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 45 46 47 48 49 50 51
 *		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.
 */

#include <asm/uaccess.h>
#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 76 77 78 79 80
#include <linux/skbuff.h>
#include <net/sock.h>
#include <net/arp.h>
#include <net/icmp.h>
#include <net/checksum.h>
#include <net/inetpeer.h>
#include <linux/igmp.h>
#include <linux/netfilter_ipv4.h>
#include <linux/netfilter_bridge.h>
#include <linux/mroute.h>
#include <linux/netlink.h>
81
#include <linux/tcp.h>
L
Linus Torvalds 已提交
82

83
int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
84
EXPORT_SYMBOL(sysctl_ip_default_ttl);
L
Linus Torvalds 已提交
85

86
static int ip_fragment(struct sock *sk, struct sk_buff *skb,
87
		       unsigned int mtu,
88 89
		       int (*output)(struct sock *, struct sk_buff *));

L
Linus Torvalds 已提交
90
/* Generate a checksum for an outgoing IP datagram. */
91
void ip_send_check(struct iphdr *iph)
L
Linus Torvalds 已提交
92 93 94 95
{
	iph->check = 0;
	iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
}
E
Eric Dumazet 已提交
96
EXPORT_SYMBOL(ip_send_check);
L
Linus Torvalds 已提交
97

98
static int __ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
99
{
100
	struct net *net = dev_net(skb_dst(skb)->dev);
H
Herbert Xu 已提交
101 102 103 104
	struct iphdr *iph = ip_hdr(skb);

	iph->tot_len = htons(skb->len);
	ip_send_check(iph);
105 106
	return nf_hook(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
		       net, sk, skb, NULL, skb_dst(skb)->dev,
107
		       dst_output_okfn);
108 109 110 111 112
}

int __ip_local_out(struct sk_buff *skb)
{
	return __ip_local_out_sk(skb->sk, skb);
H
Herbert Xu 已提交
113 114
}

115
int ip_local_out_sk(struct sock *sk, struct sk_buff *skb)
H
Herbert Xu 已提交
116 117 118 119 120
{
	int err;

	err = __ip_local_out(skb);
	if (likely(err == 1))
121
		err = dst_output(sk, skb);
H
Herbert Xu 已提交
122 123 124

	return err;
}
125
EXPORT_SYMBOL_GPL(ip_local_out_sk);
H
Herbert Xu 已提交
126

L
Linus Torvalds 已提交
127 128 129 130 131
static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
{
	int ttl = inet->uc_ttl;

	if (ttl < 0)
132
		ttl = ip4_dst_hoplimit(dst);
L
Linus Torvalds 已提交
133 134 135
	return ttl;
}

136
/*
L
Linus Torvalds 已提交
137 138 139
 *		Add an ip header to a skbuff and send it out.
 *
 */
140
int ip_build_and_send_pkt(struct sk_buff *skb, const struct sock *sk,
141
			  __be32 saddr, __be32 daddr, struct ip_options_rcu *opt)
L
Linus Torvalds 已提交
142 143
{
	struct inet_sock *inet = inet_sk(sk);
E
Eric Dumazet 已提交
144
	struct rtable *rt = skb_rtable(skb);
L
Linus Torvalds 已提交
145 146 147
	struct iphdr *iph;

	/* Build the IP header. */
148
	skb_push(skb, sizeof(struct iphdr) + (opt ? opt->opt.optlen : 0));
149
	skb_reset_network_header(skb);
150
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
151 152 153
	iph->version  = 4;
	iph->ihl      = 5;
	iph->tos      = inet->tos;
154
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
155 156
	iph->daddr    = (opt && opt->opt.srr ? opt->opt.faddr : daddr);
	iph->saddr    = saddr;
L
Linus Torvalds 已提交
157
	iph->protocol = sk->sk_protocol;
158 159 160 161 162 163 164
	if (ip_dont_fragment(sk, &rt->dst)) {
		iph->frag_off = htons(IP_DF);
		iph->id = 0;
	} else {
		iph->frag_off = 0;
		__ip_select_ident(sock_net(sk), iph, 1);
	}
L
Linus Torvalds 已提交
165

166 167 168
	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
169 170 171
	}

	skb->priority = sk->sk_priority;
172
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
173 174

	/* Send it out. */
H
Herbert Xu 已提交
175
	return ip_local_out(skb);
L
Linus Torvalds 已提交
176
}
177 178
EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);

179
static int ip_finish_output2(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
180
{
E
Eric Dumazet 已提交
181
	struct dst_entry *dst = skb_dst(skb);
182
	struct rtable *rt = (struct rtable *)dst;
L
Linus Torvalds 已提交
183
	struct net_device *dev = dst->dev;
184
	struct net *net = dev_net(dev);
185
	unsigned int hh_len = LL_RESERVED_SPACE(dev);
186
	struct neighbour *neigh;
187
	u32 nexthop;
L
Linus Torvalds 已提交
188

189
	if (rt->rt_type == RTN_MULTICAST) {
190
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);
191
	} else if (rt->rt_type == RTN_BROADCAST)
192
		IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);
193

L
Linus Torvalds 已提交
194
	/* Be paranoid, rather than too clever. */
195
	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
L
Linus Torvalds 已提交
196 197 198
		struct sk_buff *skb2;

		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
199
		if (!skb2) {
L
Linus Torvalds 已提交
200 201 202 203 204
			kfree_skb(skb);
			return -ENOMEM;
		}
		if (skb->sk)
			skb_set_owner_w(skb2, skb->sk);
205
		consume_skb(skb);
L
Linus Torvalds 已提交
206 207 208
		skb = skb2;
	}

209
	rcu_read_lock_bh();
J
Julian Anastasov 已提交
210
	nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);
211 212 213
	neigh = __ipv4_neigh_lookup_noref(dev, nexthop);
	if (unlikely(!neigh))
		neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);
214
	if (!IS_ERR(neigh)) {
215
		int res = dst_neigh_output(dst, neigh, skb);
216

217
		rcu_read_unlock_bh();
218 219
		return res;
	}
220
	rcu_read_unlock_bh();
221

222 223
	net_dbg_ratelimited("%s: No header cache and no neighbour!\n",
			    __func__);
L
Linus Torvalds 已提交
224 225 226 227
	kfree_skb(skb);
	return -EINVAL;
}

228 229
static int ip_finish_output_gso(struct sock *sk, struct sk_buff *skb,
				unsigned int mtu)
230 231 232 233 234 235 236
{
	netdev_features_t features;
	struct sk_buff *segs;
	int ret = 0;

