ip_output.c 39.5 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7
/*
 * INET		An implementation of the TCP/IP protocol suite for the LINUX
 *		operating system.  INET is implemented using the  BSD Socket
 *		interface as the means of communication with the user level.
 *
 *		The Internet Protocol (IP) output module.
 *
8
 * Authors:	Ross Biro
L
Linus Torvalds 已提交
9 10 11 12 13 14 15 16 17 18 19 20 21 22
 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 *		Donald Becker, <becker@super.org>
 *		Alan Cox, <Alan.Cox@linux.org>
 *		Richard Underwood
 *		Stefan Becker, <stefanb@yello.ping.de>
 *		Jorge Cwik, <jorge@laser.satlink.net>
 *		Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 *		Hirokazu Takahashi, <taka@valinux.co.jp>
 *
 *	See ip_input.c for original log
 *
 *	Fixes:
 *		Alan Cox	:	Missing nonblock feature in ip_build_xmit.
 *		Mike Kilburn	:	htons() missing in ip_build_xmit.
23
 *		Bradford Johnson:	Fix faulty handling of some frames when
L
Linus Torvalds 已提交
24 25 26 27 28 29 30 31 32 33
 *					no route is found.
 *		Alexander Demenshin:	Missing sk/skb free in ip_queue_xmit
 *					(in case if packet not accepted by
 *					output firewall rules)
 *		Mike McLagan	:	Routing by source
 *		Alexey Kuznetsov:	use new route cache
 *		Andi Kleen:		Fix broken PMTU recovery and remove
 *					some redundant tests.
 *	Vitaly E. Lavrov	:	Transparent proxy revived after year coma.
 *		Andi Kleen	: 	Replace ip_reply with ip_send_reply.
34 35 36
 *		Andi Kleen	:	Split fast and slow ip_build_xmit path
 *					for decreased register pressure on x86
 *					and more readibility.
L
Linus Torvalds 已提交
37 38 39 40 41 42 43 44
 *		Marc Boucher	:	When call_out_firewall returns FW_QUEUE,
 *					silently drop skb instead of failing with -EPERM.
 *		Detlev Wengorz	:	Copy protocol for fragments.
 *		Hirokazu Takahashi:	HW checksumming for outgoing UDP
 *					datagrams.
 *		Hirokazu Takahashi:	sendfile() on UDP works now.
 */

45
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
46 47 48 49 50 51
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/string.h>
#include <linux/errno.h>
A
Al Viro 已提交
52
#include <linux/highmem.h>
53
#include <linux/slab.h>
L
Linus Torvalds 已提交
54 55 56 57 58 59 60 61 62 63 64 65 66 67 68

#include <linux/socket.h>
#include <linux/sockios.h>
#include <linux/in.h>
#include <linux/inet.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/proc_fs.h>
#include <linux/stat.h>
#include <linux/init.h>

#include <net/snmp.h>
#include <net/ip.h>
#include <net/protocol.h>
#include <net/route.h>
69
#include <net/xfrm.h>
L
Linus Torvalds 已提交
70 71 72 73 74 75
#include <linux/skbuff.h>
#include <net/sock.h>
#include <net/arp.h>
#include <net/icmp.h>
#include <net/checksum.h>
#include <net/inetpeer.h>
76
#include <net/lwtunnel.h>
77
#include <linux/bpf-cgroup.h>
L
Linus Torvalds 已提交
78 79 80 81
#include <linux/igmp.h>
#include <linux/netfilter_ipv4.h>
#include <linux/netfilter_bridge.h>
#include <linux/netlink.h>
82
#include <linux/tcp.h>
L
Linus Torvalds 已提交
83

84 85 86 87
static int
ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
	    unsigned int mtu,
	    int (*output)(struct net *, struct sock *, struct sk_buff *));
88

L
Linus Torvalds 已提交
89
/* Generate a checksum for an outgoing IP datagram. */
90
void ip_send_check(struct iphdr *iph)
L
Linus Torvalds 已提交
91 92 93 94
{
	iph->check = 0;
	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
}
E
Eric Dumazet 已提交
95
EXPORT_SYMBOL(ip_send_check);
L
Linus Torvalds 已提交
96

97
int __ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
98 99 100 101 102
{
	struct iphdr *iph = ip_hdr(skb);

	iph->tot_len = htons(skb->len);
	ip_send_check(iph);
103 104 105 106 107 108 109 110

	/* if egress device is enslaved to an L3 master device pass the
	 * skb to its handler for processing
	 */
	skb = l3mdev_ip_out(sk, skb);
	if (unlikely(!skb))
		return 0;

111 112
	skb->protocol = htons(ETH_P_IP);

113 114
	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
		       net, sk, skb, NULL, skb_dst(skb)->dev,
115
		       dst_output);
116 117
}

118
int ip_local_out(struct net *net, struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
119 120 121
{
	int err;

122
	err = __ip_local_out(net, sk, skb);
H
Herbert Xu 已提交
123
	if (likely(err == 1))
124
		err = dst_output(net, sk, skb);
H
Herbert Xu 已提交
125 126 127

	return err;
}
128
EXPORT_SYMBOL_GPL(ip_local_out);
H
Herbert Xu 已提交
129

L
Linus Torvalds 已提交
130 131 132 133 134
static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
{
	int ttl = inet->uc_ttl;

	if (ttl < 0)
135
		ttl = ip4_dst_hoplimit(dst);
L
Linus Torvalds 已提交
136 137 138
	return ttl;
}

139
/*
L
Linus Torvalds 已提交
140 141 142
 *		Add an ip header to a skbuff and send it out.
 *
 */
143
int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
144
			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
L
Linus Torvalds 已提交
145 146
{
	struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
147
	struct rtable *rt = skb_rtable(skb);
148
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
149 150 151
	struct iphdr *iph;

	/* Build the IP header. */
152
	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
153
	skb_reset_network_header(skb);
154
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
155 156 157
	iph->version  = 4;
	iph->ihl      = 5;
	iph->tos      = inet->tos;
158
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
159 160
	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
	iph->saddr    = saddr;
L
Linus Torvalds 已提交
161
	iph->protocol = sk->sk_protocol;
162 163 164 165 166
	if (ip_dont_fragment(sk, &rt->dst)) {
		iph->frag_off = htons(IP_DF);
		iph->id = 0;
	} else {
		iph->frag_off = 0;
167
		__ip_select_ident(net, iph, 1);
168
	}
L
Linus Torvalds 已提交
169

170 171 172
	if (opt && opt->opt.optlen) {
		iph->ihl += opt->opt.optlen>>2;
		ip_options_build(skb, &opt->opt, daddr, rt, 0);
L
Linus Torvalds 已提交
173 174 175
	}

	skb->priority = sk->sk_priority;
176 177
	if (!skb->mark)
		skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
178 179

	/* Send it out. */
180
	return ip_local_out(net, skb->sk, skb);
L
Linus Torvalds 已提交
181
}
182 183
EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);

184
static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
185
{
E
Eric Dumazet 已提交
186
	struct dst_entry *dst = skb_dst(skb);
187
	struct rtable *rt = (struct rtable *)dst;
L
Linus Torvalds 已提交
188
	struct net_device *dev = dst->dev;
189
	unsigned int hh_len = LL_RESERVED_SPACE(dev);
190
	struct neighbour *neigh;
191
	u32 nexthop;
L
Linus Torvalds 已提交
192

