying.
 803	 *
 804	 * LATER: this step can be merged to real generation of fragments,
 805	 * we can switch to copy when see the first bad fragment.
 806	 */
 807	if (skb_has_frag_list(skb)) {
 808		struct sk_buff *frag, *frag2;
 809		unsigned int first_len = skb_pagelen(skb);
 810
 811		if (first_len - hlen > mtu ||
 812		    ((first_len - hlen) & 7) ||
 813		    ip_is_fragment(iph) ||
 814		    skb_cloned(skb) ||
 815		    skb_headroom(skb) < ll_rs)
 816			goto slow_path;
 817
 818		skb_walk_frags(skb, frag) {
 819			/* Correct geometry. */
 820			if (frag->len > mtu ||
 821			    ((frag->len & 7) && frag->next) ||
 822			    skb_headroom(frag) < hlen + ll_rs)
 823				goto slow_path_clean;
 824
 825			/* Partially cloned skb? */
 826			if (skb_shared(frag))
 827				goto slow_path_clean;
 828
 829			BUG_ON(frag->sk);
 830			if (skb->sk) {
 831				frag->sk = skb->sk;
 832				frag->destructor = sock_wfree;
 833			}
 834			skb->truesize -= frag->truesize;
 835		}
 836
 837		/* Everything is OK. Generate! */
 838		ip_fraglist_init(skb, iph, hlen, &iter);
 839
 840		for (;;) {
 841			/* Prepare header of the next frame,
 842			 * before previous one went down. */
 843			if (iter.frag) {
 844				bool first_frag = (iter.offset == 0);
 845
 846				IPCB(iter.frag)->flags = IPCB(skb)->flags;
 847				ip_fraglist_prepare(skb, &iter);
 848				if (first_frag && IPCB(skb)->opt.optlen) {
 849					/* ipcb->opt is not populated for frags
 850					 * coming from __ip_make_skb(),
 851					 * ip_options_fragment() needs optlen
 852					 */
 853					IPCB(iter.frag)->opt.optlen =
 854						IPCB(skb)->opt.optlen;
 855					ip_options_fragment(iter.frag);
 856					ip_send_check(iter.iph);
 857				}
 858			}
 859
 860			skb_set_delivery_time(skb, tstamp, tstamp_type);
 861			err = output(net, sk, skb);
 862
 863			if (!err)
 864				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
 865			if (err || !iter.frag)
 866				break;
 867
 868			skb = ip_fraglist_next(&iter);
 869		}
 870
 871		if (err == 0) {
 872			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
 873			return 0;
 874		}
 875
 876		kfree_skb_list(iter.frag);
 877
 878		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 879		return err;
 880
 881slow_path_clean:
 882		skb_walk_frags(skb, frag2) {
 883			if (frag2 == frag)
 884				break;
 885			frag2->sk = NULL;
 886			frag2->destructor = NULL;
 887			skb->truesize += frag2->truesize;
 888		}
 889	}
 890
 891slow_path:
 892	/*
 893	 *	Fragment the datagram.
 894	 */
 895
 896	ip_frag_init(skb, hlen, ll_rs, mtu, IPCB(skb)->flags & IPSKB_FRAG_PMTU,
 897		     &state);
 898
 899	/*
 900	 *	Keep copying data until we run out.
 901	 */
 902
 903	while (state.left > 0) {
 904		bool first_frag = (state.offset == 0);
 905
 906		skb2 = ip_frag_next(skb, &state);
 907		if (IS_ERR(skb2)) {
 908			err = PTR_ERR(skb2);
 909			goto fail;
 910		}
 911		ip_frag_ipcb(skb, skb2, first_frag);
 912
 913		/*
 914		 *	Put this fragment into the sending queue.