	/* common case: locally created skb or seglen is <= mtu */
	if (((IPCB(skb)->flags & IPSKB_FORWARDED) == 0) ||
237
	      skb_gso_network_seglen(skb) <= mtu)
238
		return ip_finish_output2(sk, skb);
239 240 241 242 243 244 245 246 247 248

	/* Slowpath -  GSO segment length is exceeding the dst MTU.
	 *
	 * This can happen in two cases:
	 * 1) TCP GRO packet, DF bit not set
	 * 2) skb arrived via virtio-net, we thus get TSO/GSO skbs directly
	 * from host network stack.
	 */
	features = netif_skb_features(skb);
	segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK);
249
	if (IS_ERR_OR_NULL(segs)) {
250 251 252 253 254 255 256 257 258 259 260
		kfree_skb(skb);
		return -ENOMEM;
	}

	consume_skb(skb);

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

		segs->next = NULL;
261
		err = ip_fragment(sk, segs, mtu, ip_finish_output2);
262 263 264 265 266 267 268 269 270

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

	return ret;
}

271
static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
272
{
273 274
	unsigned int mtu;

275 276
#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
	/* Policy lookup after SNAT yielded a new policy */
277
	if (skb_dst(skb)->xfrm) {
278
		IPCB(skb)->flags |= IPSKB_REROUTED;
279
		return dst_output(sk, skb);
280
	}
281
#endif
282
	mtu = ip_skb_dst_mtu(skb);
283
	if (skb_is_gso(skb))
284
		return ip_finish_output_gso(sk, skb, mtu);
285

286
	if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))
287
		return ip_fragment(sk, skb, mtu, ip_finish_output2);
288

289
	return ip_finish_output2(sk, skb);
L
Linus Torvalds 已提交
290 291
}

292
int ip_mc_output(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
293
{
E
Eric Dumazet 已提交
294
	struct rtable *rt = skb_rtable(skb);
295
	struct net_device *dev = rt->dst.dev;
296
	struct net *net = dev_net(dev);
L
Linus Torvalds 已提交
297 298 299 300

	/*
	 *	If the indicated interface is up and running, send the packet.
	 */
301
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
302 303 304 305 306 307 308 309 310

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

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

	if (rt->rt_flags&RTCF_MULTICAST) {
311
		if (sk_mc_loop(sk)
L
Linus Torvalds 已提交
312 313 314 315 316 317 318 319 320
#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.
		 */
321 322 323
		    &&
		    ((rt->rt_flags & RTCF_LOCAL) ||
		     !(IPCB(skb)->flags & IPSKB_FORWARDED))
L
Linus Torvalds 已提交
324
#endif
325
		   ) {
L
Linus Torvalds 已提交
326 327
			struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
			if (newskb)
328
				NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
329
					net, sk, newskb, NULL, newskb->dev,
330
					dev_loopback_xmit);
L
Linus Torvalds 已提交
331 332 333 334
		}

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

335
		if (ip_hdr(skb)->ttl == 0) {
L
Linus Torvalds 已提交
336 337 338 339 340 341 342 343
			kfree_skb(skb);
			return 0;
		}
	}

	if (rt->rt_flags&RTCF_BROADCAST) {
		struct sk_buff *newskb = skb_clone(skb, GFP_ATOMIC);
		if (newskb)
344 345 346
			NF_HOOK(NFPROTO_IPV4, NF_INET_POST_ROUTING,
				net, sk, newskb, NULL, newskb->dev,
				dev_loopback_xmit);
L
Linus Torvalds 已提交
347 348
	}

349 350 351
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, skb->dev,
			    ip_finish_output,
352
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
353 354
}

355
int ip_output(struct sock *sk, struct sk_buff *skb)
L
Linus Torvalds 已提交
356
{
E
Eric Dumazet 已提交
357
	struct net_device *dev = skb_dst(skb)->dev;
358
	struct net *net = dev_net(dev);
359

360
	IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);
L
Linus Torvalds 已提交
361

362 363 364
	skb->dev = dev;
	skb->protocol = htons(ETH_P_IP);

365 366
	return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,
			    net, sk, skb, NULL, dev,
367
			    ip_finish_output,
368
			    !(IPCB(skb)->flags & IPSKB_REROUTED));
L
Linus Torvalds 已提交
369 370
}

371 372 373 374 375 376 377 378 379 380 381 382 383 384
/*
 * 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));
}

385 386
/* Note: skb->sk can be different from sk, in case of tunnels */
int ip_queue_xmit(struct sock *sk, struct sk_buff *skb, struct flowi *fl)
L
Linus Torvalds 已提交
387 388
{
	struct inet_sock *inet = inet_sk(sk);
389
	struct ip_options_rcu *inet_opt;
390
	struct flowi4 *fl4;
L
Linus Torvalds 已提交
391 392
	struct rtable *rt;
	struct iphdr *iph;
393
	int res;
L
Linus Torvalds 已提交
394 395 396 397

	/* Skip all of this if the packet is already routed,
	 * f.e. by something like SCTP.
	 */
398
	rcu_read_lock();
399
	inet_opt = rcu_dereference(inet->inet_opt);
400
	fl4 = &fl->u.ip4;
E
Eric Dumazet 已提交
401
	rt = skb_rtable(skb);
402
	if (rt)
L
Linus Torvalds 已提交
403 404 405 406
		goto packet_routed;

	/* Make sure we can route this packet. */
	rt = (struct rtable *)__sk_dst_check(sk, 0);
407
	if (!rt) {
A
Al Viro 已提交
408
		__be32 daddr;
L
Linus Torvalds 已提交
409 410

		/* Use correct destination address if we have options. */
E
Eric Dumazet 已提交
411
		daddr = inet->inet_daddr;
412 413
		if (inet_opt && inet_opt->opt.srr)
			daddr = inet_opt->opt.faddr;
L
Linus Torvalds 已提交
414

415 416 417 418
		/* If this fails, retransmit mechanism of transport layer will
		 * keep trying until route appears or the connection times
		 * itself out.
		 */
419
		rt = ip_route_output_ports(sock_net(sk), fl4, sk,
420 421 422 423 424 425 426 427
					   daddr, inet->inet_saddr,
					   inet->inet_dport,
					   inet->inet_sport,
					   sk->sk_protocol,
					   RT_CONN_FLAGS(sk),
					   sk->sk_bound_dev_if);
		if (IS_ERR(rt))
			goto no_route;
428
		sk_setup_caps(sk, &rt->dst);
L
Linus Torvalds 已提交
429
	}
430
	skb_dst_set_noref(skb, &rt->dst);
L
Linus Torvalds 已提交
431 432

packet_routed:
J
Julian Anastasov 已提交
433
	if (inet_opt && inet_opt->opt.is_strictroute && rt->rt_uses_gateway)
L
Linus Torvalds 已提交
434 435 436
		goto no_route;