193
	if (rt->rt_type == RTN_MULTICAST) {
194
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
195
	} else if (rt->rt_type == RTN_BROADCAST)
196
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
197

L
Linus Torvalds 已提交
198
	/* Be paranoid, rather than too clever. */
199
	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
L
Linus Torvalds 已提交
200 201 202
		struct sk_buff *skb2;

		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
203
		if (!skb2) {
L
Linus Torvalds 已提交
204 205 206 207 208
			kfree_skb(skb);
			return -ENOMEM;
		}
		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);
209
		consume_skb(skb);
L
Linus Torvalds 已提交
210 211 212
		skb = skb2;
	}

213 214 215 216 217 218 219
	if (lwtunnel_xmit_redirect(dst->lwtstate)) {
		int res = lwtunnel_xmit(skb);

		if (res < 0 || res == LWTUNNEL_XMIT_DONE)
			return res;
	}

220
	rcu_read_lock_bh();
J
Julian Anastasov 已提交
221
	nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
222 223 224
	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
	if (unlikely(!neigh))
		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
225
	if (!IS_ERR(neigh)) {
226 227 228
		int res;

		sock_confirm_neigh(skb, neigh);
229
		res = neigh_output(neigh, skb);
230

231
		rcu_read_unlock_bh();
232 233
		return res;
	}
234
	rcu_read_unlock_bh();
235

236 237
	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
			    __func__);
L
Linus Torvalds 已提交
238 239 240 241
	kfree_skb(skb);
	return -EINVAL;
}

242 243
static int ip_finish_output_gso(struct net *net, struct sock *sk,
				struct sk_buff *skb, unsigned int mtu)
244 245 246 247 248
{
	netdev_features_t features;
	struct sk_buff *segs;
	int ret = 0;

249
	/* common case: seglen is <= mtu
250
	 */
251
	if (skb_gso_validate_network_len(skb, mtu))
252
		return ip_finish_output2(net, sk, skb);
253

254
	/* Slowpath -  GSO segment length exceeds the egress MTU.
255
	 *
256 257 258 259 260 261 262 263 264 265
	 * This can happen in several cases:
	 *  - Forwarding of a TCP GRO skb, when DF flag is not set.
	 *  - Forwarding of an skb that arrived on a virtualization interface
	 *    (virtio-net/vhost/tap) with TSO/GSO size set by other network
	 *    stack.
	 *  - Local GSO skb transmitted on an NETIF_F_TSO tunnel stacked over an
	 *    interface with a smaller MTU.
	 *  - Arriving GRO skb (or GSO skb in a virtualized environment) that is
	 *    bridged to a NETIF_F_TSO tunnel stacked over an interface with an
	 *    insufficent MTU.
266 267
	 */
	features = netif_skb_features(skb);
268
	BUILD_BUG_ON(sizeof(*IPCB(skb)) > SKB_SGO_CB_OFFSET);
269
	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
270
	if (IS_ERR_OR_NULL(segs)) {
271 272 273 274 275 276 277 278 279 280 281
		kfree_skb(skb);
		return -ENOMEM;
	}

	consume_skb(skb);

	do {
		struct sk_buff *nskb = segs->next;
		int err;

		segs->next = NULL;
282
		err = ip_fragment(net, sk, segs, mtu, ip_finish_output2);
283 284 285 286 287 288 289 290 291

		if (err && ret == 0)
			ret = err;
		segs = nskb;
	} while (segs);

	return ret;
}

292
static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
293
{
294
	unsigned int mtu;
295 296 297 298 299 300 301
	int ret;

	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
	if (ret) {
		kfree_skb(skb);
		return ret;
	}
302

303 304
#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
	/* Policy lookup after SNAT yielded a new policy */
305
	if (skb_dst(skb)->xfrm) {
306
		IPCB(skb)->flags |= IPSKB_REROUTED;
307
		return dst_output(net, sk, skb);
308
	}
309
#endif
310
	mtu = ip_skb_dst_mtu(sk, skb);
311
	if (skb_is_gso(skb))
312
		return ip_finish_output_gso(net, sk, skb, mtu);
313

314
	if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
315
		return ip_fragment(net, sk, skb, mtu, ip_finish_output2);
316

317
	return ip_finish_output2(net, sk, skb);
L
Linus Torvalds 已提交
318 319
}

320 321 322 323 324 325 326 327 328 329 330 331 332 333
static int ip_mc_finish_output(struct net *net, struct sock *sk,
			       struct sk_buff *skb)
{
	int ret;

	ret = BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb);
	if (ret) {
		kfree_skb(skb);
		return ret;
	}

	return dev_loopback_xmit(net, sk, skb);
}

E
Eric W. Biederman 已提交
334
int ip_mc_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
335
{
E
Eric Dumazet 已提交
336
	struct rtable *rt = skb_rtable(skb);
337
	struct net_device *dev = rt->dst.dev;
L
Linus Torvalds 已提交
338 339 340 341

	/*
	 *	If the indicated interface is up and running, send the packet.
	 */
342
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
343 344 345 346 347 348 349 350 351

	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

	/*
	 *	Multicasts are looped back for other local users
	 */

	if (rt->rt_flags&RTCF_MULTICAST) {
352
		if (sk_mc_loop(sk)
L
Linus Torvalds 已提交
353 354 355 356 357 358 359 360 361
#ifdef CONFIG_IP_MROUTE
		/* Small optimization: do not loopback not local frames,
		   which returned after forwarding; they will be  dropped
		   by ip_mr_input in any case.
		   Note, that local frames are looped back to be delivered
		   to local recipients.

		   This check is duplicated in ip_mr_input at the moment.
		 */
362 363 364
		    &&
		    ((rt->rt_flags & RTCF_LOCAL) ||
		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
L
Linus Torvalds 已提交
365
#endif
366
		   ) {
L
Linus Torvalds 已提交
367 368
			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
			if (newskb)
369
				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
370
					net, sk, newskb, NULL, newskb->dev,
371
					ip_mc_finish_output);
L
Linus Torvalds 已提交
372 373 374 375
		}

		/* Multicasts with ttl 0 must not go beyond the host */

376
		if (ip_hdr(skb)->ttl == 0) {
L
Linus Torvalds 已提交
377 378 379 380 381 382 383 384
			kfree_skb(skb);
			return 0;
		}
	}

	if (rt->rt_flags&RTCF_BROADCAST) {
		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
		if (newskb)
385 386
			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
				net, sk, newskb, NULL, newskb->dev,
387
				ip_mc_finish_output);
L
Linus Torvalds 已提交
388 389
	}

390 391 392
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, skb->dev,
			    ip_finish_output,
393
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
394 395
}

E
Eric W. Biederman 已提交
396
int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
397
{
E
Eric Dumazet 已提交
398
	struct net_device *dev = skb_dst(skb)->dev;
399

400
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
401

402 403 404
	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

405 406
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, dev,
407
			    ip_finish_output,
408
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
409 410
}

411 412 413 414 415 416 417 418 419 420 421 422 423 424
/*
 * copy saddr and daddr, possibly using 64bit load/stores
 * Equivalent to :
 *   iph->saddr = fl4->saddr;
 *   iph->daddr = fl4->daddr;
 */
static void ip_copy_addrs(struct iphdr *iph, const struct flowi4 *fl4)
{
	BUILD_BUG_ON(offsetof(typeof(*fl4), daddr) !=
		     offsetof(typeof(*fl4), saddr) + sizeof(fl4->saddr));
	memcpy(&iph->saddr, &fl4->saddr,
	       sizeof(fl4->saddr) + sizeof(fl4->daddr));
}