 915		 */
 916		skb_set_delivery_time(skb2, tstamp, tstamp_type);
 917		err = output(net, sk, skb2);
 918		if (err)
 919			goto fail;
 920
 921		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
 922	}
 923	consume_skb(skb);
 924	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
 925	return err;
 926
 927fail:
 928	kfree_skb(skb);
 929	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
 930	return err;
 931}
 932EXPORT_SYMBOL(ip_do_fragment);
 933
 934int
 935ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 936{
 937	struct msghdr *msg = from;
 938
 939	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 940		if (!copy_from_iter_full(to, len, &msg->msg_iter))
 941			return -EFAULT;
 942	} else {
 943		__wsum csum = 0;
 944		if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
 945			return -EFAULT;
 946		skb->csum = csum_block_add(skb->csum, csum, odd);
 947	}
 948	return 0;
 949}
 950EXPORT_SYMBOL(ip_generic_getfrag);
 951
 952static int __ip_append_data(struct sock *sk,
 953			    struct flowi4 *fl4,
 954			    struct sk_buff_head *queue,
 955			    struct inet_cork *cork,
 956			    struct page_frag *pfrag,
 957			    int getfrag(void *from, char *to, int offset,
 958					int len, int odd, struct sk_buff *skb),
 959			    void *from, int length, int transhdrlen,
 960			    unsigned int flags)
 961{
 962	struct inet_sock *inet = inet_sk(sk);
 963	struct ubuf_info *uarg = NULL;
 964	struct sk_buff *skb;
 965	struct ip_options *opt = cork->opt;
 966	int hh_len;
 967	int exthdrlen;
 968	int mtu;
 969	int copy;
 970	int err;
 971	int offset = 0;
 972	bool zc = false;
 973	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
 974	int csummode = CHECKSUM_NONE;
 975	struct rtable *rt = dst_rtable(cork->dst);
 976	bool paged, hold_tskey = false, extra_uref = false;
 977	unsigned int wmem_alloc_delta = 0;
 978	u32 tskey = 0;
 979
 980	skb = skb_peek_tail(queue);
 981
 982	exthdrlen = !skb ? rt->dst.header_len : 0;
 983	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
 984	paged = !!cork->gso_size;
 985
 986	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
 987
 988	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
 989	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
 990	maxnonfragsize = ip_sk_ignore_df(sk) ? IP_MAX_MTU : mtu;
 991
 992	if (cork->length + length > maxnonfragsize - fragheaderlen) {
 993		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
 994			       mtu - (opt ? opt->optlen : 0));
 995		return -EMSGSIZE;
 996	}
 997
 998	/*
 999	 * transhdrlen > 0 means that this is the first fragment and we wish
1000	 * it won't be fragmented in the future.
1001	 */
1002	if (transhdrlen &&
1003	    length + fragheaderlen <= mtu &&
1004	    rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
1005	    (!(flags & MSG_MORE) || cork->gso_size) &&
1006	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
1007		csummode = CHECKSUM_PARTIAL;
1008
1009	if ((flags & MSG_ZEROCOPY) && length) {
1010		struct msghdr *msg = from;
1011
1012		if (getfrag == ip_generic_getfrag && msg->msg_ubuf) {
1013			if (skb_zcopy(skb) && msg->msg_ubuf != skb_zcopy(skb))
1014				return -EINVAL;
1015
1016			/* Leave uarg NULL if can't zerocopy, callers should
1017			 * be able to handle it.