	/* OK, we know where to send it, allocate and build IP header. */
437
	skb_push(skb, sizeof(struct iphdr) + (inet_opt ? inet_opt->opt.optlen : 0));
438
	skb_reset_network_header(skb);
439
	iph = ip_hdr(skb);
440
	*((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
W
WANG Cong 已提交
441
	if (ip_dont_fragment(sk, &rt->dst) && !skb->ignore_df)
L
Linus Torvalds 已提交
442 443 444
		iph->frag_off = htons(IP_DF);
	else
		iph->frag_off = 0;
445
	iph->ttl      = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
446
	iph->protocol = sk->sk_protocol;
447 448
	ip_copy_addrs(iph, fl4);

L
Linus Torvalds 已提交
449 450
	/* Transport layer set skb->h.foo itself. */

451 452 453
	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
454 455
	}

456 457
	ip_select_ident_segs(sock_net(sk), skb, sk,
			     skb_shinfo(skb)->gso_segs ?: 1);
L
Linus Torvalds 已提交
458

459
	/* TODO : should we use skb->sk here instead of sk ? */
L
Linus Torvalds 已提交
460
	skb->priority = sk->sk_priority;
461
	skb->mark = sk->sk_mark;
L
Linus Torvalds 已提交
462

463 464 465
	res = ip_local_out(skb);
	rcu_read_unlock();
	return res;
L
Linus Torvalds 已提交
466 467

no_route:
468
	rcu_read_unlock();
P
Pavel Emelyanov 已提交
469
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
L
Linus Torvalds 已提交
470 471 472
	kfree_skb(skb);
	return -EHOSTUNREACH;
}
E
Eric Dumazet 已提交
473
EXPORT_SYMBOL(ip_queue_xmit);
L
Linus Torvalds 已提交
474 475 476 477 478 479

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
480
	skb_dst_drop(to);
481
	skb_dst_copy(to, from);
L
Linus Torvalds 已提交
482
	to->dev = from->dev;
T
Thomas Graf 已提交
483
	to->mark = from->mark;
L
Linus Torvalds 已提交
484 485 486 487 488 489 490

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

#ifdef CONFIG_NET_SCHED
	to->tc_index = from->tc_index;
#endif
491
	nf_copy(to, from);
492 493
#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
	to->ipvs_property = from->ipvs_property;
L
Linus Torvalds 已提交
494
#endif
495
	skb_copy_secmark(to, from);
L
Linus Torvalds 已提交
496 497
}

498
static int ip_fragment(struct sock *sk, struct sk_buff *skb,
499
		       unsigned int mtu,
500 501 502 503
		       int (*output)(struct sock *, struct sk_buff *))
{
	struct iphdr *iph = ip_hdr(skb);

504 505 506 507
	if ((iph->frag_off & htons(IP_DF)) == 0)
		return ip_do_fragment(sk, skb, output);

	if (unlikely(!skb->ignore_df ||
508 509
		     (IPCB(skb)->frag_max_size &&
		      IPCB(skb)->frag_max_size > mtu))) {
510
		struct net *net = dev_net(skb_rtable(skb)->dst.dev);
511

512
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
513 514 515 516 517 518 519 520 521
		icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
			  htonl(mtu));
		kfree_skb(skb);
		return -EMSGSIZE;
	}

	return ip_do_fragment(sk, skb, output);
}

L
Linus Torvalds 已提交
522 523 524 525 526 527 528
/*
 *	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.
 */

529 530
int ip_do_fragment(struct sock *sk, struct sk_buff *skb,
		   int (*output)(struct sock *, struct sk_buff *))
L
Linus Torvalds 已提交
531 532 533 534 535
{
	struct iphdr *iph;
	int ptr;
	struct net_device *dev;
	struct sk_buff *skb2;
536
	unsigned int mtu, hlen, left, len, ll_rs;
L
Linus Torvalds 已提交
537
	int offset;
538
	__be16 not_last_frag;
E
Eric Dumazet 已提交
539
	struct rtable *rt = skb_rtable(skb);
540
	struct net *net;
L
Linus Torvalds 已提交
541 542
	int err = 0;

543
	dev = rt->dst.dev;
544
	net = dev_net(dev);
L
Linus Torvalds 已提交
545 546 547 548 549

	/*
	 *	Point into the IP datagram header.
	 */

550
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
551

552
	mtu = ip_skb_dst_mtu(skb);
553 554
	if (IPCB(skb)->frag_max_size && IPCB(skb)->frag_max_size < mtu)
		mtu = IPCB(skb)->frag_max_size;
L
Linus Torvalds 已提交
555 556 557 558 559 560

	/*
	 *	Setup starting values.
	 */

	hlen = iph->ihl * 4;
561
	mtu = mtu - hlen;	/* Size of data space */
H
Herbert Xu 已提交
562
	IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
L
Linus Torvalds 已提交
563 564 565 566 567 568 569 570

	/* 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.
	 */
571
	if (skb_has_frag_list(skb)) {
572
		struct sk_buff *frag, *frag2;
L
Linus Torvalds 已提交
573 574 575 576
		int first_len = skb_pagelen(skb);

		if (first_len - hlen > mtu ||
		    ((first_len - hlen) & 7) ||
577
		    ip_is_fragment(iph) ||
L
Linus Torvalds 已提交
578 579 580
		    skb_cloned(skb))
			goto slow_path;

581
		skb_walk_frags(skb, frag) {
L
Linus Torvalds 已提交
582 583 584 585
			/* Correct geometry. */
			if (frag->len > mtu ||
			    ((frag->len & 7) && frag->next) ||
			    skb_headroom(frag) < hlen)
586
				goto slow_path_clean;
L
Linus Torvalds 已提交
587 588 589

			/* Partially cloned skb? */
			if (skb_shared(frag))
590
				goto slow_path_clean;
591 592 593 594 595 596

			BUG_ON(frag->sk);
			if (skb->sk) {
				frag->sk = skb->sk;
				frag->destructor = sock_wfree;
			}
597
			skb->truesize -= frag->truesize;
L
Linus Torvalds 已提交
598 599 600 601 602 603 604
		}