425
/* Note: skb->sk can be different from sk, in case of tunnels */
426 427
int __ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl,
		    __u8 tos)
L
Linus Torvalds 已提交
428 429
{
	struct inet_sock *inet = inet_sk(sk);
430
	struct net *net = sock_net(sk);
431
	struct ip_options_rcu *inet_opt;
432
	struct flowi4 *fl4;
L
Linus Torvalds 已提交
433 434
	struct rtable *rt;
	struct iphdr *iph;
435
	int res;
L
Linus Torvalds 已提交
436 437 438 439

	/* Skip all of this if the packet is already routed,
	 * f.e. by something like SCTP.
	 */
440
	rcu_read_lock();
441
	inet_opt = rcu_dereference(inet->inet_opt);
442
	fl4 = &fl->u.ip4;
E
Eric Dumazet 已提交
443
	rt = skb_rtable(skb);
444
	if (rt)
L
Linus Torvalds 已提交
445 446 447 448
		goto packet_routed;

	/* Make sure we can route this packet. */
	rt = (struct rtable *)__sk_dst_check(sk, 0);
449
	if (!rt) {
A
Al Viro 已提交
450
		__be32 daddr;
L
Linus Torvalds 已提交
451 452

		/* Use correct destination address if we have options. */
E
Eric Dumazet 已提交
453
		daddr = inet->inet_daddr;
454 455
		if (inet_opt && inet_opt->opt.srr)
			daddr = inet_opt->opt.faddr;
L
Linus Torvalds 已提交
456

457 458 459 460
		/* If this fails, retransmit mechanism of transport layer will
		 * keep trying until route appears or the connection times
		 * itself out.
		 */
461
		rt = ip_route_output_ports(net, fl4, sk,
462 463 464 465
					   daddr, inet->inet_saddr,
					   inet->inet_dport,
					   inet->inet_sport,
					   sk->sk_protocol,
466
					   RT_CONN_FLAGS_TOS(sk, tos),
467 468 469
					   sk->sk_bound_dev_if);
		if (IS_ERR(rt))
			goto no_route;
470
		sk_setup_caps(sk, &rt->dst);
L
Linus Torvalds 已提交
471
	}
472
	skb_dst_set_noref(skb, &rt->dst);
L
Linus Torvalds 已提交
473 474

packet_routed:
J
Julian Anastasov 已提交
475
	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
L
Linus Torvalds 已提交
476 477 478
		goto no_route;

	/* OK, we know where to send it, allocate and build IP header. */
479
	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
480
	skb_reset_network_header(skb);
481
	iph = ip_hdr(skb);
482
	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (tos & 0xff));
W
WANG Cong 已提交
483
	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
L
Linus Torvalds 已提交
484 485 486
		iph->frag_off = htons(IP_DF);
	else
		iph->frag_off = 0;
487
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
488
	iph->protocol = sk->sk_protocol;
489 490
	ip_copy_addrs(iph, fl4);

L
Linus Torvalds 已提交
491 492
	/* Transport layer set skb->h.foo itself. */

493 494 495
	if (inet_opt && inet_opt->opt.optlen) {
		iph->ihl += inet_opt->opt.optlen >> 2;
		ip_options_build(skb, &inet_opt->opt, inet->inet_daddr, rt, 0);
L
Linus Torvalds 已提交
496 497
	}

498
	ip_select_ident_segs(net, skb, sk,
499
			     skb_shinfo(skb)->gso_segs ?: 1);
L
Linus Torvalds 已提交
500

501
	/* TODO : should we use skb->sk here instead of sk ? */
L
Linus Torvalds 已提交
502
	skb->priority = sk->sk_priority;
503
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
504

505
	res = ip_local_out(net, sk, skb);
506 507
	rcu_read_unlock();
	return res;
L
Linus Torvalds 已提交
508 509

no_route:
510
	rcu_read_unlock();
511
	IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
L
Linus Torvalds 已提交
512 513 514
	kfree_skb(skb);
	return -EHOSTUNREACH;
}
515
EXPORT_SYMBOL(__ip_queue_xmit);
L
Linus Torvalds 已提交
516 517 518 519 520 521

static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
{
	to->pkt_type = from->pkt_type;
	to->priority = from->priority;
	to->protocol = from->protocol;
E
Eric Dumazet 已提交
522
	skb_dst_drop(to);
523
	skb_dst_copy(to, from);
L
Linus Torvalds 已提交
524
	to->dev = from->dev;
T
Thomas Graf 已提交
525
	to->mark = from->mark;
L
Linus Torvalds 已提交
526 527 528 529 530 531 532

	/* Copy the flags to each fragment. */
	IPCB(to)->flags = IPCB(from)->flags;

#ifdef CONFIG_NET_SCHED
	to->tc_index = from->tc_index;
#endif
533
	nf_copy(to, from);
534
#if IS_ENABLED(CONFIG_IP_VS)
535
	to->ipvs_property = from->ipvs_property;
L
Linus Torvalds 已提交
536
#endif
537
	skb_copy_secmark(to, from);
L
Linus Torvalds 已提交
538 539
}

540
static int ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
541
		       unsigned int mtu,
542
		       int (*output)(struct net *, struct sock *, struct sk_buff *))
543 544 545
{
	struct iphdr *iph = ip_hdr(skb);

546
	if ((iph->frag_off & htons(IP_DF)) == 0)
547
		return ip_do_fragment(net, sk, skb, output);
548 549

	if (unlikely(!skb->ignore_df ||
550 551
		     (IPCB(skb)->frag_max_size &&
		      IPCB(skb)->frag_max_size > mtu))) {
552
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
553 554 555 556 557 558
		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
			  htonl(mtu));
		kfree_skb(skb);
		return -EMSGSIZE;
	}

559
	return ip_do_fragment(net, sk, skb, output);
560 561
}

L
Linus Torvalds 已提交
562 563 564 565 566 567 568
/*
 *	This IP datagram is too large to be sent in one piece.  Break it up into
 *	smaller pieces (each of size equal to IP header plus
 *	a block of the data of the original IP data part) that will yet fit in a
 *	single device frame, and queue such a frame for sending.
 */

569 570
int ip_do_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
		   int (*output)(struct net *, struct sock *, struct sk_buff *))
L
Linus Torvalds 已提交
571 572 573 574
{
	struct iphdr *iph;
	int ptr;
	struct sk_buff *skb2;
575
	unsigned int mtu, hlen, left, len, ll_rs;
L
Linus Torvalds 已提交
576
	int offset;
577
	__be16 not_last_frag;
E
Eric Dumazet 已提交
578
	struct rtable *rt = skb_rtable(skb);
L
Linus Torvalds 已提交
579 580
	int err = 0;

581 582 583 584 585
	/* for offloaded checksums cleanup checksum before fragmentation */
	if (skb->ip_summed == CHECKSUM_PARTIAL &&
	    (err = skb_checksum_help(skb)))
		goto fail;

L
Linus Torvalds 已提交
586 587 588 589
	/*
	 *	Point into the IP datagram header.
	 */