1018			 */
1019			if ((rt->dst.dev->features & NETIF_F_SG) &&
1020			    csummode == CHECKSUM_PARTIAL) {
1021				paged = true;
1022				zc = true;
1023				uarg = msg->msg_ubuf;
1024			}
1025		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
1026			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
1027			if (!uarg)
1028				return -ENOBUFS;
1029			extra_uref = !skb_zcopy(skb);	/* only ref on new uarg */
1030			if (rt->dst.dev->features & NETIF_F_SG &&
1031			    csummode == CHECKSUM_PARTIAL) {
1032				paged = true;
1033				zc = true;
1034			} else {
1035				uarg_to_msgzc(uarg)->zerocopy = 0;
1036				skb_zcopy_set(skb, uarg, &extra_uref);
1037			}
1038		}
1039	} else if ((flags & MSG_SPLICE_PAGES) && length) {
1040		if (inet_test_bit(HDRINCL, sk))
1041			return -EPERM;
1042		if (rt->dst.dev->features & NETIF_F_SG &&
1043		    getfrag == ip_generic_getfrag)
1044			/* We need an empty buffer to attach stuff to */
1045			paged = true;
1046		else
1047			flags &= ~MSG_SPLICE_PAGES;
1048	}
1049
1050	cork->length += length;
1051
1052	if (cork->tx_flags & SKBTX_ANY_TSTAMP &&
1053	    READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
1054		if (cork->flags & IPCORK_TS_OPT_ID) {
1055			tskey = cork->ts_opt_id;
1056		} else {
1057			tskey = atomic_inc_return(&sk->sk_tskey) - 1;
1058			hold_tskey = true;
1059		}
1060	}
1061
1062	/* So, what's going on in the loop below?
1063	 *
1064	 * We use calculated fragment length to generate chained skb,
1065	 * each of segments is IP fragment ready for sending to network after
1066	 * adding appropriate IP header.
1067	 */
1068
1069	if (!skb)
1070		goto alloc_new_skb;
1071
1072	while (length > 0) {
1073		/* Check if the remaining data fits into current packet. */
1074		copy = mtu - skb->len;
1075		if (copy < length)
1076			copy = maxfraglen - skb->len;
1077		if (copy <= 0) {
1078			char *data;
1079			unsigned int datalen;
1080			unsigned int fraglen;
1081			unsigned int fraggap;
1082			unsigned int alloclen, alloc_extra;
1083			unsigned int pagedlen;
1084			struct sk_buff *skb_prev;
1085alloc_new_skb:
1086			skb_prev = skb;
1087			if (skb_prev)
1088				fraggap = skb_prev->len - maxfraglen;
1089			else
1090				fraggap = 0;
1091
1092			/*
1093			 * If remaining data exceeds the mtu,
1094			 * we know we need more fragment(s).
1095			 */
1096			datalen = length + fraggap;
1097			if (datalen > mtu - fragheaderlen)
1098				datalen = maxfraglen - fragheaderlen;
1099			fraglen = datalen + fragheaderlen;
1100			pagedlen = 0;
1101
1102			alloc_extra = hh_len + 15;
1103			alloc_extra += exthdrlen;
1104
1105			/* The last fragment gets additional space at tail.
1106			 * Note, with MSG_MORE we overallocate on fragments,
1107			 * because we have no idea what fragment will be
1108			 * the last.
1109			 */
1110			if (datalen == length + fraggap)
1111				alloc_extra += rt->dst.trailer_len;
1112
1113			if ((flags & MSG_MORE) &&
1114			    !(rt->dst.dev->features&NETIF_F_SG))
1115				alloclen = mtu;
1116			else if (!paged &&
1117				 (fraglen + alloc_extra < SKB_MAX_ALLOC ||
1118				  !(rt->dst.dev->features & NETIF_F_SG)))
1119				alloclen = fraglen;
1120			else {
1121				alloclen = fragheaderlen + transhdrlen;
1122				pagedlen = datalen - transhdrlen;
1123			}
1124
1125			alloclen += alloc_extra;
1126
1127			if (transhdrlen) {
1128				skb = sock_alloc_send_skb(sk, alloclen,
1129						(flags & MSG_DONTWAIT), &err);
1130			} else {
1131				skb = NULL;
1132				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
1133				    2 * sk->sk_sndbuf)
1134					skb = alloc_skb(alloclen,
1135							sk->sk_allocation);
1136				if (unlikely(!skb))
1137					err = -ENOBUFS;
1138			}
1139			if (!skb)
1140				goto error;
1141
1142			/*
1143			 *	Fill in the control structures
1144			 */
1145			skb->ip_summed = csummode;
1146			skb->csum = 0;
1147			skb_reserve(skb, hh_len);
1148
1149			/*
1150			 *	Find where to start putting bytes.