		/* Everything is OK. Generate! */

		err = 0;
		offset = 0;
		frag = skb_shinfo(skb)->frag_list;
605
		skb_frag_list_init(skb);
L
Linus Torvalds 已提交
606 607 608 609 610 611 612 613 614 615 616
		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;
617
				skb_reset_transport_header(frag);
618 619
				__skb_push(frag, hlen);
				skb_reset_network_header(frag);
620
				memcpy(skb_network_header(frag), iph, hlen);
621
				iph = ip_hdr(frag);
L
Linus Torvalds 已提交
622 623 624 625 626 627
				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);
628
				if (frag->next)
L
Linus Torvalds 已提交
629 630 631 632 633
					iph->frag_off |= htons(IP_MF);
				/* Ready, complete checksum */
				ip_send_check(iph);
			}

634
			err = output(sk, skb);
L
Linus Torvalds 已提交
635

636
			if (!err)
637
				IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
638 639 640 641 642 643 644 645 646
			if (err || !frag)
				break;

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

		if (err == 0) {
647
			IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
648 649 650 651 652 653 654 655
			return 0;
		}

		while (frag) {
			skb = frag->next;
			kfree_skb(frag);
			frag = skb;
		}
656
		IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
657
		return err;
658 659 660 661 662 663 664 665 666

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
667 668 669
	}

slow_path:
670 671 672
	/* for offloaded checksums cleanup checksum before fragmentation */
	if ((skb->ip_summed == CHECKSUM_PARTIAL) && skb_checksum_help(skb))
		goto fail;
673
	iph = ip_hdr(skb);
674

L
Linus Torvalds 已提交
675
	left = skb->len - hlen;		/* Space per frame */
676
	ptr = hlen;		/* Where to start from */
L
Linus Torvalds 已提交
677

678
	ll_rs = LL_RESERVED_SPACE(rt->dst.dev);
679

L
Linus Torvalds 已提交
680 681 682 683 684 685 686 687 688 689 690
	/*
	 *	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
691
	while (left > 0) {
L
Linus Torvalds 已提交
692 693 694 695
		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
696
		/* IF: we are not sending up to and including the packet end
L
Linus Torvalds 已提交
697 698 699 700 701
		   then align the next start on an eight byte boundary */
		if (len < left)	{
			len &= ~7;
		}

702 703 704
		/* Allocate buffer */
		skb2 = alloc_skb(len + hlen + ll_rs, GFP_ATOMIC);
		if (!skb2) {
L
Linus Torvalds 已提交
705 706 707 708 709 710 711 712 713 714 715
			err = -ENOMEM;
			goto fail;
		}

		/*
		 *	Set up data on packet
		 */

		ip_copy_metadata(skb2, skb);
		skb_reserve(skb2, ll_rs);
		skb_put(skb2, len + hlen);
716
		skb_reset_network_header(skb2);
717
		skb2->transport_header = skb2->network_header + hlen;
L
Linus Torvalds 已提交
718 719 720 721 722 723 724 725 726 727 728 729 730

		/*
		 *	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.
		 */

731
		skb_copy_from_linear_data(skb, skb_network_header(skb2), hlen);
L
Linus Torvalds 已提交
732 733 734 735

		/*
		 *	Copy a block of the IP datagram.
		 */
736
		if (skb_copy_bits(skb, ptr, skb_transport_header(skb2), len))
L
Linus Torvalds 已提交
737 738 739 740 741 742
			BUG();
		left -= len;

		/*
		 *	Fill in the new header fields.
		 */
743
		iph = ip_hdr(skb2);
L
Linus Torvalds 已提交
744 745
		iph->frag_off = htons((offset >> 3));

746 747 748
		if (IPCB(skb)->flags & IPSKB_FRAG_PMTU)
			iph->frag_off |= htons(IP_DF);

L
Linus Torvalds 已提交
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773
		/* 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);

774
		err = output(sk, skb2);
L
Linus Torvalds 已提交
775 776
		if (err)
			goto fail;
777

778
		IP_INC_STATS(net, IPSTATS_MIB_FRAGCREATES);
L
Linus Torvalds 已提交
779
	}
780
	consume_skb(skb);
781
	IP_INC_STATS(net, IPSTATS_MIB_FRAGOKS);
L
Linus Torvalds 已提交
782 783 784
	return err;

fail:
785
	kfree_skb(skb);
786
	IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
L
Linus Torvalds 已提交
787 788
	return err;
}
789
EXPORT_SYMBOL(ip_do_fragment);
790

L
Linus Torvalds 已提交
791 792 793
int
ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
{
794
	struct msghdr *msg = from;
L
Linus Torvalds 已提交
795

796
	if (skb->ip_summed == CHECKSUM_PARTIAL) {
797
		if (copy_from_iter(to, len, &msg->msg_iter) != len)
L
Linus Torvalds 已提交
798 799
			return -EFAULT;
	} else {
800
		__wsum csum = 0;
801
		if (csum_and_copy_from_iter(to, len, &csum, &msg->msg_iter) != len)
L
Linus Torvalds 已提交
802 803 804 805 806
			return -EFAULT;
		skb->csum = csum_block_add(skb->csum, csum, odd);
	}
	return 0;
}
E
Eric Dumazet 已提交
807
EXPORT_SYMBOL(ip_generic_getfrag);
L
Linus Torvalds 已提交
808

809
static inline __wsum
L
Linus Torvalds 已提交
810 811 812
csum_page(struct page *page, int offset, int copy)
{
	char *kaddr;
813
	__wsum csum;
L
Linus Torvalds 已提交
814 815 816 817 818 819
	kaddr = kmap(page);
	csum = csum_partial(kaddr + offset, copy, 0);
	kunmap(page);
	return csum;
}

A
Adrian Bunk 已提交
820
static inline int ip_ufo_append_data(struct sock *sk,
821
			struct sk_buff_head *queue,
822 823 824
			int getfrag(void *from, char *to, int offset, int len,
			       int odd, struct sk_buff *skb),
			void *from, int length, int hh_len, int fragheaderlen,
825
			int transhdrlen, int maxfraglen, unsigned int flags)
826 827 828 829 830 831 832 833
{
	struct sk_buff *skb;
	int err;

	/* There is support for UDP fragmentation offload by network
	 * device, so create one single skb packet containing complete
	 * udp datagram
	 */
834 835
	skb = skb_peek_tail(queue);
	if (!skb) {
836 837 838 839
		skb = sock_alloc_send_skb(sk,
			hh_len + fragheaderlen + transhdrlen + 20,
			(flags & MSG_DONTWAIT), &err);