590
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
591

592
	mtu = ip_skb_dst_mtu(sk, skb);
593 594
	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
		mtu = IPCB(skb)->frag_max_size;
L
Linus Torvalds 已提交
595 596 597 598 599 600

	/*
	 *	Setup starting values.
	 */

	hlen = iph->ihl * 4;
601
	mtu = mtu - hlen;	/* Size of data space */
H
Herbert Xu 已提交
602
	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
603
	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
L
Linus Torvalds 已提交
604 605 606 607 608 609 610 611

	/* When frag_list is given, use it. First, check its validity:
	 * some transformers could create wrong frag_list or break existing
	 * one, it is not prohibited. In this case fall back to copying.
	 *
	 * LATER: this step can be merged to real generation of fragments,
	 * we can switch to copy when see the first bad fragment.
	 */
612
	if (skb_has_frag_list(skb)) {
613
		struct sk_buff *frag, *frag2;
614
		unsigned int first_len = skb_pagelen(skb);
L
Linus Torvalds 已提交
615 616 617

		if (first_len - hlen > mtu ||
		    ((first_len - hlen) & 7) ||
618
		    ip_is_fragment(iph) ||
619 620
		    skb_cloned(skb) ||
		    skb_headroom(skb) < ll_rs)
L
Linus Torvalds 已提交
621 622
			goto slow_path;

623
		skb_walk_frags(skb, frag) {
L
Linus Torvalds 已提交
624 625 626
			/* Correct geometry. */
			if (frag->len > mtu ||
			    ((frag->len & 7) && frag->next) ||
627
			    skb_headroom(frag) < hlen + ll_rs)
628
				goto slow_path_clean;
L
Linus Torvalds 已提交
629 630 631

			/* Partially cloned skb? */
			if (skb_shared(frag))
632
				goto slow_path_clean;
633 634 635 636 637 638

			BUG_ON(frag->sk);
			if (skb->sk) {
				frag->sk = skb->sk;
				frag->destructor = sock_wfree;
			}
639
			skb->truesize -= frag->truesize;
L
Linus Torvalds 已提交
640 641 642 643 644 645 646
		}

		/* Everything is OK. Generate! */

		err = 0;
		offset = 0;
		frag = skb_shinfo(skb)->frag_list;
647
		skb_frag_list_init(skb);
L
Linus Torvalds 已提交
648 649 650 651 652 653 654 655 656 657 658
		skb->data_len = first_len - skb_headlen(skb);
		skb->len = first_len;
		iph->tot_len = htons(first_len);
		iph->frag_off = htons(IP_MF);
		ip_send_check(iph);

		for (;;) {
			/* Prepare header of the next frame,
			 * before previous one went down. */
			if (frag) {
				frag->ip_summed = CHECKSUM_NONE;
659
				skb_reset_transport_header(frag);
660 661
				__skb_push(frag, hlen);
				skb_reset_network_header(frag);
662
				memcpy(skb_network_header(frag), iph, hlen);
663
				iph = ip_hdr(frag);
L
Linus Torvalds 已提交
664 665 666 667 668 669
				iph->tot_len = htons(frag->len);
				ip_copy_metadata(frag, skb);
				if (offset == 0)
					ip_options_fragment(frag);
				offset += skb->len - hlen;
				iph->frag_off = htons(offset>>3);
670
				if (frag->next)
L
Linus Torvalds 已提交
671 672 673 674 675
					iph->frag_off |= htons(IP_MF);
				/* Ready, complete checksum */
				ip_send_check(iph);
			}

676
			err = output(net, sk, skb);
L
Linus Torvalds 已提交
677

678
			if (!err)
679
				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
680 681 682 683 684 685 686 687 688
			if (err || !frag)
				break;

			skb = frag;
			frag = skb->next;
			skb->next = NULL;
		}

		if (err == 0) {
689
			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
690 691 692 693 694 695 696 697
			return 0;
		}

		while (frag) {
			skb = frag->next;
			kfree_skb(frag);
			frag = skb;
		}
698
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
699
		return err;
700 701 702 703 704 705 706 707 708

slow_path_clean:
		skb_walk_frags(skb, frag2) {
			if (frag2 == frag)
				break;
			frag2->sk = NULL;
			frag2->destructor = NULL;
			skb->truesize += frag2->truesize;
		}
L
Linus Torvalds 已提交
709 710 711
	}

slow_path:
712
	iph = ip_hdr(skb);
713

L
Linus Torvalds 已提交
714
	left = skb->len - hlen;		/* Space per frame */
715
	ptr = hlen;		/* Where to start from */
L
Linus Torvalds 已提交
716 717 718 719 720 721 722 723 724 725 726 727

	/*
	 *	Fragment the datagram.
	 */

	offset = (ntohs(iph->frag_off) & IP_OFFSET) << 3;
	not_last_frag = iph->frag_off & htons(IP_MF);

	/*
	 *	Keep copying data until we run out.
	 */

S
Stephen Hemminger 已提交
728
	while (left > 0) {
L
Linus Torvalds 已提交
729 730 731 732
		len = left;
		/* IF: it doesn't fit, use 'mtu' - the data space left */
		if (len > mtu)
			len = mtu;
L
Lucas De Marchi 已提交
733
		/* IF: we are not sending up to and including the packet end
L
Linus Torvalds 已提交
734 735 736 737 738
		   then align the next start on an eight byte boundary */
		if (len < left)	{
			len &= ~7;
		}

739 740 741
		/* Allocate buffer */
		skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
		if (!skb2) {
L
Linus Torvalds 已提交
742 743 744 745 746 747 748 749 750 751 752
			err = -ENOMEM;
			goto fail;
		}

		/*
		 *	Set up data on packet
		 */

		ip_copy_metadata(skb2, skb);
		skb_reserve(skb2, ll_rs);
		skb_put(skb2, len + hlen);
753
		skb_reset_network_header(skb2);
754
		skb2->transport_header = skb2->network_header + hlen;
L
Linus Torvalds 已提交
755 756 757 758 759 760 761 762 763 764 765 766 767

		/*
		 *	Charge the memory for the fragment to any owner
		 *	it might possess
		 */

		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);

		/*
		 *	Copy the packet header into the new buffer.
		 */

768
		skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
L
Linus Torvalds 已提交
769 770 771 772

		/*
		 *	Copy a block of the IP datagram.
		 */
773
		if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
L
Linus Torvalds 已提交
774 775 776 777 778 779
			BUG();
		left -= len;

		/*
		 *	Fill in the new header fields.
		 */
780
		iph = ip_hdr(skb2);
L
Linus Torvalds 已提交
781 782
		iph->frag_off = htons((offset >> 3));

783 784 785
		if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
			iph->frag_off |= htons(IP_DF);

L
Linus Torvalds 已提交
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810
		/* ANK: dirty, but effective trick. Upgrade options only if
		 * the segment to be fragmented was THE FIRST (otherwise,
		 * options are already fixed) and make it ONCE
		 * on the initial skb, so that all the following fragments
		 * will inherit fixed options.
		 */
		if (offset == 0)
			ip_options_fragment(skb);

		/*
		 *	Added AC : If we are fragmenting a fragment that's not the
		 *		   last fragment then keep MF on each bit
		 */
		if (left > 0 || not_last_frag)
			iph->frag_off |= htons(IP_MF);
		ptr += len;
		offset += len;

		/*
		 *	Put this fragment into the sending queue.
		 */
		iph->tot_len = htons(len + hlen);

		ip_send_check(iph);