1151			 */
1152			data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1153			skb_set_network_header(skb, exthdrlen);
1154			skb->transport_header = (skb->network_header +
1155						 fragheaderlen);
1156			data += fragheaderlen + exthdrlen;
1157
1158			if (fraggap) {
1159				skb->csum = skb_copy_and_csum_bits(
1160					skb_prev, maxfraglen,
1161					data + transhdrlen, fraggap);
1162				skb_prev->csum = csum_sub(skb_prev->csum,
1163							  skb->csum);
1164				data += fraggap;
1165				pskb_trim_unique(skb_prev, maxfraglen);
1166			}
1167
1168			copy = datalen - transhdrlen - fraggap - pagedlen;
1169			/* [!] NOTE: copy will be negative if pagedlen>0
1170			 * because then the equation reduces to -fraggap.
1171			 */
1172			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
1173				err = -EFAULT;
1174				kfree_skb(skb);
1175				goto error;
1176			} else if (flags & MSG_SPLICE_PAGES) {
1177				copy = 0;
1178			}
1179
1180			offset += copy;
1181			length -= copy + transhdrlen;
1182			transhdrlen = 0;
1183			exthdrlen = 0;
1184			csummode = CHECKSUM_NONE;
1185
1186			/* only the initial fragment is time stamped */
1187			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1188			cork->tx_flags = 0;
1189			skb_shinfo(skb)->tskey = tskey;
1190			tskey = 0;
1191			skb_zcopy_set(skb, uarg, &extra_uref);
1192
1193			if ((flags & MSG_CONFIRM) && !skb_prev)
1194				skb_set_dst_pending_confirm(skb, 1);
1195
1196			/*
1197			 * Put the packet on the pending queue.
1198			 */
1199			if (!skb->destructor) {
1200				skb->destructor = sock_wfree;
1201				skb->sk = sk;
1202				wmem_alloc_delta += skb->truesize;
1203			}
1204			__skb_queue_tail(queue, skb);
1205			continue;
1206		}
1207
1208		if (copy > length)
1209			copy = length;
1210
1211		if (!(rt->dst.dev->features&NETIF_F_SG) &&
1212		    skb_tailroom(skb) >= copy) {
1213			unsigned int off;
1214
1215			off = skb->len;
1216			if (getfrag(from, skb_put(skb, copy),
1217					offset, copy, off, skb) < 0) {
1218				__skb_trim(skb, off);
1219				err = -EFAULT;
1220				goto error;
1221			}
1222		} else if (flags & MSG_SPLICE_PAGES) {
1223			struct msghdr *msg = from;
1224
1225			err = -EIO;
1226			if (WARN_ON_ONCE(copy > msg->msg_iter.count))
1227				goto error;
1228
1229			err = skb_splice_from_iter(skb, &msg->msg_iter, copy,
1230						   sk->sk_allocation);
1231			if (err < 0)
1232				goto error;
1233			copy = err;
1234			wmem_alloc_delta += copy;
1235		} else if (!zc) {
1236			int i = skb_shinfo(skb)->nr_frags;
1237
1238			err = -ENOMEM;
1239			if (!sk_page_frag_refill(sk, pfrag))
1240				goto error;
1241
1242			skb_zcopy_downgrade_managed(skb);
1243			if (!skb_can_coalesce(skb, i, pfrag->page,
1244					      pfrag->offset)) {
1245				err = -EMSGSIZE;
1246				if (i == MAX_SKB_FRAGS)
1247					goto error;
1248
1249				__skb_fill_page_desc(skb, i, pfrag->page,
1250						     pfrag->offset, 0);
1251				skb_shinfo(skb)->nr_frags = ++i;