840
		if (!skb)
841 842 843 844 845 846
			return err;

		/* reserve space for Hardware header */
		skb_reserve(skb, hh_len);

		/* create space for UDP/IP header */
847
		skb_put(skb, fragheaderlen + transhdrlen);
848 849

		/* initialize network header pointer */
850
		skb_reset_network_header(skb);
851 852

		/* initialize protocol header pointer */
853
		skb->transport_header = skb->network_header + fragheaderlen;
854 855 856

		skb->csum = 0;

857
		__skb_queue_tail(queue, skb);
858 859
	} else if (skb_is_gso(skb)) {
		goto append;
860
	}
861

862 863 864 865 866 867
	skb->ip_summed = CHECKSUM_PARTIAL;
	/* specify the length of each IP datagram fragment */
	skb_shinfo(skb)->gso_size = maxfraglen - fragheaderlen;
	skb_shinfo(skb)->gso_type = SKB_GSO_UDP;

append:
868 869
	return skb_append_datato_frags(sk, skb, getfrag, from,
				       (length - transhdrlen));
870 871
}

872 873 874
static int __ip_append_data(struct sock *sk,
			    struct flowi4 *fl4,
			    struct sk_buff_head *queue,
875
			    struct inet_cork *cork,
876
			    struct page_frag *pfrag,
877 878 879 880
			    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
881 882 883 884
{
	struct inet_sock *inet = inet_sk(sk);
	struct sk_buff *skb;

885
	struct ip_options *opt = cork->opt;
L
Linus Torvalds 已提交
886 887 888 889 890 891
	int hh_len;
	int exthdrlen;
	int mtu;
	int copy;
	int err;
	int offset = 0;
892
	unsigned int maxfraglen, fragheaderlen, maxnonfragsize;
L
Linus Torvalds 已提交
893
	int csummode = CHECKSUM_NONE;
894
	struct rtable *rt = (struct rtable *)cork->dst;
895
	u32 tskey = 0;
L
Linus Torvalds 已提交
896

897 898 899
	skb = skb_peek_tail(queue);

	exthdrlen = !skb ? rt->dst.header_len : 0;
900
	mtu = cork->fragsize;
901 902 903
	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
904

905
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
L
Linus Torvalds 已提交
906 907 908

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
909
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
910

911
	if (cork->length + length > maxnonfragsize - fragheaderlen) {
912
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
913
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
914 915 916 917 918 919 920 921 922
		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 &&
923
	    rt->dst.dev->features & NETIF_F_V4_CSUM &&
L
Linus Torvalds 已提交
924
	    !exthdrlen)
925
		csummode = CHECKSUM_PARTIAL;
L
Linus Torvalds 已提交
926

927
	cork->length += length;
928
	if (((length > mtu) || (skb && skb_is_gso(skb))) &&
929
	    (sk->sk_protocol == IPPROTO_UDP) &&
930 931
	    (rt->dst.dev->features & NETIF_F_UFO) && !rt->dst.header_len &&
	    (sk->sk_type == SOCK_DGRAM)) {
932 933
		err = ip_ufo_append_data(sk, queue, getfrag, from, length,
					 hh_len, fragheaderlen, transhdrlen,
934
					 maxfraglen, flags);
935
		if (err)
936 937 938
			goto error;
		return 0;
	}
L
Linus Torvalds 已提交
939 940 941 942 943 944 945 946

	/* 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.
	 */

947
	if (!skb)
L
Linus Torvalds 已提交
948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977
		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;
			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;

978
			if ((flags & MSG_MORE) &&
979
			    !(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
980 981
				alloclen = mtu;
			else
982
				alloclen = fraglen;
L
Linus Torvalds 已提交
983

984 985
			alloclen += exthdrlen;

L
Linus Torvalds 已提交
986 987 988 989 990
			/* 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.
			 */
991
			if (datalen == length + fraggap)
992
				alloclen += rt->dst.trailer_len;
993

L
Linus Torvalds 已提交
994
			if (transhdrlen) {
995
				skb = sock_alloc_send_skb(sk,
L
Linus Torvalds 已提交
996 997 998 999 1000 1001
						alloclen + hh_len + 15,
						(flags & MSG_DONTWAIT), &err);
			} else {
				skb = NULL;
				if (atomic_read(&sk->sk_wmem_alloc) <=
				    2 * sk->sk_sndbuf)
1002
					skb = sock_wmalloc(sk,
L
Linus Torvalds 已提交
1003 1004
							   alloclen + hh_len + 15, 1,
							   sk->sk_allocation);
1005
				if (unlikely(!skb))
L
Linus Torvalds 已提交
1006 1007
					err = -ENOBUFS;
			}
1008
			if (!skb)
L
Linus Torvalds 已提交
1009 1010 1011 1012 1013 1014 1015 1016
				goto error;

			/*
			 *	Fill in the control structures
			 */
			skb->ip_summed = csummode;
			skb->csum = 0;
			skb_reserve(skb, hh_len);
1017 1018

			/* only the initial fragment is time stamped */
1019
			skb_shinfo(skb)->tx_flags = cork->tx_flags;
1020
			cork->tx_flags = 0;
1021 1022
			skb_shinfo(skb)->tskey = tskey;
			tskey = 0;
L
Linus Torvalds 已提交
1023 1024 1025 1026

			/*
			 *	Find where to start putting bytes.
			 */
1027
			data = skb_put(skb, fraglen + exthdrlen);
1028
			skb_set_network_header(skb, exthdrlen);
1029 1030
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
1031
			data += fragheaderlen + exthdrlen;
L
Linus Torvalds 已提交
1032 1033 1034 1035 1036 1037 1038 1039

			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;
1040
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058
			}

			copy = datalen - transhdrlen - fraggap;
			if (copy > 0 && getfrag(from, data + transhdrlen, offset, copy, fraggap, skb) < 0) {
				err = -EFAULT;
				kfree_skb(skb);
				goto error;
			}

			offset += copy;
			length -= datalen - fraggap;
			transhdrlen = 0;
			exthdrlen = 0;
			csummode = CHECKSUM_NONE;

			/*
			 * Put the packet on the pending queue.
			 */
1059
			__skb_queue_tail(queue, skb);
L
Linus Torvalds 已提交
1060 1061 1062 1063 1064 1065
			continue;
		}

		if (copy > length)
			copy = length;