811
		err = output(net, sk, skb2);
L
Linus Torvalds 已提交
812 813
		if (err)
			goto fail;
814

815
		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
816
	}
817
	consume_skb(skb);
818
	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
819 820 821
	return err;

fail:
822
	kfree_skb(skb);
823
	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
824 825
	return err;
}
826
EXPORT_SYMBOL(ip_do_fragment);
827

L
Linus Torvalds 已提交
828 829 830
int
ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
{
831
	struct msghdr *msg = from;
L
Linus Torvalds 已提交
832

833
	if (skb->ip_summed == CHECKSUM_PARTIAL) {
834
		if (!copy_from_iter_full(to, len, &msg->msg_iter))
L
Linus Torvalds 已提交
835 836
			return -EFAULT;
	} else {
837
		__wsum csum = 0;
838
		if (!csum_and_copy_from_iter_full(to, len, &csum, &msg->msg_iter))
L
Linus Torvalds 已提交
839 840 841 842 843
			return -EFAULT;
		skb->csum = csum_block_add(skb->csum, csum, odd);
	}
	return 0;
}
E
Eric Dumazet 已提交
844
EXPORT_SYMBOL(ip_generic_getfrag);
L
Linus Torvalds 已提交
845

846
static inline __wsum
L
Linus Torvalds 已提交
847 848 849
csum_page(struct page *page, int offset, int copy)
{
	char *kaddr;
850
	__wsum csum;
L
Linus Torvalds 已提交
851 852 853 854 855 856
	kaddr = kmap(page);
	csum = csum_partial(kaddr + offset, copy, 0);
	kunmap(page);
	return csum;
}

857 858 859
static int __ip_append_data(struct sock *sk,
			    struct flowi4 *fl4,
			    struct sk_buff_head *queue,
860
			    struct inet_cork *cork,
861
			    struct page_frag *pfrag,
862 863 864 865
			    int getfrag(void *from, char *to, int offset,
					int len, int odd, struct sk_buff *skb),
			    void *from, int length, int transhdrlen,
			    unsigned int flags)
L
Linus Torvalds 已提交
866 867 868 869
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;

870
	struct ip_options *opt = cork->opt;
L
Linus Torvalds 已提交
871 872 873 874 875 876
	int hh_len;
	int exthdrlen;
	int mtu;
	int copy;
	int err;
	int offset = 0;
877
	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
L
Linus Torvalds 已提交
878
	int csummode = CHECKSUM_NONE;
879
	struct rtable *rt = (struct rtable *)cork->dst;
880
	unsigned int wmem_alloc_delta = 0;
881
	u32 tskey = 0;
882
	bool paged;
L
Linus Torvalds 已提交
883

884 885 886
	skb = skb_peek_tail(queue);

	exthdrlen = !skb ? rt->dst.header_len : 0;
887
	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
888
	paged = !!cork->gso_size;
889

890 891 892
	if (cork->tx_flags & SKBTX_ANY_SW_TSTAMP &&
	    sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)
		tskey = sk->sk_tskey++;
L
Linus Torvalds 已提交
893

894
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
L
Linus Torvalds 已提交
895 896 897

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
898
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
899

900
	if (cork->length + length > maxnonfragsize - fragheaderlen) {
901
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
902
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
903 904 905 906 907 908 909 910 911
		return -EMSGSIZE;
	}

	/*
	 * transhdrlen > 0 means that this is the first fragment and we wish
	 * it won't be fragmented in the future.
	 */
	if (transhdrlen &&
	    length + fragheaderlen <= mtu &&
912
	    rt->dst.dev->features & (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM) &&
913
	    (!(flags & MSG_MORE) || cork->gso_size) &&
914
	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
915
		csummode = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
916

917
	cork->length += length;
L
Linus Torvalds 已提交
918 919 920 921 922 923 924 925

	/* So, what's going on in the loop below?
	 *
	 * We use calculated fragment length to generate chained skb,
	 * each of segments is IP fragment ready for sending to network after
	 * adding appropriate IP header.
	 */

926
	if (!skb)
L
Linus Torvalds 已提交
927 928 929 930 931 932 933 934 935 936 937 938 939
		goto alloc_new_skb;

	while (length > 0) {
		/* Check if the remaining data fits into current packet. */
		copy = mtu - skb->len;
		if (copy < length)
			copy = maxfraglen - skb->len;
		if (copy <= 0) {
			char *data;
			unsigned int datalen;
			unsigned int fraglen;
			unsigned int fraggap;
			unsigned int alloclen;
940
			unsigned int pagedlen = 0;
L
Linus Torvalds 已提交
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
			struct sk_buff *skb_prev;
alloc_new_skb:
			skb_prev = skb;
			if (skb_prev)
				fraggap = skb_prev->len - maxfraglen;
			else
				fraggap = 0;

			/*
			 * If remaining data exceeds the mtu,
			 * we know we need more fragment(s).
			 */
			datalen = length + fraggap;
			if (datalen > mtu - fragheaderlen)
				datalen = maxfraglen - fragheaderlen;
			fraglen = datalen + fragheaderlen;

958
			if ((flags & MSG_MORE) &&
959
			    !(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
960
				alloclen = mtu;
961
			else if (!paged)
962
				alloclen = fraglen;
963 964 965 966
			else {
				alloclen = min_t(int, fraglen, MAX_HEADER);
				pagedlen = fraglen - alloclen;
			}
L
Linus Torvalds 已提交
967

968 969
			alloclen += exthdrlen;

L
Linus Torvalds 已提交
970 971 972 973 974
			/* The last fragment gets additional space at tail.
			 * Note, with MSG_MORE we overallocate on fragments,
			 * because we have no idea what fragment will be
			 * the last.
			 */
975
			if (datalen == length + fraggap)
976
				alloclen += rt->dst.trailer_len;
977

L
Linus Torvalds 已提交
978
			if (transhdrlen) {
979
				skb = sock_alloc_send_skb(sk,
L
Linus Torvalds 已提交
980 981 982 983
						alloclen + hh_len + 15,
						(flags & MSG_DONTWAIT), &err);
			} else {
				skb = NULL;
984
				if (refcount_read(&sk->sk_wmem_alloc) + wmem_alloc_delta <=
L
Linus Torvalds 已提交
985
				    2 * sk->sk_sndbuf)
986 987
					skb = alloc_skb(alloclen + hh_len + 15,
							sk->sk_allocation);
988
				if (unlikely(!skb))
L
Linus Torvalds 已提交
989 990
					err = -ENOBUFS;
			}
991
			if (!skb)
L
Linus Torvalds 已提交
992 993 994 995 996 997 998 999
				goto error;

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = csummode;
			skb->csum = 0;
			skb_reserve(skb, hh_len);
1000 1001

			/* only the initial fragment is time stamped */
1002
			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1003
			cork->tx_flags = 0;
1004 1005
			skb_shinfo(skb)->tskey = tskey;
			tskey = 0;
L
Linus Torvalds 已提交
1006 1007 1008 1009

			/*
			 *	Find where to start putting bytes.
			 */
1010
			data = skb_put(skb, fraglen + exthdrlen - pagedlen);
1011
			skb_set_network_header(skb, exthdrlen);
1012 1013
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
1014
			data += fragheaderlen + exthdrlen;
L
Linus Torvalds 已提交
1015 1016 1017 1018 1019 1020 1021 1022

			if (fraggap) {
				skb->csum = skb_copy_and_csum_bits(
					skb_prev, maxfraglen,
					data + transhdrlen, fraggap, 0);
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
				data += fraggap;
1023
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1024 1025
			}