1252				get_page(pfrag->page);
1253			}
1254			copy = min_t(int, copy, pfrag->size - pfrag->offset);
1255			if (getfrag(from,
1256				    page_address(pfrag->page) + pfrag->offset,
1257				    offset, copy, skb->len, skb) < 0)
1258				goto error_efault;
1259
1260			pfrag->offset += copy;
1261			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
1262			skb_len_add(skb, copy);
1263			wmem_alloc_delta += copy;
1264		} else {
1265			err = skb_zerocopy_iter_dgram(skb, from, copy);
1266			if (err < 0)
1267				goto error;
1268		}
1269		offset += copy;
1270		length -= copy;
1271	}
1272
1273	if (wmem_alloc_delta)
1274		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1275	return 0;
1276
1277error_efault:
1278	err = -EFAULT;
1279error:
1280	net_zcopy_put_abort(uarg, extra_uref);
1281	cork->length -= length;
1282	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1283	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1284	if (hold_tskey)
1285		atomic_dec(&sk->sk_tskey);
1286	return err;
1287}
1288
1289static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
1290			 struct ipcm_cookie *ipc, struct rtable **rtp)
1291{
1292	struct ip_options_rcu *opt;
1293	struct rtable *rt;
1294
1295	rt = *rtp;
1296	if (unlikely(!rt))
1297		return -EFAULT;
1298
1299	cork->fragsize = ip_sk_use_pmtu(sk) ?
1300			 dst_mtu(&rt->dst) : READ_ONCE(rt->dst.dev->mtu);
1301
1302	if (!inetdev_valid_mtu(cork->fragsize))
1303		return -ENETUNREACH;
1304
1305	/*
1306	 * setup for corking.
1307	 */
1308	opt = ipc->opt;
1309	if (opt) {
1310		if (!cork->opt) {
1311			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
1312					    sk->sk_allocation);
1313			if (unlikely(!cork->opt))
1314				return -ENOBUFS;
1315		}
1316		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1317		cork->flags |= IPCORK_OPT;
1318		cork->addr = ipc->addr;
1319	}
1320
1321	cork->gso_size = ipc->gso_size;
1322
1323	cork->dst = &rt->dst;
1324	/* We stole this route, caller should not release it. */
1325	*rtp = NULL;
1326
1327	cork->length = 0;
1328	cork->ttl = ipc->ttl;
1329	cork->tos = ipc->tos;
1330	cork->mark = ipc->sockc.mark;
1331	cork->priority = ipc->priority;
1332	cork->transmit_time = ipc->sockc.transmit_time;
1333	cork->tx_flags = 0;
1334	sock_tx_timestamp(sk, &ipc->sockc, &cork->tx_flags);
1335	if (ipc->sockc.tsflags & SOCKCM_FLAG_TS_OPT_ID) {
1336		cork->flags |= IPCORK_TS_OPT_ID;
1337		cork->ts_opt_id = ipc->sockc.ts_opt_id;
1338	}
1339
1340	return 0;
1341}
1342
1343/*
1344 *	ip_append_data() can make one large IP datagram from many pieces of
1345 *	data.  Each piece will be held on the socket until
1346 *	ip_push_pending_frames() is called. Each piece can be a page or
1347 *	non-page data.
1348 *
1349 *	Not only UDP, other transport protocols - e.g. raw sockets - can use
1350 *	this interface potentially.
1351 *
1352 *	LATER: length must be adjusted by pad at tail, when it is required.