1066
		if (!(rt->dst.dev->features&NETIF_F_SG)) {
L
Linus Torvalds 已提交
1067 1068 1069
			unsigned int off;

			off = skb->len;
1070
			if (getfrag(from, skb_put(skb, copy),
L
Linus Torvalds 已提交
1071 1072 1073 1074 1075 1076 1077 1078
					offset, copy, off, skb) < 0) {
				__skb_trim(skb, off);
				err = -EFAULT;
				goto error;
			}
		} else {
			int i = skb_shinfo(skb)->nr_frags;

1079 1080
			err = -ENOMEM;
			if (!sk_page_frag_refill(sk, pfrag))
L
Linus Torvalds 已提交
1081
				goto error;
1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092

			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
1093
			}
1094 1095 1096 1097 1098 1099 1100 1101
			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
1102 1103
			skb->len += copy;
			skb->data_len += copy;
1104 1105
			skb->truesize += copy;
			atomic_add(copy, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1106 1107 1108 1109 1110 1111 1112
		}
		offset += copy;
		length -= copy;
	}

	return 0;

1113 1114
error_efault:
	err = -EFAULT;
L
Linus Torvalds 已提交
1115
error:
1116
	cork->length -= length;
P
Pavel Emelyanov 已提交
1117
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
1118
	return err;
L
Linus Torvalds 已提交
1119 1120
}

1121 1122 1123
static int ip_setup_cork(struct sock *sk, struct inet_cork *cork,
			 struct ipcm_cookie *ipc, struct rtable **rtp)
{
1124
	struct ip_options_rcu *opt;
1125 1126 1127 1128 1129 1130 1131
	struct rtable *rt;

	/*
	 * setup for corking.
	 */
	opt = ipc->opt;
	if (opt) {
1132
		if (!cork->opt) {
1133 1134
			cork->opt = kmalloc(sizeof(struct ip_options) + 40,
					    sk->sk_allocation);
1135
			if (unlikely(!cork->opt))
1136 1137
				return -ENOBUFS;
		}
1138
		memcpy(cork->opt, &opt->opt, sizeof(struct ip_options) + opt->opt.optlen);
1139 1140 1141 1142 1143 1144 1145 1146 1147 1148
		cork->flags |= IPCORK_OPT;
		cork->addr = ipc->addr;
	}
	rt = *rtp;
	if (unlikely(!rt))
		return -EFAULT;
	/*
	 * We steal reference to this route, caller should not release it
	 */
	*rtp = NULL;
1149 1150
	cork->fragsize = ip_sk_use_pmtu(sk) ?
			 dst_mtu(&rt->dst) : rt->dst.dev->mtu;
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 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
	cork->tx_flags = ipc->tx_flags;

	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.
 */
1172
int ip_append_data(struct sock *sk, struct flowi4 *fl4,
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185
		   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)) {
1186
		err = ip_setup_cork(sk, &inet->cork.base, ipc, rtp);
1187 1188 1189 1190 1191 1192
		if (err)
			return err;
	} else {
		transhdrlen = 0;
	}

1193 1194
	return __ip_append_data(sk, fl4, &sk->sk_write_queue, &inet->cork.base,
				sk_page_frag(sk), getfrag,
1195 1196 1197
				from, length, transhdrlen, flags);
}

1198
ssize_t	ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
L
Linus Torvalds 已提交
1199 1200 1201 1202 1203 1204
		       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;
1205
	struct inet_cork *cork;
L
Linus Torvalds 已提交
1206 1207 1208 1209
	int hh_len;
	int mtu;
	int len;
	int err;
1210
	unsigned int maxfraglen, fragheaderlen, fraggap, maxnonfragsize;
L
Linus Torvalds 已提交
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220

	if (inet->hdrincl)
		return -EPERM;

	if (flags&MSG_PROBE)
		return 0;

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

1221 1222 1223 1224
	cork = &inet->cork.base;
	rt = (struct rtable *)cork->dst;
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1225

1226
	if (!(rt->dst.dev->features&NETIF_F_SG))
L
Linus Torvalds 已提交
1227 1228
		return -EOPNOTSUPP;

1229
	hh_len = LL_RESERVED_SPACE(rt->dst.dev);
1230
	mtu = cork->fragsize;
L
Linus Torvalds 已提交
1231 1232 1233

	fragheaderlen = sizeof(struct iphdr) + (opt ? opt->optlen : 0);
	maxfraglen = ((mtu - fragheaderlen) & ~7) + fragheaderlen;
W
WANG Cong 已提交
1234
	maxnonfragsize = ip_sk_ignore_df(sk) ? 0xFFFF : mtu;
L
Linus Torvalds 已提交
1235

1236
	if (cork->length + size > maxnonfragsize - fragheaderlen) {
1237 1238
		ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
			       mtu - (opt ? opt->optlen : 0));
L
Linus Torvalds 已提交
1239 1240 1241
		return -EMSGSIZE;
	}

1242 1243
	skb = skb_peek_tail(&sk->sk_write_queue);
	if (!skb)
L
Linus Torvalds 已提交
1244 1245
		return -EINVAL;

1246
	cork->length += size;
1247 1248
	if ((size + skb->len > mtu) &&
	    (sk->sk_protocol == IPPROTO_UDP) &&
1249
	    (rt->dst.dev->features & NETIF_F_UFO)) {
1250
		skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
H
Herbert Xu 已提交
1251
		skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
1252
	}
1253

L
Linus Torvalds 已提交
1254
	while (size > 0) {
1255
		if (skb_is_gso(skb)) {
1256
			len = size;
1257
		} else {
1258 1259 1260 1261 1262 1263

			/* Check if the remaining data fits into current packet. */
			len = mtu - skb->len;
			if (len < size)
				len = maxfraglen - skb->len;
		}
L
Linus Torvalds 已提交
1264 1265 1266 1267 1268
		if (len <= 0) {
			struct sk_buff *skb_prev;
			int alloclen;

			skb_prev = skb;
1269
			fraggap = skb_prev->len - maxfraglen;
L
Linus Torvalds 已提交
1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287

			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.
			 */
1288
			skb_put(skb, fragheaderlen + fraggap);
1289
			skb_reset_network_header(skb);
1290 1291
			skb->transport_header = (skb->network_header +
						 fragheaderlen);
L
Linus Torvalds 已提交
1292
			if (fraggap) {
1293 1294
				skb->csum = skb_copy_and_csum_bits(skb_prev,
								   maxfraglen,
1295
						    skb_transport_header(skb),
1296
								   fraggap, 0);
L
Linus Torvalds 已提交
1297 1298
				skb_prev->csum = csum_sub(skb_prev->csum,
							  skb->csum);
1299
				pskb_trim_unique(skb_prev, maxfraglen);
L
Linus Torvalds 已提交
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
			}