1026
			copy = datalen - transhdrlen - fraggap - pagedlen;
L
Linus Torvalds 已提交
1027 1028 1029 1030 1031 1032 1033
			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
				err = -EFAULT;
				kfree_skb(skb);
				goto error;
			}

			offset += copy;
1034
			length -= copy + transhdrlen;
L
Linus Torvalds 已提交
1035 1036 1037 1038
			transhdrlen = 0;
			exthdrlen = 0;
			csummode = CHECKSUM_NONE;

1039 1040 1041
			if ((flags & MSG_CONFIRM) && !skb_prev)
				skb_set_dst_pending_confirm(skb, 1);

L
Linus Torvalds 已提交
1042 1043 1044
			/*
			 * Put the packet on the pending queue.
			 */
1045 1046 1047 1048 1049
			if (!skb->destructor) {
				skb->destructor = sock_wfree;
				skb->sk = sk;
				wmem_alloc_delta += skb->truesize;
			}
1050
			__skb_queue_tail(queue, skb);
L
Linus Torvalds 已提交
1051 1052 1053 1054 1055 1056
			continue;
		}

		if (copy > length)
			copy = length;

1057 1058
		if (!(rt->dst.dev->features&NETIF_F_SG) &&
		    skb_tailroom(skb) >= copy) {
L
Linus Torvalds 已提交
1059 1060 1061
			unsigned int off;

			off = skb->len;
1062
			if (getfrag(from, skb_put(skb, copy),
L
Linus Torvalds 已提交
1063 1064 1065 1066 1067 1068 1069 1070
					offset, copy, off, skb) < 0) {
				__skb_trim(skb, off);
				err = -EFAULT;
				goto error;
			}
		} else {
			int i = skb_shinfo(skb)->nr_frags;

1071 1072
			err = -ENOMEM;
			if (!sk_page_frag_refill(sk, pfrag))
L
Linus Torvalds 已提交
1073
				goto error;
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084

			if (!skb_can_coalesce(skb, i, pfrag->page,
					      pfrag->offset)) {
				err = -EMSGSIZE;
				if (i == MAX_SKB_FRAGS)
					goto error;

				__skb_fill_page_desc(skb, i, pfrag->page,
						     pfrag->offset, 0);
				skb_shinfo(skb)->nr_frags = ++i;
				get_page(pfrag->page);
L
Linus Torvalds 已提交
1085
			}
1086 1087 1088 1089 1090 1091 1092 1093
			copy = min_t(int, copy, pfrag->size - pfrag->offset);
			if (getfrag(from,
				    page_address(pfrag->page) + pfrag->offset,
				    offset, copy, skb->len, skb) < 0)
				goto error_efault;

			pfrag->offset += copy;
			skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], copy);
L
Linus Torvalds 已提交
1094 1095
			skb->len += copy;
			skb->data_len += copy;
1096
			skb->truesize += copy;
1097
			wmem_alloc_delta += copy;
L
Linus Torvalds 已提交
1098 1099 1100 1101 1102
		}
		offset += copy;
		length -= copy;
	}

1103 1104
	if (wmem_alloc_delta)
		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1105 1106
	return 0;

1107 1108
error_efault:
	err = -EFAULT;
L
Linus Torvalds 已提交
1109
error:
1110
	cork->length -= length;
P
Pavel Emelyanov 已提交
1111
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1112
	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
1113
	return err;
L
Linus Torvalds 已提交
1114 1115
}

1116 1117 1118
static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
			 struct ipcm_cookie *ipc, struct rtable **rtp)
{
1119
	struct ip_options_rcu *opt;
1120 1121
	struct rtable *rt;

1122 1123 1124 1125
	rt = *rtp;
	if (unlikely(!rt))
		return -EFAULT;

1126 1127 1128 1129 1130
	/*
	 * setup for corking.
	 */
	opt = ipc->opt;
	if (opt) {
1131
		if (!cork->opt) {
1132 1133
			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
					    sk->sk_allocation);
1134
			if (unlikely(!cork->opt))
1135 1136
				return -ENOBUFS;
		}
1137
		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1138 1139 1140
		cork->flags |= IPCORK_OPT;
		cork->addr = ipc->addr;
	}
1141

1142 1143 1144 1145
	/*
	 * We steal reference to this route, caller should not release it
	 */
	*rtp = NULL;
1146 1147
	cork->fragsize = ip_sk_use_pmtu(sk) ?
			 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
1148

1149 1150
	cork->gso_size = sk->sk_type == SOCK_DGRAM &&
			 sk->sk_protocol == IPPROTO_UDP ? ipc->gso_size : 0;
1151 1152
	cork->dst = &rt->dst;
	cork->length = 0;
1153 1154 1155
	cork->ttl = ipc->ttl;
	cork->tos = ipc->tos;
	cork->priority = ipc->priority;
1156
	cork->transmit_time = ipc->sockc.transmit_time;
1157 1158
	cork->tx_flags = 0;
	sock_tx_timestamp(sk, ipc->sockc.tsflags, &cork->tx_flags);
1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173

	return 0;
}

/*
 *	ip_append_data() and ip_append_page() can make one large IP datagram
 *	from many pieces of data. Each pieces will be holded on the socket
 *	until ip_push_pending_frames() is called. Each piece can be a page
 *	or non-page data.
 *
 *	Not only UDP, other transport protocols - e.g. raw sockets - can use
 *	this interface potentially.
 *
 *	LATER: length must be adjusted by pad at tail, when it is required.
 */
1174
int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187
		   int getfrag(void *from, char *to, int offset, int len,
			       int odd, struct sk_buff *skb),
		   void *from, int length, int transhdrlen,
		   struct ipcm_cookie *ipc, struct rtable **rtp,
		   unsigned int flags)
{
	struct inet_sock *inet = inet_sk(sk);
	int err;

	if (flags&MSG_PROBE)
		return 0;

	if (skb_queue_empty(&sk->sk_write_queue)) {
1188
		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1189 1190 1191 1192 1193 1194
		if (err)
			return err;
	} else {
		transhdrlen = 0;
	}

1195 1196
	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
				sk_page_frag(sk), getfrag,
1197 1198 1199
				from, length, transhdrlen, flags);
}

1200
ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
L
Linus Torvalds 已提交
1201 1202 1203 1204 1205 1206
		       int offset, size_t size, int flags)
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;
	struct rtable *rt;
	struct ip_options *opt = NULL;
1207
	struct inet_cork *cork;
L
Linus Torvalds 已提交
1208 1209 1210 1211
	int hh_len;
	int mtu;
	int len;
	int err;
1212
	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
L
Linus Torvalds 已提交
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222

	if (inet->hdrincl)
		return -EPERM;

	if (flags&MSG_PROBE)
		return 0;

	if (skb_queue_empty(&sk->sk_write_queue))
		return -EINVAL;

1223 1224 1225 1226
	cork = &inet->cork.base;
	rt = (struct rtable *)cork->dst;
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1227

1228
	if (!(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
1229 1230
		return -EOPNOTSUPP;

1231
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1232
	mtu = cork->gso_size ? IP_MAX_MTU : cork->fragsize;
L
Linus Torvalds 已提交
1233 1234 1235