1353 */
1354int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1355		   int getfrag(void *from, char *to, int offset, int len,
1356			       int odd, struct sk_buff *skb),
1357		   void *from, int length, int transhdrlen,
1358		   struct ipcm_cookie *ipc, struct rtable **rtp,
1359		   unsigned int flags)
1360{
1361	struct inet_sock *inet = inet_sk(sk);
1362	int err;
1363
1364	if (flags&MSG_PROBE)
1365		return 0;
1366
1367	if (skb_queue_empty(&sk->sk_write_queue)) {
1368		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1369		if (err)
1370			return err;
1371	} else {
1372		transhdrlen = 0;
1373	}
1374
1375	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
1376				sk_page_frag(sk), getfrag,
1377				from, length, transhdrlen, flags);
1378}
1379
1380static void ip_cork_release(struct inet_cork *cork)
1381{
1382	cork->flags &= ~IPCORK_OPT;
1383	kfree(cork->opt);
1384	cork->opt = NULL;
1385	dst_release(cork->dst);
1386	cork->dst = NULL;
1387}
1388
1389/*
1390 *	Combined all pending IP fragments on the socket as one IP datagram
1391 *	and push them out.
1392 */
1393struct sk_buff *__ip_make_skb(struct sock *sk,
1394			      struct flowi4 *fl4,
1395			      struct sk_buff_head *queue,
1396			      struct inet_cork *cork)
1397{
1398	struct sk_buff *skb, *tmp_skb;
1399	struct sk_buff **tail_skb;
1400	struct inet_sock *inet = inet_sk(sk);
1401	struct net *net = sock_net(sk);
1402	struct ip_options *opt = NULL;
1403	struct rtable *rt = dst_rtable(cork->dst);
1404	struct iphdr *iph;
1405	u8 pmtudisc, ttl;
1406	__be16 df = 0;
1407
1408	skb = __skb_dequeue(queue);
1409	if (!skb)
1410		goto out;
1411	tail_skb = &(skb_shinfo(skb)->frag_list);
1412
1413	/* move skb->data to ip header from ext header */
1414	if (skb->data < skb_network_header(skb))
1415		__skb_pull(skb, skb_network_offset(skb));
1416	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1417		__skb_pull(tmp_skb, skb_network_header_len(skb));
1418		*tail_skb = tmp_skb;
1419		tail_skb = &(tmp_skb->next);
1420		skb->len += tmp_skb->len;
1421		skb->data_len += tmp_skb->len;
1422		skb->truesize += tmp_skb->truesize;
1423		tmp_skb->destructor = NULL;
1424		tmp_skb->sk = NULL;
1425	}
1426
1427	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
1428	 * to fragment the frame generated here. No matter, what transforms
1429	 * how transforms change size of the packet, it will come out.
1430	 */
1431	skb->ignore_df = ip_sk_ignore_df(sk);
1432
1433	/* DF bit is set when we want to see DF on outgoing frames.
1434	 * If ignore_df is set too, we still allow to fragment this frame
1435	 * locally. */
1436	pmtudisc = READ_ONCE(inet->pmtudisc);
1437	if (pmtudisc == IP_PMTUDISC_DO ||
1438	    pmtudisc == IP_PMTUDISC_PROBE ||
1439	    (skb->len <= dst_mtu(&rt->dst) &&
1440	     ip_dont_fragment(sk, &rt->dst)))
1441		df = htons(IP_DF);
1442
1443	if (cork->flags & IPCORK_OPT)
1444		opt = cork->opt;
1445
1446	if (cork->ttl != 0)
1447		ttl = cork->ttl;
1448	else if (rt->rt_type == RTN_MULTICAST)
1449		ttl = READ_ONCE(inet->mc_ttl);
1450	else
1451		ttl = ip_select_ttl(inet, &rt->dst);
1452
1453	iph = ip_hdr(skb);
1454	iph->version = 4;
1455	iph->ihl = 5;
1456	iph->tos = (cork->tos != -1) ? cork->tos : READ_ONCE(inet->tos);
1457	iph->frag_off = df;
1458	iph->ttl = ttl;
1459	iph->protocol = sk->sk_protocol;
1460	ip_copy_addrs(iph, fl4);
1461	ip_select_ident(net, skb, sk);
1462
1463	if (opt) {
1464		iph->ihl += opt->optlen >> 2;
1465		ip_options_build(skb, opt, cork->addr, rt);
1466	}
1467
1468	skb->priority = (cork->tos != -1) ? cork->priority: READ_ONCE(sk->sk_priority);
1469	skb->mark = cork->mark;
1470	if (sk_is_tcp(sk))
1471		skb_set_delivery_time(skb, cork->transmit_time, SKB_CLOCK_MONOTONIC);
1472	else
1473		skb_set_delivery_type_by_clockid(skb, cork->transmit_time, sk->sk_clockid);
1474	/*
1475	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
1476	 * on dst refcount
1477	 */
1478	cork->dst = NULL;
1479	skb_dst_set(skb, &rt->dst);
1480
1481	if (iph->protocol == IPPROTO_ICMP) {
1482		u8 icmp_type;
1483
1484		/* For such sockets, transhdrlen is zero when do ip_append_data(),
1485		 * so icmphdr does not in skb linear region and can not get icmp_type
1486		 * by icmp_hdr(skb)->type.