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

		if (len > size)
			len = size;
1311 1312

		if (skb_append_pagefrags(skb, page, offset, len)) {
L
Linus Torvalds 已提交
1313 1314 1315 1316 1317
			err = -EMSGSIZE;
			goto error;
		}

		if (skb->ip_summed == CHECKSUM_NONE) {
1318
			__wsum csum;
L
Linus Torvalds 已提交
1319 1320 1321 1322 1323 1324
			csum = csum_page(page, offset, len);
			skb->csum = csum_block_add(skb->csum, csum, skb->len);
		}

		skb->len += len;
		skb->data_len += len;
1325 1326
		skb->truesize += len;
		atomic_add(len, &sk->sk_wmem_alloc);
L
Linus Torvalds 已提交
1327 1328 1329 1330 1331 1332
		offset += len;
		size -= len;
	}
	return 0;

error:
1333
	cork->length -= size;
P
Pavel Emelyanov 已提交
1334
	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1335 1336 1337
	return err;
}

1338
static void ip_cork_release(struct inet_cork *cork)
1339
{
1340 1341 1342 1343 1344
	cork->flags &= ~IPCORK_OPT;
	kfree(cork->opt);
	cork->opt = NULL;
	dst_release(cork->dst);
	cork->dst = NULL;
1345 1346
}

L
Linus Torvalds 已提交
1347 1348 1349 1350
/*
 *	Combined all pending IP fragments on the socket as one IP datagram
 *	and push them out.
 */
1351
struct sk_buff *__ip_make_skb(struct sock *sk,
1352
			      struct flowi4 *fl4,
1353 1354
			      struct sk_buff_head *queue,
			      struct inet_cork *cork)
L
Linus Torvalds 已提交
1355 1356 1357 1358
{
	struct sk_buff *skb, *tmp_skb;
	struct sk_buff **tail_skb;
	struct inet_sock *inet = inet_sk(sk);
1359
	struct net *net = sock_net(sk);
L
Linus Torvalds 已提交
1360
	struct ip_options *opt = NULL;
1361
	struct rtable *rt = (struct rtable *)cork->dst;
L
Linus Torvalds 已提交
1362
	struct iphdr *iph;
1363
	__be16 df = 0;
L
Linus Torvalds 已提交
1364 1365
	__u8 ttl;

1366 1367
	skb = __skb_dequeue(queue);
	if (!skb)
L
Linus Torvalds 已提交
1368 1369 1370 1371
		goto out;
	tail_skb = &(skb_shinfo(skb)->frag_list);

	/* move skb->data to ip header from ext header */
1372
	if (skb->data < skb_network_header(skb))
1373
		__skb_pull(skb, skb_network_offset(skb));
1374
	while ((tmp_skb = __skb_dequeue(queue)) != NULL) {
1375
		__skb_pull(tmp_skb, skb_network_header_len(skb));
L
Linus Torvalds 已提交
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388
		*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
1389
	skb->ignore_df = ip_sk_ignore_df(sk);
L
Linus Torvalds 已提交
1390 1391

	/* DF bit is set when we want to see DF on outgoing frames.
W
WANG Cong 已提交
1392
	 * If ignore_df is set too, we still allow to fragment this frame
L
Linus Torvalds 已提交
1393
	 * locally. */
1394 1395
	if (inet->pmtudisc == IP_PMTUDISC_DO ||
	    inet->pmtudisc == IP_PMTUDISC_PROBE ||
1396 1397
	    (skb->len <= dst_mtu(&rt->dst) &&
	     ip_dont_fragment(sk, &rt->dst)))
L
Linus Torvalds 已提交
1398 1399
		df = htons(IP_DF);

1400 1401
	if (cork->flags & IPCORK_OPT)
		opt = cork->opt;
L
Linus Torvalds 已提交
1402

1403 1404 1405
	if (cork->ttl != 0)
		ttl = cork->ttl;
	else if (rt->rt_type == RTN_MULTICAST)
L
Linus Torvalds 已提交
1406 1407
		ttl = inet->mc_ttl;
	else
1408
		ttl = ip_select_ttl(inet, &rt->dst);
L
Linus Torvalds 已提交
1409

1410
	iph = ip_hdr(skb);
L
Linus Torvalds 已提交
1411 1412
	iph->version = 4;
	iph->ihl = 5;
1413
	iph->tos = (cork->tos != -1) ? cork->tos : inet->tos;
L
Linus Torvalds 已提交
1414 1415 1416
	iph->frag_off = df;
	iph->ttl = ttl;
	iph->protocol = sk->sk_protocol;
1417
	ip_copy_addrs(iph, fl4);
1418
	ip_select_ident(net, skb, sk);
L
Linus Torvalds 已提交
1419

1420 1421 1422 1423 1424
	if (opt) {
		iph->ihl += opt->optlen>>2;
		ip_options_build(skb, opt, cork->addr, rt, 0);
	}

1425
	skb->priority = (cork->tos != -1) ? cork->priority: sk->sk_priority;
1426
	skb->mark = sk->sk_mark;
1427 1428 1429 1430
	/*
	 * Steal rt from cork.dst to avoid a pair of atomic_inc/atomic_dec
	 * on dst refcount
	 */
1431
	cork->dst = NULL;
1432
	skb_dst_set(skb, &rt->dst);
L
Linus Torvalds 已提交
1433

1434
	if (iph->protocol == IPPROTO_ICMP)
1435
		icmp_out_count(net, ((struct icmphdr *)
1436 1437
			skb_transport_header(skb))->type);

1438 1439 1440 1441 1442
	ip_cork_release(cork);
out:
	return skb;
}

E
Eric Dumazet 已提交
1443
int ip_send_skb(struct net *net, struct sk_buff *skb)
1444 1445 1446
{
	int err;

H
Herbert Xu 已提交
1447
	err = ip_local_out(skb);
L
Linus Torvalds 已提交
1448 1449
	if (err) {
		if (err > 0)
E
Eric Dumazet 已提交
1450
			err = net_xmit_errno(err);
L
Linus Torvalds 已提交
1451
		if (err)
1452
			IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
L
Linus Torvalds 已提交
1453 1454 1455 1456 1457
	}

	return err;
}

1458
int ip_push_pending_frames(struct sock *sk, struct flowi4 *fl4)
1459
{
1460 1461
	struct sk_buff *skb;