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
1236
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
1237

1238
	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1239 1240
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
1241 1242 1243
		return -EMSGSIZE;
	}

1244 1245
	skb = skb_peek_tail(&sk->sk_write_queue);
	if (!skb)
L
Linus Torvalds 已提交
1246 1247
		return -EINVAL;

1248
	cork->length += size;
1249

L
Linus Torvalds 已提交
1250
	while (size > 0) {
1251 1252 1253 1254
		/* Check if the remaining data fits into current packet. */
		len = mtu - skb->len;
		if (len < size)
			len = maxfraglen - skb->len;
1255

L
Linus Torvalds 已提交
1256 1257 1258 1259 1260
		if (len <= 0) {
			struct sk_buff *skb_prev;
			int alloclen;

			skb_prev = skb;
1261
			fraggap = skb_prev->len - maxfraglen;
L
Linus Torvalds 已提交
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279

			alloclen = fragheaderlen + hh_len + fraggap + 15;
			skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
			if (unlikely(!skb)) {
				err = -ENOBUFS;
				goto error;
			}

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = CHECKSUM_NONE;
			skb->csum = 0;
			skb_reserve(skb, hh_len);

			/*
			 *	Find where to start putting bytes.
			 */
1280
			skb_put(skb, fragheaderlen + fraggap);
1281
			skb_reset_network_header(skb);
1282 1283
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
L
Linus Torvalds 已提交
1284
			if (fraggap) {
1285 1286
				skb->csum = skb_copy_and_csum_bits(skb_prev,
								   maxfraglen,
1287
						    skb_transport_header(skb),
1288
								   fraggap, 0);
L
Linus Torvalds 已提交
1289 1290
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
1291
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302
			}

			/*
			 * Put the packet on the pending queue.
			 */
			__skb_queue_tail(&sk->sk_write_queue, skb);
			continue;
		}

		if (len > size)
			len = size;
1303 1304

		if (skb_append_pagefrags(skb, page, offset, len)) {
L
Linus Torvalds 已提交
1305 1306 1307 1308 1309
			err = -EMSGSIZE;
			goto error;
		}

		if (skb->ip_summed == CHECKSUM_NONE) {
1310
			__wsum csum;
L
Linus Torvalds 已提交
1311 1312 1313 1314 1315 1316
			csum = csum_page(page, offset, len);
			skb->csum = csum_block_add(skb->csum, csum, skb->len);
		}

		skb->len += len;
		skb->data_len += len;
1317
		skb->truesize += len;
1318
		refcount_add(len, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1319 1320 1321 1322 1323 1324
		offset += len;
		size -= len;
	}
	return 0;

error:
1325
	cork->length -= size;
P
Pavel Emelyanov 已提交
1326
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1327 1328 1329
	return err;
}

1330
static void ip_cork_release(struct inet_cork *cork)
1331
{
1332 1333 1334 1335 1336
	cork->flags &= ~IPCORK_OPT;
	kfree(cork->opt);
	cork->opt = NULL;
	dst_release(cork->dst);
	cork->dst = NULL;
1337 1338
}

L
Linus Torvalds 已提交
1339 1340 1341 1342
/*
 *	Combined all pending IP fragments on the socket as one IP datagram
 *	and push them out.
 */
1343
struct sk_buff *__ip_make_skb(struct sock *sk,
1344
			      struct flowi4 *fl4,
1345 1346
			      struct sk_buff_head *queue,
			      struct inet_cork *cork)
L
Linus Torvalds 已提交
1347 1348 1349 1350
{
	struct sk_buff *skb, *tmp_skb;
	struct sk_buff **tail_skb;
	struct inet_sock *inet = inet_sk(sk);
1351
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
1352
	struct ip_options *opt = NULL;
1353
	struct rtable *rt = (struct rtable *)cork->dst;
L
Linus Torvalds 已提交
1354
	struct iphdr *iph;
1355
	__be16 df = 0;
L
Linus Torvalds 已提交
1356 1357
	__u8 ttl;

1358 1359
	skb = __skb_dequeue(queue);
	if (!skb)
L
Linus Torvalds 已提交
1360 1361 1362 1363
		goto out;
	tail_skb = &(skb_shinfo(skb)->frag_list);

	/* move skb->data to ip header from ext header */
1364
	if (skb->data < skb_network_header(skb))
1365
		__skb_pull(skb, skb_network_offset(skb));
1366
	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1367
		__skb_pull(tmp_skb, skb_network_header_len(skb));
L
Linus Torvalds 已提交
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380
		*tail_skb = tmp_skb;
		tail_skb = &(tmp_skb->next);
		skb->len += tmp_skb->len;
		skb->data_len += tmp_skb->len;
		skb->truesize += tmp_skb->truesize;
		tmp_skb->destructor = NULL;
		tmp_skb->sk = NULL;
	}

	/* Unless user demanded real pmtu discovery (IP_PMTUDISC_DO), we allow
	 * to fragment the frame generated here. No matter, what transforms
	 * how transforms change size of the packet, it will come out.
	 */
W
WANG Cong 已提交
1381
	skb->ignore_df = ip_sk_ignore_df(sk);
L
Linus Torvalds 已提交
1382 1383

	/* DF bit is set when we want to see DF on outgoing frames.
W
WANG Cong 已提交
1384
	 * If ignore_df is set too, we still allow to fragment this frame
L
Linus Torvalds 已提交
1385
	 * locally. */
1386 1387
	if (inet->pmtudisc == IP_PMTUDISC_DO ||
	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
1388 1389
	    (skb->len <= dst_mtu(&rt->dst) &&
	     ip_dont_fragment(sk, &rt->dst)))
L
Linus Torvalds 已提交
1390 1391
		df = htons(IP_DF);

1392 1393
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1394

1395 1396 1397
	if (cork->ttl != 0)
		ttl = cork->ttl;
	else if (rt->rt_type == RTN_MULTICAST)
L
Linus Torvalds 已提交
1398 1399
		ttl = inet->mc_ttl;
	else
1400
		ttl = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
1401

1402
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
1403 1404
	iph->version = 4;
	iph->ihl = 5;
1405
	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
L
Linus Torvalds 已提交
1406 1407 1408
	iph->frag_off = df;
	iph->ttl = ttl;
	iph->protocol = sk->sk_protocol;
1409
	ip_copy_addrs(iph, fl4);
1410
	ip_select_ident(net, skb, sk);
L
Linus Torvalds 已提交
1411

1412 1413 1414 1415 1416
	if (opt) {
		iph->ihl += opt->optlen>>2;
		ip_options_build(skb, opt, cork->addr, rt, 0);
	}

1417
	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1418
	skb->mark = sk->sk_mark;
1419
	skb->tstamp = cork->transmit_time;
1420 1421 1422 1423
	/*
	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
	 * on dst refcount
	 */
1424
	cork->dst = NULL;
1425
	skb_dst_set(skb, &rt->dst);
L
Linus Torvalds 已提交
1426

1427
	if (iph->protocol == IPPROTO_ICMP)
1428
		icmp_out_count(net, ((struct icmphdr *)
1429 1430
			skb_transport_header(skb))->type);

1431 1432 1433 1434 1435
	ip_cork_release(cork);
out:
	return skb;
}

E
Eric Dumazet 已提交
1436
int ip_send_skb(struct net *net, struct sk_buff *skb)
1437 1438 1439
{
	int err;