1487		 */
1488		if (sk->sk_type == SOCK_RAW &&
1489		    !(fl4->flowi4_flags & FLOWI_FLAG_KNOWN_NH))
1490			icmp_type = fl4->fl4_icmp_type;
1491		else
1492			icmp_type = icmp_hdr(skb)->type;
1493		icmp_out_count(net, icmp_type);
1494	}
1495
1496	ip_cork_release(cork);
1497out:
1498	return skb;
1499}
1500
1501int ip_send_skb(struct net *net, struct sk_buff *skb)
1502{
1503	int err;
1504
1505	err = ip_local_out(net, skb->sk, skb);
1506	if (err) {
1507		if (err > 0)
1508			err = net_xmit_errno(err);
1509		if (err)
1510			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
1511	}
1512
1513	return err;
1514}
1515
1516int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1517{
1518	struct sk_buff *skb;
1519
1520	skb = ip_finish_skb(sk, fl4);
1521	if (!skb)
1522		return 0;
1523
1524	/* Netfilter gets whole the not fragmented skb. */
1525	return ip_send_skb(sock_net(sk), skb);
1526}
1527
1528/*
1529 *	Throw away all pending data on the socket.
1530 */
1531static void __ip_flush_pending_frames(struct sock *sk,
1532				      struct sk_buff_head *queue,
1533				      struct inet_cork *cork)
1534{
1535	struct sk_buff *skb;
1536
1537	while ((skb = __skb_dequeue_tail(queue)) != NULL)
1538		kfree_skb(skb);
1539
1540	ip_cork_release(cork);
1541}
1542
1543void ip_flush_pending_frames(struct sock *sk)
1544{
1545	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
1546}
1547
1548struct sk_buff *ip_make_skb(struct sock *sk,
1549			    struct flowi4 *fl4,
1550			    int getfrag(void *from, char *to, int offset,
1551					int len, int odd, struct sk_buff *skb),
1552			    void *from, int length, int transhdrlen,
1553			    struct ipcm_cookie *ipc, struct rtable **rtp,
1554			    struct inet_cork *cork, unsigned int flags)
1555{
1556	struct sk_buff_head queue;
1557	int err;
1558
1559	if (flags & MSG_PROBE)
1560		return NULL;
1561
1562	__skb_queue_head_init(&queue);
1563
1564	cork->flags = 0;
1565	cork->addr = 0;
1566	cork->opt = NULL;
1567	err = ip_setup_cork(sk, cork, ipc, rtp);
1568	if (err)
1569		return ERR_PTR(err);
1570
1571	err = __ip_append_data(sk, fl4, &queue, cork,
1572			       &current->task_frag, getfrag,
1573			       from, length, transhdrlen, flags);
1574	if (err) {
1575		__ip_flush_pending_frames(sk, &queue, cork);
1576		return ERR_PTR(err);
1577	}
1578
1579	return __ip_make_skb(sk, fl4, &queue, cork);
1580}
1581
1582/*
1583 *	Fetch data from kernel space and fill in checksum if needed.