1462
	skb = ip_finish_skb(sk, fl4);
1463 1464 1465 1466
	if (!skb)
		return 0;

	/* Netfilter gets whole the not fragmented skb. */
E
Eric Dumazet 已提交
1467
	return ip_send_skb(sock_net(sk), skb);
1468 1469
}

L
Linus Torvalds 已提交
1470 1471 1472
/*
 *	Throw away all pending data on the socket.
 */
1473 1474 1475
static void __ip_flush_pending_frames(struct sock *sk,
				      struct sk_buff_head *queue,
				      struct inet_cork *cork)
L
Linus Torvalds 已提交
1476 1477 1478
{
	struct sk_buff *skb;

1479
	while ((skb = __skb_dequeue_tail(queue)) != NULL)
L
Linus Torvalds 已提交
1480 1481
		kfree_skb(skb);

1482 1483 1484 1485 1486
	ip_cork_release(cork);
}

void ip_flush_pending_frames(struct sock *sk)
{
1487
	__ip_flush_pending_frames(sk, &sk->sk_write_queue, &inet_sk(sk)->cork.base);
L
Linus Torvalds 已提交
1488 1489
}

1490
struct sk_buff *ip_make_skb(struct sock *sk,
1491
			    struct flowi4 *fl4,
1492 1493 1494 1495 1496 1497
			    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)
{
1498
	struct inet_cork cork;
1499 1500 1501 1502 1503 1504 1505 1506
	struct sk_buff_head queue;
	int err;

	if (flags & MSG_PROBE)
		return NULL;

	__skb_queue_head_init(&queue);

1507 1508
	cork.flags = 0;
	cork.addr = 0;
1509
	cork.opt = NULL;
1510 1511 1512 1513
	err = ip_setup_cork(sk, &cork, ipc, rtp);
	if (err)
		return ERR_PTR(err);

1514 1515
	err = __ip_append_data(sk, fl4, &queue, &cork,
			       &current->task_frag, getfrag,
1516 1517 1518 1519 1520 1521
			       from, length, transhdrlen, flags);
	if (err) {
		__ip_flush_pending_frames(sk, &queue, &cork);
		return ERR_PTR(err);
	}

1522
	return __ip_make_skb(sk, fl4, &queue, &cork);
1523
}
L
Linus Torvalds 已提交
1524 1525 1526 1527

/*
 *	Fetch data from kernel space and fill in checksum if needed.
 */
1528
static int ip_reply_glue_bits(void *dptr, char *to, int offset,
L
Linus Torvalds 已提交
1529 1530
			      int len, int odd, struct sk_buff *skb)
{
1531
	__wsum csum;
L
Linus Torvalds 已提交
1532 1533 1534

	csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
	skb->csum = csum_block_add(skb->csum, csum, odd);
1535
	return 0;
L
Linus Torvalds 已提交
1536 1537
}

1538
/*
L
Linus Torvalds 已提交
1539
 *	Generic function to send a packet as reply to another packet.
1540
 *	Used to send some TCP resets/acks so far.
L
Linus Torvalds 已提交
1541
 */
1542
void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb,
1543 1544 1545
			   const struct ip_options *sopt,
			   __be32 daddr, __be32 saddr,
			   const struct ip_reply_arg *arg,
1546
			   unsigned int len)
L
Linus Torvalds 已提交
1547
{
1548
	struct ip_options_data replyopts;
L
Linus Torvalds 已提交
1549
	struct ipcm_cookie ipc;
1550
	struct flowi4 fl4;
E
Eric Dumazet 已提交
1551
	struct rtable *rt = skb_rtable(skb);
1552
	struct net *net = sock_net(sk);
1553
	struct sk_buff *nskb;
1554
	int err;
1555
	int oif;
L
Linus Torvalds 已提交
1556

1557
	if (__ip_options_echo(&replyopts.opt.opt, skb, sopt))
L
Linus Torvalds 已提交
1558 1559
		return;

1560
	ipc.addr = daddr;
L
Linus Torvalds 已提交
1561
	ipc.opt = NULL;
1562
	ipc.tx_flags = 0;
1563 1564
	ipc.ttl = 0;
	ipc.tos = -1;
L
Linus Torvalds 已提交
1565

1566
	if (replyopts.opt.opt.optlen) {
L
Linus Torvalds 已提交
1567 1568
		ipc.opt = &replyopts.opt;

1569 1570
		if (replyopts.opt.opt.srr)
			daddr = replyopts.opt.opt.faddr;
L
Linus Torvalds 已提交
1571 1572
	}

1573 1574 1575 1576 1577
	oif = arg->bound_dev_if;
	if (!oif && netif_index_is_vrf(net, skb->skb_iif))
		oif = skb->skb_iif;

	flowi4_init_output(&fl4, oif,
1578
			   IP4_REPLY_MARK(net, skb->mark),
1579
			   RT_TOS(arg->tos),
1580
			   RT_SCOPE_UNIVERSE, ip_hdr(skb)->protocol,
1581
			   ip_reply_arg_flowi_flags(arg),
1582
			   daddr, saddr,
1583 1584
			   tcp_hdr(skb)->source, tcp_hdr(skb)->dest);
	security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
1585
	rt = ip_route_output_key(net, &fl4);
1586 1587
	if (IS_ERR(rt))
		return;
L
Linus Torvalds 已提交
1588

1589
	inet_sk(sk)->tos = arg->tos;
L
Linus Torvalds 已提交
1590 1591

	sk->sk_priority = skb->priority;
1592
	sk->sk_protocol = ip_hdr(skb)->protocol;
1593
	sk->sk_bound_dev_if = arg->bound_dev_if;
1594
	sk->sk_sndbuf = sysctl_wmem_default;
1595 1596 1597 1598 1599 1600 1601
	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;
	}

1602 1603
	nskb = skb_peek(&sk->sk_write_queue);
	if (nskb) {
L
Linus Torvalds 已提交
1604
		if (arg->csumoffset >= 0)
1605 1606
			*((__sum16 *)skb_transport_header(nskb) +
			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
1607
								arg->csum));
1608 1609
		nskb->ip_summed = CHECKSUM_NONE;
		skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
1610
		ip_push_pending_frames(sk, &fl4);
L
Linus Torvalds 已提交
1611
	}
1612
out:
L
Linus Torvalds 已提交
1613 1614 1615 1616 1617 1618 1619 1620
	ip_rt_put(rt);
}

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

1621 1622
#if defined(CONFIG_IP_MULTICAST)
	igmp_mc_init();
L
Linus Torvalds 已提交
1623 1624
#endif
}