1440
	err = ip_local_out(net, skb->sk, skb);
L
Linus Torvalds 已提交
1441 1442
	if (err) {
		if (err > 0)
E
Eric Dumazet 已提交
1443
			err = net_xmit_errno(err);
L
Linus Torvalds 已提交
1444
		if (err)
1445
			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1446 1447 1448 1449 1450
	}

	return err;
}

1451
int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1452
{
1453 1454
	struct sk_buff *skb;

1455
	skb = ip_finish_skb(sk, fl4);
1456 1457 1458 1459
	if (!skb)
		return 0;

	/* Netfilter gets whole the not fragmented skb. */
E
Eric Dumazet 已提交
1460
	return ip_send_skb(sock_net(sk), skb);
1461 1462
}

L
Linus Torvalds 已提交
1463 1464 1465
/*
 *	Throw away all pending data on the socket.
 */
1466 1467 1468
static void __ip_flush_pending_frames(struct sock *sk,
				      struct sk_buff_head *queue,
				      struct inet_cork *cork)
L
Linus Torvalds 已提交
1469 1470 1471
{
	struct sk_buff *skb;

1472
	while ((skb = __skb_dequeue_tail(queue)) != NULL)
L
Linus Torvalds 已提交
1473 1474
		kfree_skb(skb);

1475 1476 1477 1478 1479
	ip_cork_release(cork);
}

void ip_flush_pending_frames(struct sock *sk)
{
1480
	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
L
Linus Torvalds 已提交
1481 1482
}

1483
struct sk_buff *ip_make_skb(struct sock *sk,
1484
			    struct flowi4 *fl4,
1485 1486 1487 1488
			    int getfrag(void *from, char *to, int offset,
					int len, int odd, struct sk_buff *skb),
			    void *from, int length, int transhdrlen,
			    struct ipcm_cookie *ipc, struct rtable **rtp,
W
Willem de Bruijn 已提交
1489
			    struct inet_cork *cork, unsigned int flags)
1490 1491 1492 1493 1494 1495 1496 1497 1498
{
	struct sk_buff_head queue;
	int err;

	if (flags & MSG_PROBE)
		return NULL;

	__skb_queue_head_init(&queue);

W
Willem de Bruijn 已提交
1499 1500 1501 1502
	cork->flags = 0;
	cork->addr = 0;
	cork->opt = NULL;
	err = ip_setup_cork(sk, cork, ipc, rtp);
1503 1504 1505
	if (err)
		return ERR_PTR(err);

W
Willem de Bruijn 已提交
1506
	err = __ip_append_data(sk, fl4, &queue, cork,
1507
			       &current->task_frag, getfrag,
1508 1509
			       from, length, transhdrlen, flags);
	if (err) {
W
Willem de Bruijn 已提交
1510
		__ip_flush_pending_frames(sk, &queue, cork);
1511 1512 1513
		return ERR_PTR(err);
	}

W
Willem de Bruijn 已提交
1514
	return __ip_make_skb(sk, fl4, &queue, cork);
1515
}
L
Linus Torvalds 已提交
1516 1517 1518 1519

/*
 *	Fetch data from kernel space and fill in checksum if needed.
 */
1520
static int ip_reply_glue_bits(void *dptr, char *to, int offset,
L
Linus Torvalds 已提交
1521 1522
			      int len, int odd, struct sk_buff *skb)
{
1523
	__wsum csum;
L
Linus Torvalds 已提交
1524 1525 1526

	csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
	skb->csum = csum_block_add(skb->csum, csum, odd);
1527
	return 0;
L
Linus Torvalds 已提交
1528 1529
}

1530
/*
L
Linus Torvalds 已提交
1531
 *	Generic function to send a packet as reply to another packet.
1532
 *	Used to send some TCP resets/acks so far.
L
Linus Torvalds 已提交
1533
 */
1534
void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1535 1536 1537
			   const struct ip_options *sopt,
			   __be32 daddr, __be32 saddr,
			   const struct ip_reply_arg *arg,
1538
			   unsigned int len)
L
Linus Torvalds 已提交
1539
{
1540
	struct ip_options_data replyopts;
L
Linus Torvalds 已提交
1541
	struct ipcm_cookie ipc;
1542
	struct flowi4 fl4;
E
Eric Dumazet 已提交
1543
	struct rtable *rt = skb_rtable(skb);
1544
	struct net *net = sock_net(sk);
1545
	struct sk_buff *nskb;
1546
	int err;
1547
	int oif;
L
Linus Torvalds 已提交
1548

1549
	if (__ip_options_echo(net, &replyopts.opt.opt, skb, sopt))
L
Linus Torvalds 已提交
1550 1551
		return;

1552
	ipcm_init(&ipc);
1553
	ipc.addr = daddr;
L
Linus Torvalds 已提交
1554

1555
	if (replyopts.opt.opt.optlen) {
L
Linus Torvalds 已提交
1556 1557
		ipc.opt = &replyopts.opt;

1558 1559
		if (replyopts.opt.opt.srr)
			daddr = replyopts.opt.opt.faddr;
L
Linus Torvalds 已提交
1560 1561
	}

1562
	oif = arg->bound_dev_if;
1563 1564
	if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
		oif = skb->skb_iif;
1565 1566

	flowi4_init_output(&fl4, oif,
J
Jon Maxwell 已提交
1567
			   IP4_REPLY_MARK(net, skb->mark) ?: sk->sk_mark,
1568
			   RT_TOS(arg->tos),
1569
			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1570
			   ip_reply_arg_flowi_flags(arg),
1571
			   daddr, saddr,
1572 1573
			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest,
			   arg->uid);
1574
	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1575
	rt = ip_route_output_key(net, &fl4);
1576 1577
	if (IS_ERR(rt))
		return;
L
Linus Torvalds 已提交
1578

1579
	inet_sk(sk)->tos = arg->tos;
L
Linus Torvalds 已提交
1580 1581

	sk->sk_priority = skb->priority;
1582
	sk->sk_protocol = ip_hdr(skb)->protocol;
1583
	sk->sk_bound_dev_if = arg->bound_dev_if;
1584
	sk->sk_sndbuf = sysctl_wmem_default;
1585
	sk->sk_mark = fl4.flowi4_mark;
1586 1587 1588 1589 1590 1591 1592
	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
			     len, 0, &ipc, &rt, MSG_DONTWAIT);
	if (unlikely(err)) {
		ip_flush_pending_frames(sk);
		goto out;
	}

1593 1594
	nskb = skb_peek(&sk->sk_write_queue);
	if (nskb) {
L
Linus Torvalds 已提交
1595
		if (arg->csumoffset >= 0)
1596 1597
			*((__sum16 *)skb_transport_header(nskb) +
			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1598
								arg->csum));
1599
		nskb->ip_summed = CHECKSUM_NONE;
1600
		ip_push_pending_frames(sk, &fl4);
L
Linus Torvalds 已提交
1601
	}
1602
out:
L
Linus Torvalds 已提交
1603 1604 1605 1606 1607 1608 1609 1610
	ip_rt_put(rt);
}

void __init ip_init(void)
{
	ip_rt_init();
	inet_initpeers();

1611 1612
#if defined(CONFIG_IP_MULTICAST)
	igmp_mc_init();
L
Linus Torvalds 已提交
1613 1614
#endif
}