1584 */
1585static int ip_reply_glue_bits(void *dptr, char *to, int offset,
1586			      int len, int odd, struct sk_buff *skb)
1587{
1588	__wsum csum;
1589
1590	csum = csum_partial_copy_nocheck(dptr+offset, to, len);
1591	skb->csum = csum_block_add(skb->csum, csum, odd);
1592	return 0;
1593}
1594
1595/*
1596 *	Generic function to send a packet as reply to another packet.
1597 *	Used to send some TCP resets/acks so far.
1598 */
1599void ip_send_unicast_reply(struct sock *sk, const struct sock *orig_sk,
1600			   struct sk_buff *skb,
1601			   const struct ip_options *sopt,
1602			   __be32 daddr, __be32 saddr,
1603			   const struct ip_reply_arg *arg,
1604			   unsigned int len, u64 transmit_time, u32 txhash)
1605{
1606	struct ip_options_data replyopts;
1607	struct ipcm_cookie ipc;
1608	struct flowi4 fl4;
1609	struct rtable *rt = skb_rtable(skb);
1610	struct net *net = sock_net(sk);
1611	struct sk_buff *nskb;
1612	int err;
1613	int oif;
1614
1615	if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
1616		return;
1617
1618	ipcm_init(&ipc);
1619	ipc.addr = daddr;
1620	ipc.sockc.transmit_time = transmit_time;
1621
1622	if (replyopts.opt.opt.optlen) {
1623		ipc.opt = &replyopts.opt;
1624
1625		if (replyopts.opt.opt.srr)
1626			daddr = replyopts.opt.opt.faddr;
1627	}
1628
1629	oif = arg->bound_dev_if;
1630	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
1631		oif = skb->skb_iif;
1632
1633	flowi4_init_output(&fl4, oif,
1634			   IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1635			   arg->tos & INET_DSCP_MASK,
1636			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1637			   ip_reply_arg_flowi_flags(arg),
1638			   daddr, saddr,
1639			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
1640			   arg->uid);
1641	security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
1642	rt = ip_route_output_flow(net, &fl4, sk);
1643	if (IS_ERR(rt))
1644		return;
1645
1646	inet_sk(sk)->tos = arg->tos & ~INET_ECN_MASK;
1647
1648	sk->sk_protocol = ip_hdr(skb)->protocol;
1649	sk->sk_bound_dev_if = arg->bound_dev_if;
1650	sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default);
1651	ipc.sockc.mark = fl4.flowi4_mark;
1652	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
1653			     len, 0, &ipc, &rt, MSG_DONTWAIT);
1654	if (unlikely(err)) {
1655		ip_flush_pending_frames(sk);
1656		goto out;
1657	}
1658
1659	nskb = skb_peek(&sk->sk_write_queue);
1660	if (nskb) {
1661		if (arg->csumoffset >= 0)
1662			*((__sum16 *)skb_transport_header(nskb) +
1663			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1664								arg->csum));
1665		nskb->ip_summed = CHECKSUM_NONE;
1666		if (orig_sk)
1667			skb_set_owner_edemux(nskb, (struct sock *)orig_sk);
1668		if (transmit_time)
1669			nskb->tstamp_type = SKB_CLOCK_MONOTONIC;
1670		if (txhash)
1671			skb_set_hash(nskb, txhash, PKT_HASH_TYPE_L4);
1672		ip_push_pending_frames(sk, &fl4);
1673	}
1674out:
1675	ip_rt_put(rt);
1676}
1677
1678void __init ip_init(void)
1679{
1680	ip_rt_init();
1681	inet_initpeers();
1682
1683#if defined(CONFIG_IP_MULTICAST)
1684	igmp_mc_init();
1685#endif
1686}